better caster (#1480)

related to mps/rocm/cpu casting for fp16 and etc on clip

.github/CODEOWNERS

@@ -0,0 +1 @@
+*       @lllyasviel

.github/ISSUE_TEMPLATE/bug_report.md

@@ -0,0 +1,18 @@
+---
+name: Bug report
+about: Describe a problem
+title: ''
+labels: ''
+assignees: ''
+
+---
+
+**Read Troubleshoot**
+
+[x] I admit that I have read the [Troubleshoot](https://github.com/lllyasviel/Fooocus/blob/main/troubleshoot.md) before making this issue.
+
+**Describe the problem**
+A clear and concise description of what the bug is.
+
+**Full Console Log**
+Paste **full** console log here. You will make our job easier if you give a **full** log.

.github/ISSUE_TEMPLATE/feature_request.md

@@ -0,0 +1,14 @@
+---
+name: Feature request
+about: Suggest an idea for this project
+title: ''
+labels: ''
+assignees: ''
+
+---
+
+**Is your feature request related to a problem? Please describe.**
+A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]
+
+**Describe the idea you'd like**
+A clear and concise description of what you want to happen.

.gitignore

@@ -2,7 +2,23 @@ __pycache__
 *.ckpt
 *.safetensors
 *.pth
-!taesdxl_decoder.pth
+*.pt
+*.bin
+*.patch
+*.backup
+*.corrupted
+*.partial
+*.onnx
+sorted_styles.json
+/input
+/cache
+/language/default.json
+/test_imgs
+config.txt
+config_modification_tutorial.txt
+user_path_config.txt
+user_path_config-deprecated.txt
+/modules/*.png
 /repositories
 /venv
 /tmp
@@ -20,7 +36,8 @@ __pycache__
 /interrogate
 /user.css
 /.idea
-notification.mp3
+/notification.ogg
+/notification.mp3
 /SwinIR
 /textual_inversion
 .vscode
@@ -32,3 +49,4 @@ notification.mp3
 /node_modules
 /package-lock.json
 /.coverage*
+/auth.json

LICENSE

@@ -0,0 +1,674 @@
+                    GNU GENERAL PUBLIC LICENSE
+                       Version 3, 29 June 2007
+
+ Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+                            Preamble
+
+  The GNU General Public License is a free, copyleft license for
+software and other kinds of works.
+
+  The licenses for most software and other practical works are designed
+to take away your freedom to share and change the works.  By contrast,
+the GNU General Public License is intended to guarantee your freedom to
+share and change all versions of a program--to make sure it remains free
+software for all its users.  We, the Free Software Foundation, use the
+GNU General Public License for most of our software; it applies also to
+any other work released this way by its authors.  You can apply it to
+your programs, too.
+
+  When we speak of free software, we are referring to freedom, not
+price.  Our General Public Licenses are designed to make sure that you
+have the freedom to distribute copies of free software (and charge for
+them if you wish), that you receive source code or can get it if you
+want it, that you can change the software or use pieces of it in new
+free programs, and that you know you can do these things.
+
+  To protect your rights, we need to prevent others from denying you
+these rights or asking you to surrender the rights.  Therefore, you have
+certain responsibilities if you distribute copies of the software, or if
+you modify it: responsibilities to respect the freedom of others.
+
+  For example, if you distribute copies of such a program, whether
+gratis or for a fee, you must pass on to the recipients the same
+freedoms that you received.  You must make sure that they, too, receive
+or can get the source code.  And you must show them these terms so they
+know their rights.
+
+  Developers that use the GNU GPL protect your rights with two steps:
+(1) assert copyright on the software, and (2) offer you this License
+giving you legal permission to copy, distribute and/or modify it.
+
+  For the developers' and authors' protection, the GPL clearly explains
+that there is no warranty for this free software.  For both users' and
+authors' sake, the GPL requires that modified versions be marked as
+changed, so that their problems will not be attributed erroneously to
+authors of previous versions.
+
+  Some devices are designed to deny users access to install or run
+modified versions of the software inside them, although the manufacturer
+can do so.  This is fundamentally incompatible with the aim of
+protecting users' freedom to change the software.  The systematic
+pattern of such abuse occurs in the area of products for individuals to
+use, which is precisely where it is most unacceptable.  Therefore, we
+have designed this version of the GPL to prohibit the practice for those
+products.  If such problems arise substantially in other domains, we
+stand ready to extend this provision to those domains in future versions
+of the GPL, as needed to protect the freedom of users.
+
+  Finally, every program is threatened constantly by software patents.
+States should not allow patents to restrict development and use of
+software on general-purpose computers, but in those that do, we wish to
+avoid the special danger that patents applied to a free program could
+make it effectively proprietary.  To prevent this, the GPL assures that
+patents cannot be used to render the program non-free.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.
+
+                       TERMS AND CONDITIONS
+
+  0. Definitions.
+
+  "This License" refers to version 3 of the GNU General Public License.
+
+  "Copyright" also means copyright-like laws that apply to other kinds of
+works, such as semiconductor masks.
+
+  "The Program" refers to any copyrightable work licensed under this
+License.  Each licensee is addressed as "you".  "Licensees" and
+"recipients" may be individuals or organizations.
+
+  To "modify" a work means to copy from or adapt all or part of the work
+in a fashion requiring copyright permission, other than the making of an
+exact copy.  The resulting work is called a "modified version" of the
+earlier work or a work "based on" the earlier work.
+
+  A "covered work" means either the unmodified Program or a work based
+on the Program.
+
+  To "propagate" a work means to do anything with it that, without
+permission, would make you directly or secondarily liable for
+infringement under applicable copyright law, except executing it on a
+computer or modifying a private copy.  Propagation includes copying,
+distribution (with or without modification), making available to the
+public, and in some countries other activities as well.
+
+  To "convey" a work means any kind of propagation that enables other
+parties to make or receive copies.  Mere interaction with a user through
+a computer network, with no transfer of a copy, is not conveying.
+
+  An interactive user interface displays "Appropriate Legal Notices"
+to the extent that it includes a convenient and prominently visible
+feature that (1) displays an appropriate copyright notice, and (2)
+tells the user that there is no warranty for the work (except to the
+extent that warranties are provided), that licensees may convey the
+work under this License, and how to view a copy of this License.  If
+the interface presents a list of user commands or options, such as a
+menu, a prominent item in the list meets this criterion.
+
+  1. Source Code.
+
+  The "source code" for a work means the preferred form of the work
+for making modifications to it.  "Object code" means any non-source
+form of a work.
+
+  A "Standard Interface" means an interface that either is an official
+standard defined by a recognized standards body, or, in the case of
+interfaces specified for a particular programming language, one that
+is widely used among developers working in that language.
+
+  The "System Libraries" of an executable work include anything, other
+than the work as a whole, that (a) is included in the normal form of
+packaging a Major Component, but which is not part of that Major
+Component, and (b) serves only to enable use of the work with that
+Major Component, or to implement a Standard Interface for which an
+implementation is available to the public in source code form.  A
+"Major Component", in this context, means a major essential component
+(kernel, window system, and so on) of the specific operating system
+(if any) on which the executable work runs, or a compiler used to
+produce the work, or an object code interpreter used to run it.
+
+  The "Corresponding Source" for a work in object code form means all
+the source code needed to generate, install, and (for an executable
+work) run the object code and to modify the work, including scripts to
+control those activities.  However, it does not include the work's
+System Libraries, or general-purpose tools or generally available free
+programs which are used unmodified in performing those activities but
+which are not part of the work.  For example, Corresponding Source
+includes interface definition files associated with source files for
+the work, and the source code for shared libraries and dynamically
+linked subprograms that the work is specifically designed to require,
+such as by intimate data communication or control flow between those
+subprograms and other parts of the work.
+
+  The Corresponding Source need not include anything that users
+can regenerate automatically from other parts of the Corresponding
+Source.
+
+  The Corresponding Source for a work in source code form is that
+same work.
+
+  2. Basic Permissions.
+
+  All rights granted under this License are granted for the term of
+copyright on the Program, and are irrevocable provided the stated
+conditions are met.  This License explicitly affirms your unlimited
+permission to run the unmodified Program.  The output from running a
+covered work is covered by this License only if the output, given its
+content, constitutes a covered work.  This License acknowledges your
+rights of fair use or other equivalent, as provided by copyright law.
+
+  You may make, run and propagate covered works that you do not
+convey, without conditions so long as your license otherwise remains
+in force.  You may convey covered works to others for the sole purpose
+of having them make modifications exclusively for you, or provide you
+with facilities for running those works, provided that you comply with
+the terms of this License in conveying all material for which you do
+not control copyright.  Those thus making or running the covered works
+for you must do so exclusively on your behalf, under your direction
+and control, on terms that prohibit them from making any copies of
+your copyrighted material outside their relationship with you.
+
+  Conveying under any other circumstances is permitted solely under
+the conditions stated below.  Sublicensing is not allowed; section 10
+makes it unnecessary.
+
+  3. Protecting Users' Legal Rights From Anti-Circumvention Law.
+
+  No covered work shall be deemed part of an effective technological
+measure under any applicable law fulfilling obligations under article
+11 of the WIPO copyright treaty adopted on 20 December 1996, or
+similar laws prohibiting or restricting circumvention of such
+measures.
+
+  When you convey a covered work, you waive any legal power to forbid
+circumvention of technological measures to the extent such circumvention
+is effected by exercising rights under this License with respect to
+the covered work, and you disclaim any intention to limit operation or
+modification of the work as a means of enforcing, against the work's
+users, your or third parties' legal rights to forbid circumvention of
+technological measures.
+
+  4. Conveying Verbatim Copies.
+
+  You may convey verbatim copies of the Program's source code as you
+receive it, in any medium, provided that you conspicuously and
+appropriately publish on each copy an appropriate copyright notice;
+keep intact all notices stating that this License and any
+non-permissive terms added in accord with section 7 apply to the code;
+keep intact all notices of the absence of any warranty; and give all
+recipients a copy of this License along with the Program.
+
+  You may charge any price or no price for each copy that you convey,
+and you may offer support or warranty protection for a fee.
+
+  5. Conveying Modified Source Versions.
+
+  You may convey a work based on the Program, or the modifications to
+produce it from the Program, in the form of source code under the
+terms of section 4, provided that you also meet all of these conditions:
+
+    a) The work must carry prominent notices stating that you modified
+    it, and giving a relevant date.
+
+    b) The work must carry prominent notices stating that it is
+    released under this License and any conditions added under section
+    7.  This requirement modifies the requirement in section 4 to
+    "keep intact all notices".
+
+    c) You must license the entire work, as a whole, under this
+    License to anyone who comes into possession of a copy.  This
+    License will therefore apply, along with any applicable section 7
+    additional terms, to the whole of the work, and all its parts,
+    regardless of how they are packaged.  This License gives no
+    permission to license the work in any other way, but it does not
+    invalidate such permission if you have separately received it.
+
+    d) If the work has interactive user interfaces, each must display
+    Appropriate Legal Notices; however, if the Program has interactive
+    interfaces that do not display Appropriate Legal Notices, your
+    work need not make them do so.
+
+  A compilation of a covered work with other separate and independent
+works, which are not by their nature extensions of the covered work,
+and which are not combined with it such as to form a larger program,
+in or on a volume of a storage or distribution medium, is called an
+"aggregate" if the compilation and its resulting copyright are not
+used to limit the access or legal rights of the compilation's users
+beyond what the individual works permit.  Inclusion of a covered work
+in an aggregate does not cause this License to apply to the other
+parts of the aggregate.
+
+  6. Conveying Non-Source Forms.
+
+  You may convey a covered work in object code form under the terms
+of sections 4 and 5, provided that you also convey the
+machine-readable Corresponding Source under the terms of this License,
+in one of these ways:
+
+    a) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by the
+    Corresponding Source fixed on a durable physical medium
+    customarily used for software interchange.
+
+    b) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by a
+    written offer, valid for at least three years and valid for as
+    long as you offer spare parts or customer support for that product
+    model, to give anyone who possesses the object code either (1) a
+    copy of the Corresponding Source for all the software in the
+    product that is covered by this License, on a durable physical
+    medium customarily used for software interchange, for a price no
+    more than your reasonable cost of physically performing this
+    conveying of source, or (2) access to copy the
+    Corresponding Source from a network server at no charge.
+
+    c) Convey individual copies of the object code with a copy of the
+    written offer to provide the Corresponding Source.  This
+    alternative is allowed only occasionally and noncommercially, and
+    only if you received the object code with such an offer, in accord
+    with subsection 6b.
+
+    d) Convey the object code by offering access from a designated
+    place (gratis or for a charge), and offer equivalent access to the
+    Corresponding Source in the same way through the same place at no
+    further charge.  You need not require recipients to copy the
+    Corresponding Source along with the object code.  If the place to
+    copy the object code is a network server, the Corresponding Source
+    may be on a different server (operated by you or a third party)
+    that supports equivalent copying facilities, provided you maintain
+    clear directions next to the object code saying where to find the
+    Corresponding Source.  Regardless of what server hosts the
+    Corresponding Source, you remain obligated to ensure that it is
+    available for as long as needed to satisfy these requirements.
+
+    e) Convey the object code using peer-to-peer transmission, provided
+    you inform other peers where the object code and Corresponding
+    Source of the work are being offered to the general public at no
+    charge under subsection 6d.
+
+  A separable portion of the object code, whose source code is excluded
+from the Corresponding Source as a System Library, need not be
+included in conveying the object code work.
+
+  A "User Product" is either (1) a "consumer product", which means any
+tangible personal property which is normally used for personal, family,
+or household purposes, or (2) anything designed or sold for incorporation
+into a dwelling.  In determining whether a product is a consumer product,
+doubtful cases shall be resolved in favor of coverage.  For a particular
+product received by a particular user, "normally used" refers to a
+typical or common use of that class of product, regardless of the status
+of the particular user or of the way in which the particular user
+actually uses, or expects or is expected to use, the product.  A product
+is a consumer product regardless of whether the product has substantial
+commercial, industrial or non-consumer uses, unless such uses represent
+the only significant mode of use of the product.
+
+  "Installation Information" for a User Product means any methods,
+procedures, authorization keys, or other information required to install
+and execute modified versions of a covered work in that User Product from
+a modified version of its Corresponding Source.  The information must
+suffice to ensure that the continued functioning of the modified object
+code is in no case prevented or interfered with solely because
+modification has been made.
+
+  If you convey an object code work under this section in, or with, or
+specifically for use in, a User Product, and the conveying occurs as
+part of a transaction in which the right of possession and use of the
+User Product is transferred to the recipient in perpetuity or for a
+fixed term (regardless of how the transaction is characterized), the
+Corresponding Source conveyed under this section must be accompanied
+by the Installation Information.  But this requirement does not apply
+if neither you nor any third party retains the ability to install
+modified object code on the User Product (for example, the work has
+been installed in ROM).
+
+  The requirement to provide Installation Information does not include a
+requirement to continue to provide support service, warranty, or updates
+for a work that has been modified or installed by the recipient, or for
+the User Product in which it has been modified or installed.  Access to a
+network may be denied when the modification itself materially and
+adversely affects the operation of the network or violates the rules and
+protocols for communication across the network.
+
+  Corresponding Source conveyed, and Installation Information provided,
+in accord with this section must be in a format that is publicly
+documented (and with an implementation available to the public in
+source code form), and must require no special password or key for
+unpacking, reading or copying.
+
+  7. Additional Terms.
+
+  "Additional permissions" are terms that supplement the terms of this
+License by making exceptions from one or more of its conditions.
+Additional permissions that are applicable to the entire Program shall
+be treated as though they were included in this License, to the extent
+that they are valid under applicable law.  If additional permissions
+apply only to part of the Program, that part may be used separately
+under those permissions, but the entire Program remains governed by
+this License without regard to the additional permissions.
+
+  When you convey a copy of a covered work, you may at your option
+remove any additional permissions from that copy, or from any part of
+it.  (Additional permissions may be written to require their own
+removal in certain cases when you modify the work.)  You may place
+additional permissions on material, added by you to a covered work,
+for which you have or can give appropriate copyright permission.
+
+  Notwithstanding any other provision of this License, for material you
+add to a covered work, you may (if authorized by the copyright holders of
+that material) supplement the terms of this License with terms:
+
+    a) Disclaiming warranty or limiting liability differently from the
+    terms of sections 15 and 16 of this License; or
+
+    b) Requiring preservation of specified reasonable legal notices or
+    author attributions in that material or in the Appropriate Legal
+    Notices displayed by works containing it; or
+
+    c) Prohibiting misrepresentation of the origin of that material, or
+    requiring that modified versions of such material be marked in
+    reasonable ways as different from the original version; or
+
+    d) Limiting the use for publicity purposes of names of licensors or
+    authors of the material; or
+
+    e) Declining to grant rights under trademark law for use of some
+    trade names, trademarks, or service marks; or
+
+    f) Requiring indemnification of licensors and authors of that
+    material by anyone who conveys the material (or modified versions of
+    it) with contractual assumptions of liability to the recipient, for
+    any liability that these contractual assumptions directly impose on
+    those licensors and authors.
+
+  All other non-permissive additional terms are considered "further
+restrictions" within the meaning of section 10.  If the Program as you
+received it, or any part of it, contains a notice stating that it is
+governed by this License along with a term that is a further
+restriction, you may remove that term.  If a license document contains
+a further restriction but permits relicensing or conveying under this
+License, you may add to a covered work material governed by the terms
+of that license document, provided that the further restriction does
+not survive such relicensing or conveying.
+
+  If you add terms to a covered work in accord with this section, you
+must place, in the relevant source files, a statement of the
+additional terms that apply to those files, or a notice indicating
+where to find the applicable terms.
+
+  Additional terms, permissive or non-permissive, may be stated in the
+form of a separately written license, or stated as exceptions;
+the above requirements apply either way.
+
+  8. Termination.
+
+  You may not propagate or modify a covered work except as expressly
+provided under this License.  Any attempt otherwise to propagate or
+modify it is void, and will automatically terminate your rights under
+this License (including any patent licenses granted under the third
+paragraph of section 11).
+
+  However, if you cease all violation of this License, then your
+license from a particular copyright holder is reinstated (a)
+provisionally, unless and until the copyright holder explicitly and
+finally terminates your license, and (b) permanently, if the copyright
+holder fails to notify you of the violation by some reasonable means
+prior to 60 days after the cessation.
+
+  Moreover, your license from a particular copyright holder is
+reinstated permanently if the copyright holder notifies you of the
+violation by some reasonable means, this is the first time you have
+received notice of violation of this License (for any work) from that
+copyright holder, and you cure the violation prior to 30 days after
+your receipt of the notice.
+
+  Termination of your rights under this section does not terminate the
+licenses of parties who have received copies or rights from you under
+this License.  If your rights have been terminated and not permanently
+reinstated, you do not qualify to receive new licenses for the same
+material under section 10.
+
+  9. Acceptance Not Required for Having Copies.
+
+  You are not required to accept this License in order to receive or
+run a copy of the Program.  Ancillary propagation of a covered work
+occurring solely as a consequence of using peer-to-peer transmission
+to receive a copy likewise does not require acceptance.  However,
+nothing other than this License grants you permission to propagate or
+modify any covered work.  These actions infringe copyright if you do
+not accept this License.  Therefore, by modifying or propagating a
+covered work, you indicate your acceptance of this License to do so.
+
+  10. Automatic Licensing of Downstream Recipients.
+
+  Each time you convey a covered work, the recipient automatically
+receives a license from the original licensors, to run, modify and
+propagate that work, subject to this License.  You are not responsible
+for enforcing compliance by third parties with this License.
+
+  An "entity transaction" is a transaction transferring control of an
+organization, or substantially all assets of one, or subdividing an
+organization, or merging organizations.  If propagation of a covered
+work results from an entity transaction, each party to that
+transaction who receives a copy of the work also receives whatever
+licenses to the work the party's predecessor in interest had or could
+give under the previous paragraph, plus a right to possession of the
+Corresponding Source of the work from the predecessor in interest, if
+the predecessor has it or can get it with reasonable efforts.
+
+  You may not impose any further restrictions on the exercise of the
+rights granted or affirmed under this License.  For example, you may
+not impose a license fee, royalty, or other charge for exercise of
+rights granted under this License, and you may not initiate litigation
+(including a cross-claim or counterclaim in a lawsuit) alleging that
+any patent claim is infringed by making, using, selling, offering for
+sale, or importing the Program or any portion of it.
+
+  11. Patents.
+
+  A "contributor" is a copyright holder who authorizes use under this
+License of the Program or a work on which the Program is based.  The
+work thus licensed is called the contributor's "contributor version".
+
+  A contributor's "essential patent claims" are all patent claims
+owned or controlled by the contributor, whether already acquired or
+hereafter acquired, that would be infringed by some manner, permitted
+by this License, of making, using, or selling its contributor version,
+but do not include claims that would be infringed only as a
+consequence of further modification of the contributor version.  For
+purposes of this definition, "control" includes the right to grant
+patent sublicenses in a manner consistent with the requirements of
+this License.
+
+  Each contributor grants you a non-exclusive, worldwide, royalty-free
+patent license under the contributor's essential patent claims, to
+make, use, sell, offer for sale, import and otherwise run, modify and
+propagate the contents of its contributor version.
+
+  In the following three paragraphs, a "patent license" is any express
+agreement or commitment, however denominated, not to enforce a patent
+(such as an express permission to practice a patent or covenant not to
+sue for patent infringement).  To "grant" such a patent license to a
+party means to make such an agreement or commitment not to enforce a
+patent against the party.
+
+  If you convey a covered work, knowingly relying on a patent license,
+and the Corresponding Source of the work is not available for anyone
+to copy, free of charge and under the terms of this License, through a
+publicly available network server or other readily accessible means,
+then you must either (1) cause the Corresponding Source to be so
+available, or (2) arrange to deprive yourself of the benefit of the
+patent license for this particular work, or (3) arrange, in a manner
+consistent with the requirements of this License, to extend the patent
+license to downstream recipients.  "Knowingly relying" means you have
+actual knowledge that, but for the patent license, your conveying the
+covered work in a country, or your recipient's use of the covered work
+in a country, would infringe one or more identifiable patents in that
+country that you have reason to believe are valid.
+
+  If, pursuant to or in connection with a single transaction or
+arrangement, you convey, or propagate by procuring conveyance of, a
+covered work, and grant a patent license to some of the parties
+receiving the covered work authorizing them to use, propagate, modify
+or convey a specific copy of the covered work, then the patent license
+you grant is automatically extended to all recipients of the covered
+work and works based on it.
+
+  A patent license is "discriminatory" if it does not include within
+the scope of its coverage, prohibits the exercise of, or is
+conditioned on the non-exercise of one or more of the rights that are
+specifically granted under this License.  You may not convey a covered
+work if you are a party to an arrangement with a third party that is
+in the business of distributing software, under which you make payment
+to the third party based on the extent of your activity of conveying
+the work, and under which the third party grants, to any of the
+parties who would receive the covered work from you, a discriminatory
+patent license (a) in connection with copies of the covered work
+conveyed by you (or copies made from those copies), or (b) primarily
+for and in connection with specific products or compilations that
+contain the covered work, unless you entered into that arrangement,
+or that patent license was granted, prior to 28 March 2007.
+
+  Nothing in this License shall be construed as excluding or limiting
+any implied license or other defenses to infringement that may
+otherwise be available to you under applicable patent law.
+
+  12. No Surrender of Others' Freedom.
+
+  If conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot convey a
+covered work so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you may
+not convey it at all.  For example, if you agree to terms that obligate you
+to collect a royalty for further conveying from those to whom you convey
+the Program, the only way you could satisfy both those terms and this
+License would be to refrain entirely from conveying the Program.
+
+  13. Use with the GNU Affero General Public License.
+
+  Notwithstanding any other provision of this License, you have
+permission to link or combine any covered work with a work licensed
+under version 3 of the GNU Affero General Public License into a single
+combined work, and to convey the resulting work.  The terms of this
+License will continue to apply to the part which is the covered work,
+but the special requirements of the GNU Affero General Public License,
+section 13, concerning interaction through a network will apply to the
+combination as such.
+
+  14. Revised Versions of this License.
+
+  The Free Software Foundation may publish revised and/or new versions of
+the GNU General Public License from time to time.  Such new versions will
+be similar in spirit to the present version, but may differ in detail to
+address new problems or concerns.
+
+  Each version is given a distinguishing version number.  If the
+Program specifies that a certain numbered version of the GNU General
+Public License "or any later version" applies to it, you have the
+option of following the terms and conditions either of that numbered
+version or of any later version published by the Free Software
+Foundation.  If the Program does not specify a version number of the
+GNU General Public License, you may choose any version ever published
+by the Free Software Foundation.
+
+  If the Program specifies that a proxy can decide which future
+versions of the GNU General Public License can be used, that proxy's
+public statement of acceptance of a version permanently authorizes you
+to choose that version for the Program.
+
+  Later license versions may give you additional or different
+permissions.  However, no additional obligations are imposed on any
+author or copyright holder as a result of your choosing to follow a
+later version.
+
+  15. Disclaimer of Warranty.
+
+  THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
+APPLICABLE LAW.  EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
+HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
+OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
+THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
+IS WITH YOU.  SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
+ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
+
+  16. Limitation of Liability.
+
+  IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
+THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
+GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
+USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
+DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
+PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
+EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
+SUCH DAMAGES.
+
+  17. Interpretation of Sections 15 and 16.
+
+  If the disclaimer of warranty and limitation of liability provided
+above cannot be given local legal effect according to their terms,
+reviewing courts shall apply local law that most closely approximates
+an absolute waiver of all civil liability in connection with the
+Program, unless a warranty or assumption of liability accompanies a
+copy of the Program in return for a fee.
+
+                     END OF TERMS AND CONDITIONS
+
+            How to Apply These Terms to Your New Programs
+
+  If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these terms.
+
+  To do so, attach the following notices to the program.  It is safest
+to attach them to the start of each source file to most effectively
+state the exclusion of warranty; and each file should have at least
+the "copyright" line and a pointer to where the full notice is found.
+
+    <one line to give the program's name and a brief idea of what it does.>
+    Copyright (C) <year>  <name of author>
+
+    This program is free software: you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation, either version 3 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License
+    along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+Also add information on how to contact you by electronic and paper mail.
+
+  If the program does terminal interaction, make it output a short
+notice like this when it starts in an interactive mode:
+
+    <program>  Copyright (C) <year>  <name of author>
+    This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
+    This is free software, and you are welcome to redistribute it
+    under certain conditions; type `show c' for details.
+
+The hypothetical commands `show w' and `show c' should show the appropriate
+parts of the General Public License.  Of course, your program's commands
+might be different; for a GUI interface, you would use an "about box".
+
+  You should also get your employer (if you work as a programmer) or school,
+if any, to sign a "copyright disclaimer" for the program, if necessary.
+For more information on this, and how to apply and follow the GNU GPL, see
+<https://www.gnu.org/licenses/>.
+
+  The GNU General Public License does not permit incorporating your program
+into proprietary programs.  If your program is a subroutine library, you
+may consider it more useful to permit linking proprietary applications with
+the library.  If this is what you want to do, use the GNU Lesser General
+Public License instead of this License.  But first, please read
+<https://www.gnu.org/licenses/why-not-lgpl.html>.

args_manager.py

@@ -0,0 +1,38 @@
+import ldm_patched.modules.args_parser as args_parser
+
+
+args_parser.parser.add_argument("--share", action='store_true', help="Set whether to share on Gradio.")
+args_parser.parser.add_argument("--preset", type=str, default=None, help="Apply specified UI preset.")
+
+args_parser.parser.add_argument("--language", type=str, default='default',
+                                help="Translate UI using json files in [language] folder. "
+                                  "For example, [--language example] will use [language/example.json] for translation.")
+
+# For example, https://github.com/lllyasviel/Fooocus/issues/849
+args_parser.parser.add_argument("--disable-offload-from-vram", action="store_true",
+                                help="Force loading models to vram when the unload can be avoided. "
+                                  "Some Mac users may need this.")
+
+args_parser.parser.add_argument("--theme", type=str, help="launches the UI with light or dark theme", default=None)
+args_parser.parser.add_argument("--disable-image-log", action='store_true',
+                                help="Prevent writing images and logs to hard drive.")
+
+args_parser.parser.add_argument("--disable-analytics", action='store_true',
+                                help="Disables analytics for Gradio", default=False)
+
+args_parser.parser.set_defaults(
+    disable_cuda_malloc=True,
+    in_browser=True,
+    port=None
+)
+
+args_parser.args = args_parser.parser.parse_args()
+
+# (Disable by default because of issues like https://github.com/lllyasviel/Fooocus/issues/724)
+args_parser.args.always_offload_from_vram = not args_parser.args.disable_offload_from_vram
+
+if args_parser.args.disable_analytics:
+    import os
+    os.environ["GRADIO_ANALYTICS_ENABLED"] = "False"
+
+args = args_parser.args

auth-example.json

@@ -0,0 +1,6 @@
+[
+    {
+        "user": "sitting-duck-1",
+        "pass": "very-bad-publicly-known-password-change-it"
+    }
+]

build_launcher.py

@@ -0,0 +1,26 @@
+import os
+
+win32_root = os.path.dirname(os.path.dirname(__file__))
+python_embeded_path = os.path.join(win32_root, 'python_embeded')
+
+is_win32_standalone_build = os.path.exists(python_embeded_path) and os.path.isdir(python_embeded_path)
+
+win32_cmd = '''
+.\python_embeded\python.exe -s Fooocus\entry_with_update.py {cmds} %*
+pause
+'''
+
+
+def build_launcher():
+    if not is_win32_standalone_build:
+        return
+
+    presets = [None, 'anime', 'realistic']
+
+    for preset in presets:
+        win32_cmd_preset = win32_cmd.replace('{cmds}', '' if preset is None else f'--preset {preset}')
+        bat_path = os.path.join(win32_root, 'run.bat' if preset is None else f'run_{preset}.bat')
+        if not os.path.exists(bat_path):
+            with open(bat_path, "w", encoding="utf-8") as f:
+                f.write(win32_cmd_preset)
+    return

css/style.css

@@ -0,0 +1,198 @@
+/* based on https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob/v1.6.0/style.css */
+
+#context-menu{
+    z-index:9999;
+    position:absolute;
+    display:block;
+    padding:0px 0;
+    border:2px solid #a55000;
+    border-radius:8px;
+    box-shadow:1px 1px 2px #CE6400;
+    width: 200px;
+}
+
+.context-menu-items{
+    list-style: none;
+    margin: 0;
+    padding: 0;
+}
+
+.context-menu-items a{
+    display:block;
+    padding:5px;
+    cursor:pointer;
+}
+
+.context-menu-items a:hover{
+    background: #a55000;
+}
+
+.canvas-tooltip-info {
+  position: absolute;
+  top: 28px;
+  left: 2px;
+  cursor: help;
+  background-color: rgba(0, 0, 0, 0.3);
+  width: 20px;
+  height: 20px;
+  border-radius: 50%;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  flex-direction: column;
+
+  z-index: 100;
+}
+
+.canvas-tooltip-info::after {
+  content: '';
+  display: block;
+  width: 2px;
+  height: 7px;
+  background-color: white;
+  margin-top: 2px;
+}
+
+.canvas-tooltip-info::before {
+  content: '';
+  display: block;
+  width: 2px;
+  height: 2px;
+  background-color: white;
+}
+
+.canvas-tooltip-content {
+  display: none;
+  background-color: #f9f9f9;
+  color: #333;
+  border: 1px solid #ddd;
+  padding: 15px;
+  position: absolute;
+  top: 40px;
+  left: 10px;
+  width: 250px;
+  font-size: 16px;
+  opacity: 0;
+  border-radius: 8px;
+  box-shadow: 0px 8px 16px 0px rgba(0,0,0,0.2);
+
+  z-index: 100;
+}
+
+.canvas-tooltip:hover .canvas-tooltip-content {
+  display: block;
+  animation: fadeIn 0.5s;
+  opacity: 1;
+}
+
+@keyframes fadeIn {
+  from {opacity: 0;}
+  to {opacity: 1;}
+}
+
+.styler {
+  overflow:inherit !important;
+}
+
+.gradio-container{
+  overflow: visible;
+}
+
+/* fullpage image viewer */
+
+#lightboxModal{
+    display: none;
+    position: fixed;
+    z-index: 1001;
+    left: 0;
+    top: 0;
+    width: 100%;
+    height: 100%;
+    overflow: auto;
+    background-color: rgba(20, 20, 20, 0.95);
+    user-select: none;
+    -webkit-user-select: none;
+    flex-direction: column;
+}
+
+.modalControls {
+    display: flex;
+    position: absolute;
+    right: 0px;
+    left: 0px;
+    gap: 1em;
+    padding: 1em;
+    background-color:rgba(0,0,0,0);
+    z-index: 1;
+    transition: 0.2s ease background-color;
+}
+.modalControls:hover {
+    background-color:rgba(0,0,0,0.9);
+}
+.modalClose {
+    margin-left: auto;
+}
+.modalControls span{
+    color: white;
+    text-shadow: 0px 0px 0.25em black;
+    font-size: 35px;
+    font-weight: bold;
+    cursor: pointer;
+    width: 1em;
+}
+
+.modalControls span:hover, .modalControls span:focus{
+    color: #999;
+    text-decoration: none;
+}
+
+#lightboxModal > img {
+    display: block;
+    margin: auto;
+    width: auto;
+}
+
+#lightboxModal > img.modalImageFullscreen{
+    object-fit: contain;
+    height: 100%;
+    width: 100%;
+    min-height: 0;
+}
+
+.modalPrev,
+.modalNext {
+  cursor: pointer;
+  position: absolute;
+  top: 50%;
+  width: auto;
+  padding: 16px;
+  margin-top: -50px;
+  color: white;
+  font-weight: bold;
+  font-size: 20px;
+  transition: 0.6s ease;
+  border-radius: 0 3px 3px 0;
+  user-select: none;
+  -webkit-user-select: none;
+}
+
+.modalNext {
+  right: 0;
+  border-radius: 3px 0 0 3px;
+}
+
+.modalPrev:hover,
+.modalNext:hover {
+  background-color: rgba(0, 0, 0, 0.8);
+}
+
+#imageARPreview {
+    position: absolute;
+    top: 0px;
+    left: 0px;
+    border: 2px solid red;
+    background: rgba(255, 0, 0, 0.3);
+    z-index: 900;
+    pointer-events: none;
+    display: none;
+}

entry_with_update.py

@@ -9,6 +9,7 @@ os.chdir(root)
 
 try:
     import pygit2
+    pygit2.option(pygit2.GIT_OPT_SET_OWNER_VALIDATION, 0)
 
     repo = pygit2.Repository(os.path.abspath(os.path.dirname(__file__)))
 
@@ -36,7 +37,7 @@ try:
         repo.reset(local_branch.target, pygit2.GIT_RESET_HARD)
         print("Fast-forward merge")
     elif merge_result & pygit2.GIT_MERGE_ANALYSIS_NORMAL:
-        print("Update failed - Did you modified any file?")
+        print("Update failed - Did you modify any file?")
 except Exception as e:
     print('Update failed.')
     print(str(e))

experiments_expansion.py

@@ -0,0 +1,8 @@
+from modules.expansion import FooocusExpansion
+
+expansion = FooocusExpansion()
+
+text = 'a handsome man'
+
+for i in range(64):
+    print(expansion(text, seed=i))

experiments_face.py

@@ -0,0 +1,7 @@
+import cv2
+import extras.face_crop as cropper
+
+
+img = cv2.imread('lena.png')
+result = cropper.crop_image(img)
+cv2.imwrite('lena_result.png', result)

experiments_interrogate.py

@@ -0,0 +1,8 @@
+import cv2
+from extras.interrogate import default_interrogator as default_interrogator_photo
+from extras.wd14tagger import default_interrogator as default_interrogator_anime
+
+img = cv2.imread('./test_imgs/red_box.jpg')[:, :, ::-1].copy()
+print(default_interrogator_photo(img))
+img = cv2.imread('./test_imgs/miku.jpg')[:, :, ::-1].copy()
+print(default_interrogator_anime(img))

extras/BLIP/configs/bert_config.json

@@ -0,0 +1,21 @@
+{
+  "architectures": [
+    "BertModel"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "type_vocab_size": 2,
+  "vocab_size": 30522,
+  "encoder_width": 768,
+  "add_cross_attention": true   
+}

extras/BLIP/configs/caption_coco.yaml

@@ -0,0 +1,33 @@
+image_root: '/export/share/datasets/vision/coco/images/'
+ann_root: 'annotation'
+coco_gt_root: 'annotation/coco_gt'
+
+# set pretrained as a file path or an url
+pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_caption_capfilt_large.pth'
+
+# size of vit model; base or large
+vit: 'base'
+vit_grad_ckpt: False
+vit_ckpt_layer: 0
+batch_size: 32
+init_lr: 1e-5
+
+# vit: 'large'
+# vit_grad_ckpt: True
+# vit_ckpt_layer: 5
+# batch_size: 16
+# init_lr: 2e-6
+
+image_size: 384
+
+# generation configs
+max_length: 20  
+min_length: 5
+num_beams: 3
+prompt: 'a picture of '
+
+# optimizer
+weight_decay: 0.05
+min_lr: 0
+max_epoch: 5
+

extras/BLIP/configs/med_config.json

@@ -0,0 +1,21 @@
+{
+  "architectures": [
+    "BertModel"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "type_vocab_size": 2,
+  "vocab_size": 30524,
+  "encoder_width": 768,
+  "add_cross_attention": true   
+}

extras/BLIP/configs/nlvr.yaml

@@ -0,0 +1,21 @@
+image_root: '/export/share/datasets/vision/NLVR2/' 
+ann_root: 'annotation'
+
+# set pretrained as a file path or an url
+pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_nlvr.pth'
+
+#size of vit model; base or large
+vit: 'base'
+batch_size_train: 16 
+batch_size_test: 64 
+vit_grad_ckpt: False
+vit_ckpt_layer: 0
+max_epoch: 15
+
+image_size: 384
+
+# optimizer
+weight_decay: 0.05
+init_lr: 3e-5
+min_lr: 0
+

extras/BLIP/configs/nocaps.yaml

@@ -0,0 +1,15 @@
+image_root: '/export/share/datasets/vision/nocaps/'
+ann_root: 'annotation'
+
+# set pretrained as a file path or an url
+pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_caption_capfilt_large.pth'
+
+vit: 'base'
+batch_size: 32
+
+image_size: 384
+
+max_length: 20
+min_length: 5
+num_beams: 3
+prompt: 'a picture of '

extras/BLIP/configs/pretrain.yaml

@@ -0,0 +1,27 @@
+train_file: ['/export/share/junnan-li/VL_pretrain/annotation/coco_karpathy_train.json',
+             '/export/share/junnan-li/VL_pretrain/annotation/vg_caption.json',
+             ]
+laion_path: ''   
+
+# size of vit model; base or large
+vit: 'base'
+vit_grad_ckpt: False
+vit_ckpt_layer: 0
+
+image_size: 224
+batch_size: 75
+
+queue_size: 57600
+alpha: 0.4
+
+# optimizer
+weight_decay: 0.05
+init_lr: 3e-4
+min_lr: 1e-6
+warmup_lr: 1e-6
+lr_decay_rate: 0.9
+max_epoch: 20
+warmup_steps: 3000
+
+
+

extras/BLIP/configs/retrieval_coco.yaml

@@ -0,0 +1,34 @@
+image_root: '/export/share/datasets/vision/coco/images/'
+ann_root: 'annotation'
+dataset: 'coco'
+
+# set pretrained as a file path or an url
+pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_retrieval_coco.pth'
+
+# size of vit model; base or large
+
+vit: 'base'
+batch_size_train: 32
+batch_size_test: 64
+vit_grad_ckpt: True
+vit_ckpt_layer: 4
+init_lr: 1e-5
+
+# vit: 'large'
+# batch_size_train: 16
+# batch_size_test: 32
+# vit_grad_ckpt: True
+# vit_ckpt_layer: 12
+# init_lr: 5e-6
+
+image_size: 384
+queue_size: 57600
+alpha: 0.4
+k_test: 256
+negative_all_rank: True
+
+# optimizer
+weight_decay: 0.05
+min_lr: 0
+max_epoch: 6
+

extras/BLIP/configs/retrieval_flickr.yaml

@@ -0,0 +1,34 @@
+image_root: '/export/share/datasets/vision/flickr30k/'
+ann_root: 'annotation'
+dataset: 'flickr'
+
+# set pretrained as a file path or an url
+pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_retrieval_flickr.pth'
+
+# size of vit model; base or large
+
+vit: 'base'
+batch_size_train: 32
+batch_size_test: 64
+vit_grad_ckpt: True
+vit_ckpt_layer: 4
+init_lr: 1e-5
+
+# vit: 'large'
+# batch_size_train: 16
+# batch_size_test: 32
+# vit_grad_ckpt: True
+# vit_ckpt_layer: 10
+# init_lr: 5e-6
+
+image_size: 384
+queue_size: 57600
+alpha: 0.4
+k_test: 128
+negative_all_rank: False
+
+# optimizer
+weight_decay: 0.05
+min_lr: 0
+max_epoch: 6
+

extras/BLIP/configs/retrieval_msrvtt.yaml

@@ -0,0 +1,12 @@
+video_root: '/export/share/dongxuli/data/msrvtt_retrieval/videos'
+ann_root: 'annotation'
+
+# set pretrained as a file path or an url
+pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_retrieval_coco.pth'
+
+# size of vit model; base or large
+vit: 'base'
+batch_size: 64
+k_test: 128
+image_size: 384
+num_frm_test: 8

extras/BLIP/configs/vqa.yaml

@@ -0,0 +1,25 @@
+vqa_root: '/export/share/datasets/vision/VQA/Images/mscoco/' #followed by train2014/
+vg_root: '/export/share/datasets/vision/visual-genome/'  #followed by image/
+train_files: ['vqa_train','vqa_val','vg_qa']
+ann_root: 'annotation'
+
+# set pretrained as a file path or an url
+pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_vqa_capfilt_large.pth'
+
+# size of vit model; base or large
+vit: 'base'
+batch_size_train: 16 
+batch_size_test: 32 
+vit_grad_ckpt: False
+vit_ckpt_layer: 0
+init_lr: 2e-5
+
+image_size: 480
+
+k_test: 128
+inference: 'rank'
+
+# optimizer
+weight_decay: 0.05
+min_lr: 0
+max_epoch: 10

extras/BLIP/models/bert_tokenizer/config.json

@@ -0,0 +1,23 @@
+{
+  "architectures": [
+    "BertForMaskedLM"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "transformers_version": "4.6.0.dev0",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 30522
+}

extras/BLIP/models/bert_tokenizer/tokenizer_config.json

@@ -0,0 +1,3 @@
+{
+  "do_lower_case": true
+}

extras/BLIP/models/blip.py

@@ -0,0 +1,239 @@
+'''
+ * Copyright (c) 2022, salesforce.com, inc.
+ * All rights reserved.
+ * SPDX-License-Identifier: BSD-3-Clause
+ * For full license text, see LICENSE.txt file in the repo root or https://opensource.org/licenses/BSD-3-Clause
+ * By Junnan Li
+'''
+import warnings
+warnings.filterwarnings("ignore")
+
+from extras.BLIP.models.vit import VisionTransformer, interpolate_pos_embed
+from extras.BLIP.models.med import BertConfig, BertModel, BertLMHeadModel
+from transformers import BertTokenizer
+
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+import os
+from urllib.parse import urlparse
+from timm.models.hub import download_cached_file
+
+class BLIP_Base(nn.Module):
+    def __init__(self,                 
+                 med_config = 'configs/med_config.json',  
+                 image_size = 224,
+                 vit = 'base',
+                 vit_grad_ckpt = False,
+                 vit_ckpt_layer = 0,                 
+                 ):
+        """
+        Args:
+            med_config (str): path for the mixture of encoder-decoder model's configuration file
+            image_size (int): input image size
+            vit (str): model size of vision transformer
+        """               
+        super().__init__()
+        
+        self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer)
+        self.tokenizer = init_tokenizer()   
+        med_config = BertConfig.from_json_file(med_config)
+        med_config.encoder_width = vision_width
+        self.text_encoder = BertModel(config=med_config, add_pooling_layer=False)  
+
+        
+    def forward(self, image, caption, mode):
+        
+        assert mode in ['image', 'text', 'multimodal'], "mode parameter must be image, text, or multimodal"
+        text = self.tokenizer(caption, return_tensors="pt").to(image.device) 
+        
+        if mode=='image':    
+            # return image features
+            image_embeds = self.visual_encoder(image)             
+            return image_embeds
+        
+        elif mode=='text':
+            # return text features
+            text_output = self.text_encoder(text.input_ids, attention_mask = text.attention_mask,                      
+                                            return_dict = True, mode = 'text')  
+            return text_output.last_hidden_state
+        
+        elif mode=='multimodal':
+            # return multimodel features
+            image_embeds = self.visual_encoder(image)    
+            image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)      
+            
+            text.input_ids[:,0] = self.tokenizer.enc_token_id
+            output = self.text_encoder(text.input_ids,
+                                       attention_mask = text.attention_mask,
+                                       encoder_hidden_states = image_embeds,
+                                       encoder_attention_mask = image_atts,      
+                                       return_dict = True,
+                                      )              
+            return output.last_hidden_state
+        
+        
+        
+class BLIP_Decoder(nn.Module):
+    def __init__(self,                 
+                 med_config = 'configs/med_config.json',  
+                 image_size = 384,
+                 vit = 'base',
+                 vit_grad_ckpt = False,
+                 vit_ckpt_layer = 0,
+                 prompt = 'a picture of ',
+                 ):
+        """
+        Args:
+            med_config (str): path for the mixture of encoder-decoder model's configuration file
+            image_size (int): input image size
+            vit (str): model size of vision transformer
+        """            
+        super().__init__()
+        
+        self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer)
+        self.tokenizer = init_tokenizer()   
+        med_config = BertConfig.from_json_file(med_config)
+        med_config.encoder_width = vision_width
+        self.text_decoder = BertLMHeadModel(config=med_config)    
+        
+        self.prompt = prompt
+        self.prompt_length = len(self.tokenizer(self.prompt).input_ids)-1
+
+        
+    def forward(self, image, caption):
+        
+        image_embeds = self.visual_encoder(image) 
+        image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)
+        
+        text = self.tokenizer(caption, padding='longest', truncation=True, max_length=40, return_tensors="pt").to(image.device) 
+        
+        text.input_ids[:,0] = self.tokenizer.bos_token_id
+        
+        decoder_targets = text.input_ids.masked_fill(text.input_ids == self.tokenizer.pad_token_id, -100)         
+        decoder_targets[:,:self.prompt_length] = -100
+     
+        decoder_output = self.text_decoder(text.input_ids, 
+                                           attention_mask = text.attention_mask, 
+                                           encoder_hidden_states = image_embeds,
+                                           encoder_attention_mask = image_atts,                  
+                                           labels = decoder_targets,
+                                           return_dict = True,   
+                                          )   
+        loss_lm = decoder_output.loss
+        
+        return loss_lm
+        
+    def generate(self, image, sample=False, num_beams=3, max_length=30, min_length=10, top_p=0.9, repetition_penalty=1.0):
+        image_embeds = self.visual_encoder(image)
+
+        if not sample:
+            image_embeds = image_embeds.repeat_interleave(num_beams,dim=0)
+            
+        image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)
+        model_kwargs = {"encoder_hidden_states": image_embeds, "encoder_attention_mask":image_atts}
+        
+        prompt = [self.prompt] * image.size(0)
+        input_ids = self.tokenizer(prompt, return_tensors="pt").input_ids.to(image.device) 
+        input_ids[:,0] = self.tokenizer.bos_token_id
+        input_ids = input_ids[:, :-1] 
+
+        if sample:
+            #nucleus sampling
+            outputs = self.text_decoder.generate(input_ids=input_ids,
+                                                  max_length=max_length,
+                                                  min_length=min_length,
+                                                  do_sample=True,
+                                                  top_p=top_p,
+                                                  num_return_sequences=1,
+                                                  eos_token_id=self.tokenizer.sep_token_id,
+                                                  pad_token_id=self.tokenizer.pad_token_id, 
+                                                  repetition_penalty=1.1,                                            
+                                                  **model_kwargs)
+        else:
+            #beam search
+            outputs = self.text_decoder.generate(input_ids=input_ids,
+                                                  max_length=max_length,
+                                                  min_length=min_length,
+                                                  num_beams=num_beams,
+                                                  eos_token_id=self.tokenizer.sep_token_id,
+                                                  pad_token_id=self.tokenizer.pad_token_id,     
+                                                  repetition_penalty=repetition_penalty,
+                                                  **model_kwargs)            
+            
+        captions = []    
+        for output in outputs:
+            caption = self.tokenizer.decode(output, skip_special_tokens=True)    
+            captions.append(caption[len(self.prompt):])
+        return captions
+    
+
+def blip_decoder(pretrained='',**kwargs):
+    model = BLIP_Decoder(**kwargs)
+    if pretrained:
+        model,msg = load_checkpoint(model,pretrained)
+        assert(len(msg.missing_keys)==0)
+    return model    
+    
+def blip_feature_extractor(pretrained='',**kwargs):
+    model = BLIP_Base(**kwargs)
+    if pretrained:
+        model,msg = load_checkpoint(model,pretrained)
+        assert(len(msg.missing_keys)==0)
+    return model        
+
+def init_tokenizer():
+    tokenizer_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), "bert_tokenizer")
+    tokenizer = BertTokenizer.from_pretrained(tokenizer_path)
+    tokenizer.add_special_tokens({'bos_token':'[DEC]'})
+    tokenizer.add_special_tokens({'additional_special_tokens':['[ENC]']})       
+    tokenizer.enc_token_id = tokenizer.additional_special_tokens_ids[0]  
+    return tokenizer
+
+
+def create_vit(vit, image_size, use_grad_checkpointing=False, ckpt_layer=0, drop_path_rate=0):
+        
+    assert vit in ['base', 'large'], "vit parameter must be base or large"
+    if vit=='base':
+        vision_width = 768
+        visual_encoder = VisionTransformer(img_size=image_size, patch_size=16, embed_dim=vision_width, depth=12, 
+                                           num_heads=12, use_grad_checkpointing=use_grad_checkpointing, ckpt_layer=ckpt_layer,
+                                           drop_path_rate=0 or drop_path_rate
+                                          )   
+    elif vit=='large':
+        vision_width = 1024
+        visual_encoder = VisionTransformer(img_size=image_size, patch_size=16, embed_dim=vision_width, depth=24, 
+                                           num_heads=16, use_grad_checkpointing=use_grad_checkpointing, ckpt_layer=ckpt_layer,
+                                           drop_path_rate=0.1 or drop_path_rate
+                                          )   
+    return visual_encoder, vision_width
+
+def is_url(url_or_filename):
+    parsed = urlparse(url_or_filename)
+    return parsed.scheme in ("http", "https")
+
+def load_checkpoint(model,url_or_filename):
+    if is_url(url_or_filename):
+        cached_file = download_cached_file(url_or_filename, check_hash=False, progress=True)
+        checkpoint = torch.load(cached_file, map_location='cpu') 
+    elif os.path.isfile(url_or_filename):        
+        checkpoint = torch.load(url_or_filename, map_location='cpu') 
+    else:
+        raise RuntimeError('checkpoint url or path is invalid')
+        
+    state_dict = checkpoint['model']
+    
+    state_dict['visual_encoder.pos_embed'] = interpolate_pos_embed(state_dict['visual_encoder.pos_embed'],model.visual_encoder) 
+    if 'visual_encoder_m.pos_embed' in model.state_dict().keys():
+        state_dict['visual_encoder_m.pos_embed'] = interpolate_pos_embed(state_dict['visual_encoder_m.pos_embed'],
+                                                                         model.visual_encoder_m)    
+    for key in model.state_dict().keys():
+        if key in state_dict.keys():
+            if state_dict[key].shape!=model.state_dict()[key].shape:
+                del state_dict[key]
+    
+    msg = model.load_state_dict(state_dict,strict=False)
+    print('load checkpoint from %s'%url_or_filename)  
+    return model,msg
+    

extras/BLIP/models/blip_itm.py

@@ -0,0 +1,76 @@
+from extras.BLIP.models.med import BertConfig, BertModel
+from transformers import BertTokenizer
+
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from extras.BLIP.models.blip import create_vit, init_tokenizer, load_checkpoint
+
+class BLIP_ITM(nn.Module):
+    def __init__(self,                 
+                 med_config = 'configs/med_config.json',  
+                 image_size = 384,
+                 vit = 'base',
+                 vit_grad_ckpt = False,
+                 vit_ckpt_layer = 0,                      
+                 embed_dim = 256,     
+                 ):
+        """
+        Args:
+            med_config (str): path for the mixture of encoder-decoder model's configuration file
+            image_size (int): input image size
+            vit (str): model size of vision transformer
+        """               
+        super().__init__()
+        
+        self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer)
+        self.tokenizer = init_tokenizer()   
+        med_config = BertConfig.from_json_file(med_config)
+        med_config.encoder_width = vision_width
+        self.text_encoder = BertModel(config=med_config, add_pooling_layer=False)          
+
+        text_width = self.text_encoder.config.hidden_size
+        
+        self.vision_proj = nn.Linear(vision_width, embed_dim)
+        self.text_proj = nn.Linear(text_width, embed_dim)
+
+        self.itm_head = nn.Linear(text_width, 2) 
+        
+        
+    def forward(self, image, caption, match_head='itm'):
+
+        image_embeds = self.visual_encoder(image) 
+        image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)        
+      
+        text = self.tokenizer(caption, padding='max_length', truncation=True, max_length=35, 
+                              return_tensors="pt").to(image.device) 
+
+                 
+        if match_head=='itm':
+            output = self.text_encoder(text.input_ids,
+                                       attention_mask = text.attention_mask,
+                                       encoder_hidden_states = image_embeds,
+                                       encoder_attention_mask = image_atts,      
+                                       return_dict = True,
+                                      )
+            itm_output = self.itm_head(output.last_hidden_state[:,0,:])     
+            return itm_output
+            
+        elif match_head=='itc':
+            text_output = self.text_encoder(text.input_ids, attention_mask = text.attention_mask,                      
+                                            return_dict = True, mode = 'text')                     
+            image_feat = F.normalize(self.vision_proj(image_embeds[:,0,:]),dim=-1)   
+            text_feat = F.normalize(self.text_proj(text_output.last_hidden_state[:,0,:]),dim=-1)    
+            
+            sim = image_feat @ text_feat.t()
+            return sim
+        
+        
+def blip_itm(pretrained='',**kwargs):
+    model = BLIP_ITM(**kwargs)
+    if pretrained:
+        model,msg = load_checkpoint(model,pretrained)
+        assert(len(msg.missing_keys)==0)
+    return model         
+            

extras/BLIP/models/blip_nlvr.py

@@ -0,0 +1,105 @@
+from extras.BLIP.models.med import BertConfig
+from extras.BLIP.models.nlvr_encoder import BertModel
+from extras.BLIP.models.vit import interpolate_pos_embed
+from extras.BLIP.models.blip import create_vit, init_tokenizer, is_url
+
+from timm.models.hub import download_cached_file
+
+import torch
+from torch import nn
+import torch.nn.functional as F
+from transformers import BertTokenizer
+import numpy as np
+import os
+
+
+class BLIP_NLVR(nn.Module):
+    def __init__(self,                 
+                 med_config = 'configs/med_config.json',  
+                 image_size = 480,
+                 vit = 'base',
+                 vit_grad_ckpt = False,
+                 vit_ckpt_layer = 0,                   
+                 ):
+        """
+        Args:
+            med_config (str): path for the mixture of encoder-decoder model's configuration file
+            image_size (int): input image size
+            vit (str): model size of vision transformer
+        """               
+        super().__init__()
+        
+        self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer, drop_path_rate=0.1)
+        self.tokenizer = init_tokenizer()   
+        med_config = BertConfig.from_json_file(med_config)
+        med_config.encoder_width = vision_width
+        self.text_encoder = BertModel(config=med_config, add_pooling_layer=False) 
+                    
+        self.cls_head = nn.Sequential(
+                  nn.Linear(self.text_encoder.config.hidden_size, self.text_encoder.config.hidden_size),
+                  nn.ReLU(),
+                  nn.Linear(self.text_encoder.config.hidden_size, 2)
+                )  
+
+    def forward(self, image, text, targets, train=True):
+        
+        image_embeds = self.visual_encoder(image) 
+        image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)        
+        image0_embeds, image1_embeds = torch.split(image_embeds,targets.size(0))     
+
+        text = self.tokenizer(text, padding='longest', return_tensors="pt").to(image.device) 
+        text.input_ids[:,0] = self.tokenizer.enc_token_id        
+
+        output = self.text_encoder(text.input_ids, 
+                                   attention_mask = text.attention_mask, 
+                                   encoder_hidden_states = [image0_embeds,image1_embeds],
+                                   encoder_attention_mask = [image_atts[:image0_embeds.size(0)],
+                                                             image_atts[image0_embeds.size(0):]],        
+                                   return_dict = True,
+                                  )  
+        hidden_state = output.last_hidden_state[:,0,:]        
+        prediction = self.cls_head(hidden_state)
+
+        if train:            
+            loss = F.cross_entropy(prediction, targets)   
+            return loss
+        else:
+            return prediction
+    
+def blip_nlvr(pretrained='',**kwargs):
+    model = BLIP_NLVR(**kwargs)
+    if pretrained:
+        model,msg = load_checkpoint(model,pretrained)
+        print("missing keys:")
+        print(msg.missing_keys)
+    return model  
+
+        
+def load_checkpoint(model,url_or_filename):
+    if is_url(url_or_filename):
+        cached_file = download_cached_file(url_or_filename, check_hash=False, progress=True)
+        checkpoint = torch.load(cached_file, map_location='cpu') 
+    elif os.path.isfile(url_or_filename):        
+        checkpoint = torch.load(url_or_filename, map_location='cpu') 
+    else:
+        raise RuntimeError('checkpoint url or path is invalid')
+    state_dict = checkpoint['model']
+    
+    state_dict['visual_encoder.pos_embed'] = interpolate_pos_embed(state_dict['visual_encoder.pos_embed'],model.visual_encoder) 
+    
+    for key in list(state_dict.keys()):
+        if 'crossattention.self.' in key:
+            new_key0 = key.replace('self','self0')
+            new_key1 = key.replace('self','self1')
+            state_dict[new_key0] = state_dict[key]
+            state_dict[new_key1] = state_dict[key]
+        elif 'crossattention.output.dense.' in key:
+            new_key0 = key.replace('dense','dense0')
+            new_key1 = key.replace('dense','dense1')
+            state_dict[new_key0] = state_dict[key]
+            state_dict[new_key1] = state_dict[key]  
+                
+    msg = model.load_state_dict(state_dict,strict=False)
+    print('load checkpoint from %s'%url_or_filename)  
+    return model,msg
+            

extras/BLIP/models/blip_pretrain.py

@@ -0,0 +1,339 @@
+'''
+ * Copyright (c) 2022, salesforce.com, inc.
+ * All rights reserved.
+ * SPDX-License-Identifier: BSD-3-Clause
+ * For full license text, see LICENSE.txt file in the repo root or https://opensource.org/licenses/BSD-3-Clause
+ * By Junnan Li
+'''
+from extras.BLIP.models.med import BertConfig, BertModel, BertLMHeadModel
+from transformers import BertTokenizer
+import transformers
+transformers.logging.set_verbosity_error()
+
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from extras.BLIP.models.blip import create_vit, init_tokenizer, load_checkpoint
+
+class BLIP_Pretrain(nn.Module):
+    def __init__(self,                 
+                 med_config = 'configs/bert_config.json',  
+                 image_size = 224,
+                 vit = 'base',
+                 vit_grad_ckpt = False,
+                 vit_ckpt_layer = 0,                    
+                 embed_dim = 256,     
+                 queue_size = 57600,
+                 momentum = 0.995,
+                 ):
+        """
+        Args:
+            med_config (str): path for the mixture of encoder-decoder model's configuration file
+            image_size (int): input image size
+            vit (str): model size of vision transformer
+        """               
+        super().__init__()
+        
+        self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer, 0)
+        
+        if vit=='base':
+            checkpoint = torch.hub.load_state_dict_from_url(
+                url="https://dl.fbaipublicfiles.com/deit/deit_base_patch16_224-b5f2ef4d.pth",
+                map_location="cpu", check_hash=True)
+            state_dict = checkpoint["model"]     
+            msg = self.visual_encoder.load_state_dict(state_dict,strict=False)
+        elif vit=='large':
+            from timm.models.helpers import load_custom_pretrained
+            from timm.models.vision_transformer import default_cfgs
+            load_custom_pretrained(self.visual_encoder,default_cfgs['vit_large_patch16_224_in21k'])        
+               
+        self.tokenizer = init_tokenizer()   
+        encoder_config = BertConfig.from_json_file(med_config)
+        encoder_config.encoder_width = vision_width
+        self.text_encoder = BertModel.from_pretrained('bert-base-uncased',config=encoder_config, add_pooling_layer=False)
+        self.text_encoder.resize_token_embeddings(len(self.tokenizer)) 
+
+        text_width = self.text_encoder.config.hidden_size
+        
+        self.vision_proj = nn.Linear(vision_width, embed_dim)
+        self.text_proj = nn.Linear(text_width, embed_dim)
+
+        self.itm_head = nn.Linear(text_width, 2) 
+        
+        # create momentum encoders  
+        self.visual_encoder_m, vision_width = create_vit(vit,image_size)              
+        self.vision_proj_m = nn.Linear(vision_width, embed_dim)
+        self.text_encoder_m = BertModel(config=encoder_config, add_pooling_layer=False)      
+        self.text_proj_m = nn.Linear(text_width, embed_dim)
+        
+        self.model_pairs = [[self.visual_encoder,self.visual_encoder_m],
+                            [self.vision_proj,self.vision_proj_m],
+                            [self.text_encoder,self.text_encoder_m],
+                            [self.text_proj,self.text_proj_m],
+                           ]       
+        self.copy_params()
+
+        # create the queue
+        self.register_buffer("image_queue", torch.randn(embed_dim, queue_size))
+        self.register_buffer("text_queue", torch.randn(embed_dim, queue_size))
+        self.register_buffer("queue_ptr", torch.zeros(1, dtype=torch.long))  
+
+        self.image_queue = nn.functional.normalize(self.image_queue, dim=0)
+        self.text_queue = nn.functional.normalize(self.text_queue, dim=0)
+        
+        self.queue_size = queue_size
+        self.momentum = momentum
+        self.temp = nn.Parameter(0.07*torch.ones([]))   
+        
+        # create the decoder
+        decoder_config = BertConfig.from_json_file(med_config)
+        decoder_config.encoder_width = vision_width        
+        self.text_decoder = BertLMHeadModel.from_pretrained('bert-base-uncased',config=decoder_config)    
+        self.text_decoder.resize_token_embeddings(len(self.tokenizer)) 
+        tie_encoder_decoder_weights(self.text_encoder,self.text_decoder.bert,'','/attention')
+        
+        
+    def forward(self, image, caption, alpha):
+        with torch.no_grad():
+            self.temp.clamp_(0.001,0.5)
+        
+        image_embeds = self.visual_encoder(image) 
+        image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)        
+        image_feat = F.normalize(self.vision_proj(image_embeds[:,0,:]),dim=-1)          
+        
+        text = self.tokenizer(caption, padding='max_length', truncation=True, max_length=30, 
+                              return_tensors="pt").to(image.device)  
+        text_output = self.text_encoder(text.input_ids, attention_mask = text.attention_mask,                      
+                                        return_dict = True, mode = 'text')            
+        text_feat = F.normalize(self.text_proj(text_output.last_hidden_state[:,0,:]),dim=-1)                 
+             
+        # get momentum features
+        with torch.no_grad():
+            self._momentum_update()
+            image_embeds_m = self.visual_encoder_m(image) 
+            image_feat_m = F.normalize(self.vision_proj_m(image_embeds_m[:,0,:]),dim=-1)  
+            image_feat_all = torch.cat([image_feat_m.t(),self.image_queue.clone().detach()],dim=1)                   
+            
+            text_output_m = self.text_encoder_m(text.input_ids, attention_mask = text.attention_mask,                      
+                                                return_dict = True, mode = 'text')    
+            text_feat_m = F.normalize(self.text_proj_m(text_output_m.last_hidden_state[:,0,:]),dim=-1) 
+            text_feat_all = torch.cat([text_feat_m.t(),self.text_queue.clone().detach()],dim=1)
+
+            sim_i2t_m = image_feat_m @ text_feat_all / self.temp  
+            sim_t2i_m = text_feat_m @ image_feat_all / self.temp 
+
+            sim_targets = torch.zeros(sim_i2t_m.size()).to(image.device)
+            sim_targets.fill_diagonal_(1)          
+
+            sim_i2t_targets = alpha * F.softmax(sim_i2t_m, dim=1) + (1 - alpha) * sim_targets
+            sim_t2i_targets = alpha * F.softmax(sim_t2i_m, dim=1) + (1 - alpha) * sim_targets        
+
+        sim_i2t = image_feat @ text_feat_all / self.temp
+        sim_t2i = text_feat @ image_feat_all / self.temp
+                             
+        loss_i2t = -torch.sum(F.log_softmax(sim_i2t, dim=1)*sim_i2t_targets,dim=1).mean()
+        loss_t2i = -torch.sum(F.log_softmax(sim_t2i, dim=1)*sim_t2i_targets,dim=1).mean() 
+
+        loss_ita = (loss_i2t+loss_t2i)/2
+
+        self._dequeue_and_enqueue(image_feat_m, text_feat_m)        
+
+        ###============== Image-text Matching ===================###
+        encoder_input_ids = text.input_ids.clone()
+        encoder_input_ids[:,0] = self.tokenizer.enc_token_id
+        
+        # forward the positve image-text pair
+        bs = image.size(0)
+        output_pos = self.text_encoder(encoder_input_ids,
+                                       attention_mask = text.attention_mask,
+                                       encoder_hidden_states = image_embeds,
+                                       encoder_attention_mask = image_atts,      
+                                       return_dict = True,
+                                      )            
+        with torch.no_grad():       
+            weights_t2i = F.softmax(sim_t2i[:,:bs],dim=1)+1e-4 
+            weights_t2i.fill_diagonal_(0)            
+            weights_i2t = F.softmax(sim_i2t[:,:bs],dim=1)+1e-4  
+            weights_i2t.fill_diagonal_(0)   
+            
+        # select a negative image for each text
+        image_embeds_neg = []    
+        for b in range(bs):
+            neg_idx = torch.multinomial(weights_t2i[b], 1).item()
+            image_embeds_neg.append(image_embeds[neg_idx])
+        image_embeds_neg = torch.stack(image_embeds_neg,dim=0)   
+
+        # select a negative text for each image
+        text_ids_neg = []
+        text_atts_neg = []
+        for b in range(bs):
+            neg_idx = torch.multinomial(weights_i2t[b], 1).item()
+            text_ids_neg.append(encoder_input_ids[neg_idx])
+            text_atts_neg.append(text.attention_mask[neg_idx])
+
+        text_ids_neg = torch.stack(text_ids_neg,dim=0)   
+        text_atts_neg = torch.stack(text_atts_neg,dim=0)      
+
+        text_ids_all = torch.cat([encoder_input_ids, text_ids_neg],dim=0)     
+        text_atts_all = torch.cat([text.attention_mask, text_atts_neg],dim=0)     
+
+        image_embeds_all = torch.cat([image_embeds_neg,image_embeds],dim=0)
+        image_atts_all = torch.cat([image_atts,image_atts],dim=0)
+
+        output_neg = self.text_encoder(text_ids_all,
+                                       attention_mask = text_atts_all,
+                                       encoder_hidden_states = image_embeds_all,
+                                       encoder_attention_mask = image_atts_all,      
+                                       return_dict = True,
+                                      )                            
+
+        vl_embeddings = torch.cat([output_pos.last_hidden_state[:,0,:], output_neg.last_hidden_state[:,0,:]],dim=0)
+        vl_output = self.itm_head(vl_embeddings)            
+
+        itm_labels = torch.cat([torch.ones(bs,dtype=torch.long),torch.zeros(2*bs,dtype=torch.long)],
+                               dim=0).to(image.device)
+        loss_itm = F.cross_entropy(vl_output, itm_labels)  
+        
+        ##================= LM ========================##     
+        decoder_input_ids = text.input_ids.clone()      
+        decoder_input_ids[:,0] = self.tokenizer.bos_token_id
+        decoder_targets = decoder_input_ids.masked_fill(decoder_input_ids == self.tokenizer.pad_token_id, -100) 
+
+        decoder_output = self.text_decoder(decoder_input_ids, 
+                                           attention_mask = text.attention_mask, 
+                                           encoder_hidden_states = image_embeds,
+                                           encoder_attention_mask = image_atts,                  
+                                           labels = decoder_targets,
+                                           return_dict = True,   
+                                          )   
+          
+        loss_lm = decoder_output.loss                
+        return loss_ita, loss_itm, loss_lm
+ 
+
+
+    @torch.no_grad()    
+    def copy_params(self):
+        for model_pair in self.model_pairs:           
+            for param, param_m in zip(model_pair[0].parameters(), model_pair[1].parameters()):
+                param_m.data.copy_(param.data)  # initialize
+                param_m.requires_grad = False  # not update by gradient    
+
+            
+    @torch.no_grad()        
+    def _momentum_update(self):
+        for model_pair in self.model_pairs:           
+            for param, param_m in zip(model_pair[0].parameters(), model_pair[1].parameters()):
+                param_m.data = param_m.data * self.momentum + param.data * (1. - self.momentum)
+
+                        
+    @torch.no_grad()
+    def _dequeue_and_enqueue(self, image_feat, text_feat):
+        # gather keys before updating queue
+        image_feats = concat_all_gather(image_feat)
+        text_feats = concat_all_gather(text_feat)
+
+        batch_size = image_feats.shape[0]
+
+        ptr = int(self.queue_ptr)
+        assert self.queue_size % batch_size == 0  # for simplicity
+
+        # replace the keys at ptr (dequeue and enqueue)
+        self.image_queue[:, ptr:ptr + batch_size] = image_feats.T
+        self.text_queue[:, ptr:ptr + batch_size] = text_feats.T
+        ptr = (ptr + batch_size) % self.queue_size  # move pointer
+
+        self.queue_ptr[0] = ptr 
+
+
+def blip_pretrain(**kwargs):
+    model = BLIP_Pretrain(**kwargs)
+    return model 
+
+
+@torch.no_grad()
+def concat_all_gather(tensor):
+    """
+    Performs all_gather operation on the provided tensors.
+    *** Warning ***: torch.distributed.all_gather has no gradient.
+    """
+    tensors_gather = [torch.ones_like(tensor)
+        for _ in range(torch.distributed.get_world_size())]
+    torch.distributed.all_gather(tensors_gather, tensor, async_op=False)
+
+    output = torch.cat(tensors_gather, dim=0)
+    return output     
+
+
+from typing import List
+def tie_encoder_decoder_weights(encoder: nn.Module, decoder: nn.Module, base_model_prefix: str, skip_key:str):
+    uninitialized_encoder_weights: List[str] = []
+    if decoder.__class__ != encoder.__class__:
+        print(
+            f"{decoder.__class__} and {encoder.__class__} are not equal. In this case make sure that all encoder weights are correctly initialized."
+        )
+
+    def tie_encoder_to_decoder_recursively(
+        decoder_pointer: nn.Module,
+        encoder_pointer: nn.Module,
+        module_name: str,
+        uninitialized_encoder_weights: List[str],
+        skip_key: str,
+        depth=0,
+    ):
+        assert isinstance(decoder_pointer, nn.Module) and isinstance(
+            encoder_pointer, nn.Module
+        ), f"{decoder_pointer} and {encoder_pointer} have to be of type torch.nn.Module"
+        if hasattr(decoder_pointer, "weight") and skip_key not in module_name:
+            assert hasattr(encoder_pointer, "weight")
+            encoder_pointer.weight = decoder_pointer.weight
+            if hasattr(decoder_pointer, "bias"):
+                assert hasattr(encoder_pointer, "bias")
+                encoder_pointer.bias = decoder_pointer.bias                
+            print(module_name+' is tied')    
+            return
+
+        encoder_modules = encoder_pointer._modules
+        decoder_modules = decoder_pointer._modules
+        if len(decoder_modules) > 0:
+            assert (
+                len(encoder_modules) > 0
+            ), f"Encoder module {encoder_pointer} does not match decoder module {decoder_pointer}"
+
+            all_encoder_weights = set([module_name + "/" + sub_name for sub_name in encoder_modules.keys()])
+            encoder_layer_pos = 0
+            for name, module in decoder_modules.items():
+                if name.isdigit():
+                    encoder_name = str(int(name) + encoder_layer_pos)
+                    decoder_name = name
+                    if not isinstance(decoder_modules[decoder_name], type(encoder_modules[encoder_name])) and len(
+                        encoder_modules
+                    ) != len(decoder_modules):
+                        # this can happen if the name corresponds to the position in a list module list of layers
+                        # in this case the decoder has added a cross-attention that the encoder does not have
+                        # thus skip this step and subtract one layer pos from encoder
+                        encoder_layer_pos -= 1
+                        continue
+                elif name not in encoder_modules:
+                    continue
+                elif depth > 500:
+                    raise ValueError(
+                        "Max depth of recursive function `tie_encoder_to_decoder` reached. It seems that there is a circular dependency between two or more `nn.Modules` of your model."
+                    )
+                else:
+                    decoder_name = encoder_name = name
+                tie_encoder_to_decoder_recursively(
+                    decoder_modules[decoder_name],
+                    encoder_modules[encoder_name],
+                    module_name + "/" + name,
+                    uninitialized_encoder_weights,
+                    skip_key,
+                    depth=depth + 1,
+                )
+                all_encoder_weights.remove(module_name + "/" + encoder_name)
+
+            uninitialized_encoder_weights += list(all_encoder_weights)
+
+    # tie weights recursively
+    tie_encoder_to_decoder_recursively(decoder, encoder, base_model_prefix, uninitialized_encoder_weights, skip_key)  

extras/BLIP/models/blip_retrieval.py

@@ -0,0 +1,319 @@
+from extras.BLIP.models.med import BertConfig, BertModel
+from transformers import BertTokenizer
+
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from extras.BLIP.models.blip import create_vit, init_tokenizer, load_checkpoint
+
+class BLIP_Retrieval(nn.Module):
+    def __init__(self,                 
+                 med_config = 'configs/med_config.json',  
+                 image_size = 384,
+                 vit = 'base',
+                 vit_grad_ckpt = False,
+                 vit_ckpt_layer = 0,                      
+                 embed_dim = 256,     
+                 queue_size = 57600,
+                 momentum = 0.995,
+                 negative_all_rank = False,
+                 ):
+        """
+        Args:
+            med_config (str): path for the mixture of encoder-decoder model's configuration file
+            image_size (int): input image size
+            vit (str): model size of vision transformer
+        """               
+        super().__init__()
+        
+        self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer)
+        self.tokenizer = init_tokenizer()   
+        med_config = BertConfig.from_json_file(med_config)
+        med_config.encoder_width = vision_width
+        self.text_encoder = BertModel(config=med_config, add_pooling_layer=False)          
+
+        text_width = self.text_encoder.config.hidden_size
+        
+        self.vision_proj = nn.Linear(vision_width, embed_dim)
+        self.text_proj = nn.Linear(text_width, embed_dim)
+
+        self.itm_head = nn.Linear(text_width, 2) 
+        
+        # create momentum encoders  
+        self.visual_encoder_m, vision_width = create_vit(vit,image_size)              
+        self.vision_proj_m = nn.Linear(vision_width, embed_dim)
+        self.text_encoder_m = BertModel(config=med_config, add_pooling_layer=False)    
+        self.text_proj_m = nn.Linear(text_width, embed_dim)
+        
+        self.model_pairs = [[self.visual_encoder,self.visual_encoder_m],
+                            [self.vision_proj,self.vision_proj_m],
+                            [self.text_encoder,self.text_encoder_m],
+                            [self.text_proj,self.text_proj_m],
+                           ]       
+        self.copy_params()
+
+        # create the queue
+        self.register_buffer("image_queue", torch.randn(embed_dim, queue_size))
+        self.register_buffer("text_queue", torch.randn(embed_dim, queue_size))
+        self.register_buffer("idx_queue", torch.full((1,queue_size),-100))
+        self.register_buffer("ptr_queue", torch.zeros(1, dtype=torch.long))  
+
+        self.image_queue = nn.functional.normalize(self.image_queue, dim=0)
+        self.text_queue = nn.functional.normalize(self.text_queue, dim=0)
+        
+        self.queue_size = queue_size
+        self.momentum = momentum
+        self.temp = nn.Parameter(0.07*torch.ones([]))   
+        
+        self.negative_all_rank = negative_all_rank
+        
+        
+    def forward(self, image, caption, alpha, idx):
+        with torch.no_grad():
+            self.temp.clamp_(0.001,0.5)
+        
+        image_embeds = self.visual_encoder(image) 
+        image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)        
+        image_feat = F.normalize(self.vision_proj(image_embeds[:,0,:]),dim=-1)    
+        
+        text = self.tokenizer(caption, padding='max_length', truncation=True, max_length=35, 
+                              return_tensors="pt").to(image.device) 
+        
+        text_output = self.text_encoder(text.input_ids, attention_mask = text.attention_mask,                      
+                                        return_dict = True, mode = 'text')            
+        text_feat = F.normalize(self.text_proj(text_output.last_hidden_state[:,0,:]),dim=-1)        
+        
+        ###============== Image-text Contrastive Learning ===================###
+        idx = idx.view(-1,1)
+        idx_all = torch.cat([idx.t(), self.idx_queue.clone().detach()],dim=1)  
+        pos_idx = torch.eq(idx, idx_all).float()       
+        sim_targets = pos_idx / pos_idx.sum(1,keepdim=True)   
+        
+        # get momentum features
+        with torch.no_grad():
+            self._momentum_update()
+            image_embeds_m = self.visual_encoder_m(image) 
+            image_feat_m = F.normalize(self.vision_proj_m(image_embeds_m[:,0,:]),dim=-1)  
+            image_feat_m_all = torch.cat([image_feat_m.t(),self.image_queue.clone().detach()],dim=1)                   
+            
+            text_output_m = self.text_encoder_m(text.input_ids, attention_mask = text.attention_mask,                      
+                                                return_dict = True, mode = 'text')    
+            text_feat_m = F.normalize(self.text_proj_m(text_output_m.last_hidden_state[:,0,:]),dim=-1) 
+            text_feat_m_all = torch.cat([text_feat_m.t(),self.text_queue.clone().detach()],dim=1)
+
+            sim_i2t_m = image_feat_m @ text_feat_m_all / self.temp  
+            sim_t2i_m = text_feat_m @ image_feat_m_all / self.temp   
+
+            sim_i2t_targets = alpha * F.softmax(sim_i2t_m, dim=1) + (1 - alpha) * sim_targets
+            sim_t2i_targets = alpha * F.softmax(sim_t2i_m, dim=1) + (1 - alpha) * sim_targets        
+
+        sim_i2t = image_feat @ text_feat_m_all / self.temp 
+        sim_t2i = text_feat @ image_feat_m_all / self.temp 
+                             
+        loss_i2t = -torch.sum(F.log_softmax(sim_i2t, dim=1)*sim_i2t_targets,dim=1).mean()
+        loss_t2i = -torch.sum(F.log_softmax(sim_t2i, dim=1)*sim_t2i_targets,dim=1).mean() 
+
+        loss_ita = (loss_i2t+loss_t2i)/2
+        
+        idxs = concat_all_gather(idx)
+        self._dequeue_and_enqueue(image_feat_m, text_feat_m, idxs)        
+
+        ###============== Image-text Matching ===================###
+        encoder_input_ids = text.input_ids.clone()
+        encoder_input_ids[:,0] = self.tokenizer.enc_token_id
+
+        # forward the positve image-text pair
+        bs = image.size(0)
+        output_pos = self.text_encoder(encoder_input_ids,
+                                       attention_mask = text.attention_mask,
+                                       encoder_hidden_states = image_embeds,
+                                       encoder_attention_mask = image_atts,      
+                                       return_dict = True,
+                                      )  
+        
+        
+        if self.negative_all_rank:    
+            # compute sample similarity
+            with torch.no_grad():                
+                mask = torch.eq(idx, idxs.t())
+
+                image_feat_world = concat_all_gather(image_feat)
+                text_feat_world = concat_all_gather(text_feat)
+
+                sim_i2t = image_feat @ text_feat_world.t() / self.temp 
+                sim_t2i = text_feat @ image_feat_world.t() / self.temp 
+
+                weights_i2t = F.softmax(sim_i2t,dim=1)
+                weights_i2t.masked_fill_(mask, 0)            
+
+                weights_t2i = F.softmax(sim_t2i,dim=1)
+                weights_t2i.masked_fill_(mask, 0)     
+
+            image_embeds_world = all_gather_with_grad(image_embeds) 
+
+            # select a negative image (from all ranks) for each text
+            image_embeds_neg = []    
+            for b in range(bs):
+                neg_idx = torch.multinomial(weights_t2i[b], 1).item()
+                image_embeds_neg.append(image_embeds_world[neg_idx])
+            image_embeds_neg = torch.stack(image_embeds_neg,dim=0)   
+
+            # select a negative text (from all ranks) for each image
+            input_ids_world = concat_all_gather(encoder_input_ids)
+            att_mask_world = concat_all_gather(text.attention_mask)        
+
+            text_ids_neg = []
+            text_atts_neg = []
+            for b in range(bs):
+                neg_idx = torch.multinomial(weights_i2t[b], 1).item()
+                text_ids_neg.append(input_ids_world[neg_idx])
+                text_atts_neg.append(att_mask_world[neg_idx])
+                
+        else:
+            with torch.no_grad():                
+                mask = torch.eq(idx, idx.t())
+                
+                sim_i2t = image_feat @ text_feat.t() / self.temp 
+                sim_t2i = text_feat @ image_feat.t() / self.temp 
+
+                weights_i2t = F.softmax(sim_i2t,dim=1)
+                weights_i2t.masked_fill_(mask, 0)            
+
+                weights_t2i = F.softmax(sim_t2i,dim=1)
+                weights_t2i.masked_fill_(mask, 0)     
+
+            # select a negative image (from same rank) for each text
+            image_embeds_neg = []    
+            for b in range(bs):
+                neg_idx = torch.multinomial(weights_t2i[b], 1).item()
+                image_embeds_neg.append(image_embeds[neg_idx])
+            image_embeds_neg = torch.stack(image_embeds_neg,dim=0)   
+
+            # select a negative text (from same rank) for each image    
+            text_ids_neg = []
+            text_atts_neg = []
+            for b in range(bs):
+                neg_idx = torch.multinomial(weights_i2t[b], 1).item()
+                text_ids_neg.append(encoder_input_ids[neg_idx])
+                text_atts_neg.append(text.attention_mask[neg_idx])            
+            
+        text_ids_neg = torch.stack(text_ids_neg,dim=0)   
+        text_atts_neg = torch.stack(text_atts_neg,dim=0)      
+
+        text_ids_all = torch.cat([encoder_input_ids, text_ids_neg],dim=0)     
+        text_atts_all = torch.cat([text.attention_mask, text_atts_neg],dim=0)     
+
+        image_embeds_all = torch.cat([image_embeds_neg,image_embeds],dim=0)
+        image_atts_all = torch.cat([image_atts,image_atts],dim=0)
+
+        output_neg = self.text_encoder(text_ids_all,
+                                       attention_mask = text_atts_all,
+                                       encoder_hidden_states = image_embeds_all,
+                                       encoder_attention_mask = image_atts_all,      
+                                       return_dict = True,
+                                      )                         
+          
+
+        vl_embeddings = torch.cat([output_pos.last_hidden_state[:,0,:], output_neg.last_hidden_state[:,0,:]],dim=0)
+        vl_output = self.itm_head(vl_embeddings)            
+
+        itm_labels = torch.cat([torch.ones(bs,dtype=torch.long),torch.zeros(2*bs,dtype=torch.long)],
+                               dim=0).to(image.device)
+        loss_itm = F.cross_entropy(vl_output, itm_labels)     
+
+        return loss_ita, loss_itm 
+ 
+
+    @torch.no_grad()    
+    def copy_params(self):
+        for model_pair in self.model_pairs:           
+            for param, param_m in zip(model_pair[0].parameters(), model_pair[1].parameters()):
+                param_m.data.copy_(param.data)  # initialize
+                param_m.requires_grad = False  # not update by gradient    
+
+            
+    @torch.no_grad()        
+    def _momentum_update(self):
+        for model_pair in self.model_pairs:           
+            for param, param_m in zip(model_pair[0].parameters(), model_pair[1].parameters()):
+                param_m.data = param_m.data * self.momentum + param.data * (1. - self.momentum)
+                
+                
+    @torch.no_grad()
+    def _dequeue_and_enqueue(self, image_feat, text_feat, idxs):
+        # gather keys before updating queue
+        image_feats = concat_all_gather(image_feat)
+        text_feats = concat_all_gather(text_feat)
+        
+
+        batch_size = image_feats.shape[0]
+
+        ptr = int(self.ptr_queue)
+        assert self.queue_size % batch_size == 0  # for simplicity
+
+        # replace the keys at ptr (dequeue and enqueue)
+        self.image_queue[:, ptr:ptr + batch_size] = image_feats.T
+        self.text_queue[:, ptr:ptr + batch_size] = text_feats.T
+        self.idx_queue[:, ptr:ptr + batch_size] = idxs.T
+        ptr = (ptr + batch_size) % self.queue_size # move pointer
+
+        self.ptr_queue[0] = ptr  
+
+
+def blip_retrieval(pretrained='',**kwargs):
+    model = BLIP_Retrieval(**kwargs)
+    if pretrained:
+        model,msg = load_checkpoint(model,pretrained)
+        print("missing keys:")
+        print(msg.missing_keys)
+    return model 
+
+
+@torch.no_grad()
+def concat_all_gather(tensor):
+    """
+    Performs all_gather operation on the provided tensors.
+    *** Warning ***: torch.distributed.all_gather has no gradient.
+    """
+    tensors_gather = [torch.ones_like(tensor)
+        for _ in range(torch.distributed.get_world_size())]
+    torch.distributed.all_gather(tensors_gather, tensor, async_op=False)
+
+    output = torch.cat(tensors_gather, dim=0)
+    return output      
+
+
+class GatherLayer(torch.autograd.Function):
+    """
+    Gather tensors from all workers with support for backward propagation:
+    This implementation does not cut the gradients as torch.distributed.all_gather does.
+    """
+
+    @staticmethod
+    def forward(ctx, x):
+        output = [torch.zeros_like(x) for _ in range(torch.distributed.get_world_size())]
+        torch.distributed.all_gather(output, x)
+        return tuple(output)
+
+    @staticmethod
+    def backward(ctx, *grads):
+        all_gradients = torch.stack(grads)
+        torch.distributed.all_reduce(all_gradients)
+        return all_gradients[torch.distributed.get_rank()]
+
+
+def all_gather_with_grad(tensors):
+    """
+    Performs all_gather operation on the provided tensors.
+    Graph remains connected for backward grad computation.
+    """
+    # Queue the gathered tensors
+    world_size = torch.distributed.get_world_size()
+    # There is no need for reduction in the single-proc case
+    if world_size == 1:
+        return tensors
+
+    tensor_all = GatherLayer.apply(tensors)
+
+    return torch.cat(tensor_all, dim=0)

extras/BLIP/models/blip_vqa.py

@@ -0,0 +1,186 @@
+from extras.BLIP.models.med import BertConfig, BertModel, BertLMHeadModel
+from extras.BLIP.models.blip import create_vit, init_tokenizer, load_checkpoint
+
+import torch
+from torch import nn
+import torch.nn.functional as F
+from transformers import BertTokenizer
+import numpy as np
+
+class BLIP_VQA(nn.Module):
+    def __init__(self,                 
+                 med_config = 'configs/med_config.json',  
+                 image_size = 480,
+                 vit = 'base',
+                 vit_grad_ckpt = False,
+                 vit_ckpt_layer = 0,                   
+                 ):
+        """
+        Args:
+            med_config (str): path for the mixture of encoder-decoder model's configuration file
+            image_size (int): input image size
+            vit (str): model size of vision transformer
+        """               
+        super().__init__()
+        
+        self.visual_encoder, vision_width = create_vit(vit, image_size, vit_grad_ckpt, vit_ckpt_layer, drop_path_rate=0.1)
+        self.tokenizer = init_tokenizer()  
+        
+        encoder_config = BertConfig.from_json_file(med_config)
+        encoder_config.encoder_width = vision_width
+        self.text_encoder = BertModel(config=encoder_config, add_pooling_layer=False) 
+        
+        decoder_config = BertConfig.from_json_file(med_config)        
+        self.text_decoder = BertLMHeadModel(config=decoder_config)          
+
+
+    def forward(self, image, question, answer=None, n=None, weights=None, train=True, inference='rank', k_test=128):
+        
+        image_embeds = self.visual_encoder(image) 
+        image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)
+        
+        question = self.tokenizer(question, padding='longest', truncation=True, max_length=35, 
+                                  return_tensors="pt").to(image.device) 
+        question.input_ids[:,0] = self.tokenizer.enc_token_id
+        
+        if train:               
+            '''
+            n: number of answers for each question
+            weights: weight for each answer
+            '''                     
+            answer = self.tokenizer(answer, padding='longest', return_tensors="pt").to(image.device) 
+            answer.input_ids[:,0] = self.tokenizer.bos_token_id
+            answer_targets = answer.input_ids.masked_fill(answer.input_ids == self.tokenizer.pad_token_id, -100)      
+
+            question_output = self.text_encoder(question.input_ids, 
+                                                attention_mask = question.attention_mask, 
+                                                encoder_hidden_states = image_embeds,
+                                                encoder_attention_mask = image_atts,                             
+                                                return_dict = True)    
+
+            question_states = []                
+            question_atts = []  
+            for b, n in enumerate(n):
+                question_states += [question_output.last_hidden_state[b]]*n
+                question_atts += [question.attention_mask[b]]*n                
+            question_states = torch.stack(question_states,0)    
+            question_atts = torch.stack(question_atts,0)     
+
+            answer_output = self.text_decoder(answer.input_ids, 
+                                              attention_mask = answer.attention_mask, 
+                                              encoder_hidden_states = question_states,
+                                              encoder_attention_mask = question_atts,                  
+                                              labels = answer_targets,
+                                              return_dict = True,   
+                                              reduction = 'none',
+                                             )      
+            
+            loss = weights * answer_output.loss
+            loss = loss.sum()/image.size(0)
+
+            return loss
+            
+
+        else: 
+            question_output = self.text_encoder(question.input_ids, 
+                                                attention_mask = question.attention_mask, 
+                                                encoder_hidden_states = image_embeds,
+                                                encoder_attention_mask = image_atts,                                    
+                                                return_dict = True) 
+            
+            if inference=='generate':
+                num_beams = 3
+                question_states = question_output.last_hidden_state.repeat_interleave(num_beams,dim=0)
+                question_atts = torch.ones(question_states.size()[:-1],dtype=torch.long).to(question_states.device)
+                model_kwargs = {"encoder_hidden_states": question_states, "encoder_attention_mask":question_atts}
+                
+                bos_ids = torch.full((image.size(0),1),fill_value=self.tokenizer.bos_token_id,device=image.device)
+                
+                outputs = self.text_decoder.generate(input_ids=bos_ids,
+                                                     max_length=10,
+                                                     min_length=1,
+                                                     num_beams=num_beams,
+                                                     eos_token_id=self.tokenizer.sep_token_id,
+                                                     pad_token_id=self.tokenizer.pad_token_id, 
+                                                     **model_kwargs)
+                
+                answers = []    
+                for output in outputs:
+                    answer = self.tokenizer.decode(output, skip_special_tokens=True)    
+                    answers.append(answer)
+                return answers
+            
+            elif inference=='rank':
+                max_ids = self.rank_answer(question_output.last_hidden_state, question.attention_mask, 
+                                           answer.input_ids, answer.attention_mask, k_test) 
+                return max_ids
+ 
+                
+                
+    def rank_answer(self, question_states, question_atts, answer_ids, answer_atts, k):
+        
+        num_ques = question_states.size(0)
+        start_ids = answer_ids[0,0].repeat(num_ques,1) # bos token
+        
+        start_output = self.text_decoder(start_ids, 
+                                         encoder_hidden_states = question_states,
+                                         encoder_attention_mask = question_atts,                                      
+                                         return_dict = True,
+                                         reduction = 'none')              
+        logits = start_output.logits[:,0,:] # first token's logit
+        
+        # topk_probs: top-k probability 
+        # topk_ids: [num_question, k]        
+        answer_first_token = answer_ids[:,1]
+        prob_first_token = F.softmax(logits,dim=1).index_select(dim=1, index=answer_first_token) 
+        topk_probs, topk_ids = prob_first_token.topk(k,dim=1) 
+        
+        # answer input: [num_question*k, answer_len]                 
+        input_ids = []
+        input_atts = []
+        for b, topk_id in enumerate(topk_ids):
+            input_ids.append(answer_ids.index_select(dim=0, index=topk_id))
+            input_atts.append(answer_atts.index_select(dim=0, index=topk_id))
+        input_ids = torch.cat(input_ids,dim=0)  
+        input_atts = torch.cat(input_atts,dim=0)  
+
+        targets_ids = input_ids.masked_fill(input_ids == self.tokenizer.pad_token_id, -100)
+
+        # repeat encoder's output for top-k answers
+        question_states = tile(question_states, 0, k)
+        question_atts = tile(question_atts, 0, k)
+        
+        output = self.text_decoder(input_ids, 
+                                   attention_mask = input_atts, 
+                                   encoder_hidden_states = question_states,
+                                   encoder_attention_mask = question_atts,     
+                                   labels = targets_ids,
+                                   return_dict = True, 
+                                   reduction = 'none')   
+        
+        log_probs_sum = -output.loss
+        log_probs_sum = log_probs_sum.view(num_ques,k)
+
+        max_topk_ids = log_probs_sum.argmax(dim=1) 
+        max_ids = topk_ids[max_topk_ids>=0,max_topk_ids]
+
+        return max_ids
+    
+    
+def blip_vqa(pretrained='',**kwargs):
+    model = BLIP_VQA(**kwargs)
+    if pretrained:
+        model,msg = load_checkpoint(model,pretrained)
+#         assert(len(msg.missing_keys)==0)
+    return model  
+
+
+def tile(x, dim, n_tile):
+    init_dim = x.size(dim)
+    repeat_idx = [1] * x.dim()
+    repeat_idx[dim] = n_tile
+    x = x.repeat(*(repeat_idx))
+    order_index = torch.LongTensor(np.concatenate([init_dim * np.arange(n_tile) + i for i in range(init_dim)]))
+    return torch.index_select(x, dim, order_index.to(x.device))    
+        
+        

extras/BLIP/models/med.py

@@ -0,0 +1,955 @@
+'''
+ * Copyright (c) 2022, salesforce.com, inc.
+ * All rights reserved.
+ * SPDX-License-Identifier: BSD-3-Clause
+ * For full license text, see LICENSE.txt file in the repo root or https://opensource.org/licenses/BSD-3-Clause
+ * By Junnan Li
+ * Based on huggingface code base
+ * https://github.com/huggingface/transformers/blob/v4.15.0/src/transformers/models/bert
+'''
+
+import math
+import os
+import warnings
+from dataclasses import dataclass
+from typing import Optional, Tuple
+
+import torch
+from torch import Tensor, device, dtype, nn
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import CrossEntropyLoss
+import torch.nn.functional as F
+
+from transformers.activations import ACT2FN
+from transformers.file_utils import (
+    ModelOutput,
+)
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPastAndCrossAttentions,
+    BaseModelOutputWithPoolingAndCrossAttentions,
+    CausalLMOutputWithCrossAttentions,
+    MaskedLMOutput,
+    MultipleChoiceModelOutput,
+    NextSentencePredictorOutput,
+    QuestionAnsweringModelOutput,
+    SequenceClassifierOutput,
+    TokenClassifierOutput,
+)
+from transformers.modeling_utils import (
+    PreTrainedModel,
+    apply_chunking_to_forward,
+    find_pruneable_heads_and_indices,
+    prune_linear_layer,
+)
+from transformers.utils import logging
+from transformers.models.bert.configuration_bert import BertConfig
+
+
+logger = logging.get_logger(__name__)
+
+
+class BertEmbeddings(nn.Module):
+    """Construct the embeddings from word and position embeddings."""
+
+    def __init__(self, config):
+        super().__init__()
+        self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id)
+        self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size)
+
+        # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
+        # any TensorFlow checkpoint file
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+        # position_ids (1, len position emb) is contiguous in memory and exported when serialized
+        self.register_buffer("position_ids", torch.arange(config.max_position_embeddings).expand((1, -1)))
+        self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
+        
+        self.config = config
+
+    def forward(
+        self, input_ids=None, position_ids=None, inputs_embeds=None, past_key_values_length=0
+    ):
+        if input_ids is not None:
+            input_shape = input_ids.size()
+        else:
+            input_shape = inputs_embeds.size()[:-1]
+
+        seq_length = input_shape[1]
+
+        if position_ids is None:
+            position_ids = self.position_ids[:, past_key_values_length : seq_length + past_key_values_length]
+
+        if inputs_embeds is None:
+            inputs_embeds = self.word_embeddings(input_ids)
+
+        embeddings = inputs_embeds
+
+        if self.position_embedding_type == "absolute":
+            position_embeddings = self.position_embeddings(position_ids)
+            embeddings += position_embeddings
+        embeddings = self.LayerNorm(embeddings)
+        embeddings = self.dropout(embeddings)
+        return embeddings
+
+
+class BertSelfAttention(nn.Module):
+    def __init__(self, config, is_cross_attention):
+        super().__init__()
+        self.config = config
+        if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"):
+            raise ValueError(
+                "The hidden size (%d) is not a multiple of the number of attention "
+                "heads (%d)" % (config.hidden_size, config.num_attention_heads)
+            )
+        
+        self.num_attention_heads = config.num_attention_heads
+        self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
+        self.all_head_size = self.num_attention_heads * self.attention_head_size
+
+        self.query = nn.Linear(config.hidden_size, self.all_head_size)
+        if is_cross_attention:
+            self.key = nn.Linear(config.encoder_width, self.all_head_size)
+            self.value = nn.Linear(config.encoder_width, self.all_head_size)
+        else:
+            self.key = nn.Linear(config.hidden_size, self.all_head_size)
+            self.value = nn.Linear(config.hidden_size, self.all_head_size)
+
+        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
+        self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
+        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
+            self.max_position_embeddings = config.max_position_embeddings
+            self.distance_embedding = nn.Embedding(2 * config.max_position_embeddings - 1, self.attention_head_size)
+        self.save_attention = False   
+            
+    def save_attn_gradients(self, attn_gradients):
+        self.attn_gradients = attn_gradients
+        
+    def get_attn_gradients(self):
+        return self.attn_gradients
+    
+    def save_attention_map(self, attention_map):
+        self.attention_map = attention_map
+        
+    def get_attention_map(self):
+        return self.attention_map
+    
+    def transpose_for_scores(self, x):
+        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
+        x = x.view(*new_x_shape)
+        return x.permute(0, 2, 1, 3)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+    ):
+        mixed_query_layer = self.query(hidden_states)
+
+        # If this is instantiated as a cross-attention module, the keys
+        # and values come from an encoder; the attention mask needs to be
+        # such that the encoder's padding tokens are not attended to.
+        is_cross_attention = encoder_hidden_states is not None
+
+        if is_cross_attention:
+            key_layer = self.transpose_for_scores(self.key(encoder_hidden_states))
+            value_layer = self.transpose_for_scores(self.value(encoder_hidden_states))
+            attention_mask = encoder_attention_mask
+        elif past_key_value is not None:
+            key_layer = self.transpose_for_scores(self.key(hidden_states))
+            value_layer = self.transpose_for_scores(self.value(hidden_states))
+            key_layer = torch.cat([past_key_value[0], key_layer], dim=2)
+            value_layer = torch.cat([past_key_value[1], value_layer], dim=2)
+        else:
+            key_layer = self.transpose_for_scores(self.key(hidden_states))
+            value_layer = self.transpose_for_scores(self.value(hidden_states))
+
+        query_layer = self.transpose_for_scores(mixed_query_layer)
+
+        past_key_value = (key_layer, value_layer)
+
+        # Take the dot product between "query" and "key" to get the raw attention scores.
+        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))
+
+        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
+            seq_length = hidden_states.size()[1]
+            position_ids_l = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(-1, 1)
+            position_ids_r = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(1, -1)
+            distance = position_ids_l - position_ids_r
+            positional_embedding = self.distance_embedding(distance + self.max_position_embeddings - 1)
+            positional_embedding = positional_embedding.to(dtype=query_layer.dtype)  # fp16 compatibility
+
+            if self.position_embedding_type == "relative_key":
+                relative_position_scores = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
+                attention_scores = attention_scores + relative_position_scores
+            elif self.position_embedding_type == "relative_key_query":
+                relative_position_scores_query = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
+                relative_position_scores_key = torch.einsum("bhrd,lrd->bhlr", key_layer, positional_embedding)
+                attention_scores = attention_scores + relative_position_scores_query + relative_position_scores_key
+
+        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
+        if attention_mask is not None:
+            # Apply the attention mask is (precomputed for all layers in BertModel forward() function)
+            attention_scores = attention_scores + attention_mask
+
+        # Normalize the attention scores to probabilities.
+        attention_probs = nn.Softmax(dim=-1)(attention_scores)
+        
+        if is_cross_attention and self.save_attention:
+            self.save_attention_map(attention_probs)
+            attention_probs.register_hook(self.save_attn_gradients)         
+
+        # This is actually dropping out entire tokens to attend to, which might
+        # seem a bit unusual, but is taken from the original Transformer paper.
+        attention_probs_dropped = self.dropout(attention_probs)
+
+        # Mask heads if we want to
+        if head_mask is not None:
+            attention_probs_dropped = attention_probs_dropped * head_mask
+
+        context_layer = torch.matmul(attention_probs_dropped, value_layer)
+
+        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
+        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
+        context_layer = context_layer.view(*new_context_layer_shape)
+
+        outputs = (context_layer, attention_probs) if output_attentions else (context_layer,)
+
+        outputs = outputs + (past_key_value,)
+        return outputs
+
+
+class BertSelfOutput(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+    def forward(self, hidden_states, input_tensor):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states + input_tensor)
+        return hidden_states
+
+
+class BertAttention(nn.Module):
+    def __init__(self, config, is_cross_attention=False):
+        super().__init__()
+        self.self = BertSelfAttention(config, is_cross_attention)
+        self.output = BertSelfOutput(config)
+        self.pruned_heads = set()
+
+    def prune_heads(self, heads):
+        if len(heads) == 0:
+            return
+        heads, index = find_pruneable_heads_and_indices(
+            heads, self.self.num_attention_heads, self.self.attention_head_size, self.pruned_heads
+        )
+
+        # Prune linear layers
+        self.self.query = prune_linear_layer(self.self.query, index)
+        self.self.key = prune_linear_layer(self.self.key, index)
+        self.self.value = prune_linear_layer(self.self.value, index)
+        self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)
+
+        # Update hyper params and store pruned heads
+        self.self.num_attention_heads = self.self.num_attention_heads - len(heads)
+        self.self.all_head_size = self.self.attention_head_size * self.self.num_attention_heads
+        self.pruned_heads = self.pruned_heads.union(heads)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+    ):
+        self_outputs = self.self(
+            hidden_states,
+            attention_mask,
+            head_mask,
+            encoder_hidden_states,
+            encoder_attention_mask,
+            past_key_value,
+            output_attentions,
+        )
+        attention_output = self.output(self_outputs[0], hidden_states)
+        outputs = (attention_output,) + self_outputs[1:]  # add attentions if we output them
+        return outputs
+
+
+class BertIntermediate(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
+        if isinstance(config.hidden_act, str):
+            self.intermediate_act_fn = ACT2FN[config.hidden_act]
+        else:
+            self.intermediate_act_fn = config.hidden_act
+
+    def forward(self, hidden_states):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.intermediate_act_fn(hidden_states)
+        return hidden_states
+
+
+class BertOutput(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+    def forward(self, hidden_states, input_tensor):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states + input_tensor)
+        return hidden_states
+
+
+class BertLayer(nn.Module):
+    def __init__(self, config, layer_num):
+        super().__init__()
+        self.config = config
+        self.chunk_size_feed_forward = config.chunk_size_feed_forward
+        self.seq_len_dim = 1
+        self.attention = BertAttention(config)      
+        self.layer_num = layer_num          
+        if self.config.add_cross_attention:
+            self.crossattention = BertAttention(config, is_cross_attention=self.config.add_cross_attention)
+        self.intermediate = BertIntermediate(config)
+        self.output = BertOutput(config)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+        mode=None,
+    ):
+        # decoder uni-directional self-attention cached key/values tuple is at positions 1,2
+        self_attn_past_key_value = past_key_value[:2] if past_key_value is not None else None
+        self_attention_outputs = self.attention(
+            hidden_states,
+            attention_mask,
+            head_mask,
+            output_attentions=output_attentions,
+            past_key_value=self_attn_past_key_value,
+        )
+        attention_output = self_attention_outputs[0]
+
+        outputs = self_attention_outputs[1:-1]
+        present_key_value = self_attention_outputs[-1]
+
+        if mode=='multimodal':
+            assert encoder_hidden_states is not None, "encoder_hidden_states must be given for cross-attention layers"
+
+            cross_attention_outputs = self.crossattention(
+                attention_output,
+                attention_mask,
+                head_mask,
+                encoder_hidden_states,
+                encoder_attention_mask,
+                output_attentions=output_attentions,
+            )
+            attention_output = cross_attention_outputs[0]
+            outputs = outputs + cross_attention_outputs[1:-1]  # add cross attentions if we output attention weights                               
+        layer_output = apply_chunking_to_forward(
+            self.feed_forward_chunk, self.chunk_size_feed_forward, self.seq_len_dim, attention_output
+        )
+        outputs = (layer_output,) + outputs
+
+        outputs = outputs + (present_key_value,)
+
+        return outputs
+
+    def feed_forward_chunk(self, attention_output):
+        intermediate_output = self.intermediate(attention_output)
+        layer_output = self.output(intermediate_output, attention_output)
+        return layer_output
+
+
+class BertEncoder(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.layer = nn.ModuleList([BertLayer(config,i) for i in range(config.num_hidden_layers)])
+        self.gradient_checkpointing = False
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_values=None,
+        use_cache=None,
+        output_attentions=False,
+        output_hidden_states=False,
+        return_dict=True,
+        mode='multimodal',
+    ):
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attentions = () if output_attentions else None
+        all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None
+
+        next_decoder_cache = () if use_cache else None
+               
+        for i in range(self.config.num_hidden_layers):
+            layer_module = self.layer[i]
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            layer_head_mask = head_mask[i] if head_mask is not None else None
+            past_key_value = past_key_values[i] if past_key_values is not None else None
+
+            if self.gradient_checkpointing and self.training:
+
+                if use_cache:
+                    logger.warn(
+                        "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                    )
+                    use_cache = False
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        return module(*inputs, past_key_value, output_attentions)
+
+                    return custom_forward
+
+                layer_outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(layer_module),
+                    hidden_states,
+                    attention_mask,
+                    layer_head_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                    mode=mode,
+                )
+            else:
+                layer_outputs = layer_module(
+                    hidden_states,
+                    attention_mask,
+                    layer_head_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                    past_key_value,
+                    output_attentions,
+                    mode=mode,
+                )
+
+            hidden_states = layer_outputs[0]
+            if use_cache:
+                next_decoder_cache += (layer_outputs[-1],)
+            if output_attentions:
+                all_self_attentions = all_self_attentions + (layer_outputs[1],)
+
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(
+                v
+                for v in [
+                    hidden_states,
+                    next_decoder_cache,
+                    all_hidden_states,
+                    all_self_attentions,
+                    all_cross_attentions,
+                ]
+                if v is not None
+            )
+        return BaseModelOutputWithPastAndCrossAttentions(
+            last_hidden_state=hidden_states,
+            past_key_values=next_decoder_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+            cross_attentions=all_cross_attentions,
+        )
+
+
+class BertPooler(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        self.activation = nn.Tanh()
+
+    def forward(self, hidden_states):
+        # We "pool" the model by simply taking the hidden state corresponding
+        # to the first token.
+        first_token_tensor = hidden_states[:, 0]
+        pooled_output = self.dense(first_token_tensor)
+        pooled_output = self.activation(pooled_output)
+        return pooled_output
+
+
+class BertPredictionHeadTransform(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        if isinstance(config.hidden_act, str):
+            self.transform_act_fn = ACT2FN[config.hidden_act]
+        else:
+            self.transform_act_fn = config.hidden_act
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+
+    def forward(self, hidden_states):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.transform_act_fn(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states)
+        return hidden_states
+
+
+class BertLMPredictionHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.transform = BertPredictionHeadTransform(config)
+
+        # The output weights are the same as the input embeddings, but there is
+        # an output-only bias for each token.
+        self.decoder = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        self.bias = nn.Parameter(torch.zeros(config.vocab_size))
+
+        # Need a link between the two variables so that the bias is correctly resized with `resize_token_embeddings`
+        self.decoder.bias = self.bias
+
+    def forward(self, hidden_states):
+        hidden_states = self.transform(hidden_states)
+        hidden_states = self.decoder(hidden_states)
+        return hidden_states
+
+
+class BertOnlyMLMHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.predictions = BertLMPredictionHead(config)
+
+    def forward(self, sequence_output):
+        prediction_scores = self.predictions(sequence_output)
+        return prediction_scores
+
+
+class BertPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = BertConfig
+    base_model_prefix = "bert"
+    _keys_to_ignore_on_load_missing = [r"position_ids"]
+
+    def _init_weights(self, module):
+        """ Initialize the weights """
+        if isinstance(module, (nn.Linear, nn.Embedding)):
+            # Slightly different from the TF version which uses truncated_normal for initialization
+            # cf https://github.com/pytorch/pytorch/pull/5617
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+        elif isinstance(module, nn.LayerNorm):
+            module.bias.data.zero_()
+            module.weight.data.fill_(1.0)
+        if isinstance(module, nn.Linear) and module.bias is not None:
+            module.bias.data.zero_()
+
+
+class BertModel(BertPreTrainedModel):
+    """
+    The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of
+    cross-attention is added between the self-attention layers, following the architecture described in `Attention is
+    all you need <https://arxiv.org/abs/1706.03762>`__ by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit,
+    Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.
+    argument and :obj:`add_cross_attention` set to :obj:`True`; an :obj:`encoder_hidden_states` is then expected as an
+    input to the forward pass.
+    """
+
+    def __init__(self, config, add_pooling_layer=True):
+        super().__init__(config)
+        self.config = config
+
+        self.embeddings = BertEmbeddings(config)
+        
+        self.encoder = BertEncoder(config)
+
+        self.pooler = BertPooler(config) if add_pooling_layer else None
+
+        self.init_weights()
+ 
+
+    def get_input_embeddings(self):
+        return self.embeddings.word_embeddings
+
+    def set_input_embeddings(self, value):
+        self.embeddings.word_embeddings = value
+
+    def _prune_heads(self, heads_to_prune):
+        """
+        Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base
+        class PreTrainedModel
+        """
+        for layer, heads in heads_to_prune.items():
+            self.encoder.layer[layer].attention.prune_heads(heads)
+
+    
+    def get_extended_attention_mask(self, attention_mask: Tensor, input_shape: Tuple[int], device: device, is_decoder: bool) -> Tensor:
+        """
+        Makes broadcastable attention and causal masks so that future and masked tokens are ignored.
+
+        Arguments:
+            attention_mask (:obj:`torch.Tensor`):
+                Mask with ones indicating tokens to attend to, zeros for tokens to ignore.
+            input_shape (:obj:`Tuple[int]`):
+                The shape of the input to the model.
+            device: (:obj:`torch.device`):
+                The device of the input to the model.
+
+        Returns:
+            :obj:`torch.Tensor` The extended attention mask, with a the same dtype as :obj:`attention_mask.dtype`.
+        """
+        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
+        # ourselves in which case we just need to make it broadcastable to all heads.
+        if attention_mask.dim() == 3:
+            extended_attention_mask = attention_mask[:, None, :, :]
+        elif attention_mask.dim() == 2:
+            # Provided a padding mask of dimensions [batch_size, seq_length]
+            # - if the model is a decoder, apply a causal mask in addition to the padding mask
+            # - if the model is an encoder, make the mask broadcastable to [batch_size, num_heads, seq_length, seq_length]
+            if is_decoder:
+                batch_size, seq_length = input_shape
+
+                seq_ids = torch.arange(seq_length, device=device)
+                causal_mask = seq_ids[None, None, :].repeat(batch_size, seq_length, 1) <= seq_ids[None, :, None]
+                # in case past_key_values are used we need to add a prefix ones mask to the causal mask
+                # causal and attention masks must have same type with pytorch version < 1.3
+                causal_mask = causal_mask.to(attention_mask.dtype)
+   
+                if causal_mask.shape[1] < attention_mask.shape[1]:
+                    prefix_seq_len = attention_mask.shape[1] - causal_mask.shape[1]
+                    causal_mask = torch.cat(
+                        [
+                            torch.ones((batch_size, seq_length, prefix_seq_len), device=device, dtype=causal_mask.dtype),
+                            causal_mask,
+                        ],
+                        axis=-1,
+                    )                     
+
+                extended_attention_mask = causal_mask[:, None, :, :] * attention_mask[:, None, None, :]
+            else:
+                extended_attention_mask = attention_mask[:, None, None, :]
+        else:
+            raise ValueError(
+                "Wrong shape for input_ids (shape {}) or attention_mask (shape {})".format(
+                    input_shape, attention_mask.shape
+                )
+            )
+
+        # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
+        # masked positions, this operation will create a tensor which is 0.0 for
+        # positions we want to attend and -10000.0 for masked positions.
+        # Since we are adding it to the raw scores before the softmax, this is
+        # effectively the same as removing these entirely.
+        extended_attention_mask = extended_attention_mask.to(dtype=self.dtype)  # fp16 compatibility
+        extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
+        return extended_attention_mask
+    
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        encoder_embeds=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_values=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+        is_decoder=False,
+        mode='multimodal',
+    ):
+        r"""
+        encoder_hidden_states  (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
+            Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
+            the model is configured as a decoder.
+        encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
+            Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
+            the cross-attention if the model is configured as a decoder. Mask values selected in ``[0, 1]``:
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+        past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
+            Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
+            If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`
+            (those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`
+            instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.
+        use_cache (:obj:`bool`, `optional`):
+            If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
+            decoding (see :obj:`past_key_values`).
+        """
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if is_decoder:
+            use_cache = use_cache if use_cache is not None else self.config.use_cache
+        else:
+            use_cache = False
+
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
+        elif input_ids is not None:
+            input_shape = input_ids.size()
+            batch_size, seq_length = input_shape
+            device = input_ids.device
+        elif inputs_embeds is not None:
+            input_shape = inputs_embeds.size()[:-1]
+            batch_size, seq_length = input_shape
+            device = inputs_embeds.device
+        elif encoder_embeds is not None:    
+            input_shape = encoder_embeds.size()[:-1]
+            batch_size, seq_length = input_shape 
+            device = encoder_embeds.device
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds or encoder_embeds")
+
+        # past_key_values_length
+        past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0
+
+        if attention_mask is None:
+            attention_mask = torch.ones(((batch_size, seq_length + past_key_values_length)), device=device)
+            
+        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
+        # ourselves in which case we just need to make it broadcastable to all heads.
+        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape, 
+                                                                                 device, is_decoder)
+
+        # If a 2D or 3D attention mask is provided for the cross-attention
+        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
+        if encoder_hidden_states is not None:
+            if type(encoder_hidden_states) == list:
+                encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states[0].size()
+            else:
+                encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
+            encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
+            
+            if type(encoder_attention_mask) == list:
+                encoder_extended_attention_mask = [self.invert_attention_mask(mask) for mask in encoder_attention_mask]
+            elif encoder_attention_mask is None:
+                encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
+                encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+            else:    
+                encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+        else:
+            encoder_extended_attention_mask = None
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x n_heads x N x N
+        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
+        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
+        head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)
+        
+        if encoder_embeds is None:
+            embedding_output = self.embeddings(
+                input_ids=input_ids,
+                position_ids=position_ids,
+                inputs_embeds=inputs_embeds,
+                past_key_values_length=past_key_values_length,
+            )
+        else:
+            embedding_output = encoder_embeds
+            
+        encoder_outputs = self.encoder(
+            embedding_output,
+            attention_mask=extended_attention_mask,
+            head_mask=head_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_extended_attention_mask,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            mode=mode,
+        )
+        sequence_output = encoder_outputs[0]
+        pooled_output = self.pooler(sequence_output) if self.pooler is not None else None
+
+        if not return_dict:
+            return (sequence_output, pooled_output) + encoder_outputs[1:]
+
+        return BaseModelOutputWithPoolingAndCrossAttentions(
+            last_hidden_state=sequence_output,
+            pooler_output=pooled_output,
+            past_key_values=encoder_outputs.past_key_values,
+            hidden_states=encoder_outputs.hidden_states,
+            attentions=encoder_outputs.attentions,
+            cross_attentions=encoder_outputs.cross_attentions,
+        )
+
+
+
+class BertLMHeadModel(BertPreTrainedModel):
+
+    _keys_to_ignore_on_load_unexpected = [r"pooler"]
+    _keys_to_ignore_on_load_missing = [r"position_ids", r"predictions.decoder.bias"]
+
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.bert = BertModel(config, add_pooling_layer=False)
+        self.cls = BertOnlyMLMHead(config)
+
+        self.init_weights()
+
+    def get_output_embeddings(self):
+        return self.cls.predictions.decoder
+
+    def set_output_embeddings(self, new_embeddings):
+        self.cls.predictions.decoder = new_embeddings
+
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        labels=None,
+        past_key_values=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+        return_logits=False,            
+        is_decoder=True,
+        reduction='mean',
+        mode='multimodal', 
+    ):
+        r"""
+        encoder_hidden_states  (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
+            Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
+            the model is configured as a decoder.
+        encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
+            Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
+            the cross-attention if the model is configured as a decoder. Mask values selected in ``[0, 1]``:
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+        labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
+            Labels for computing the left-to-right language modeling loss (next word prediction). Indices should be in
+            ``[-100, 0, ..., config.vocab_size]`` (see ``input_ids`` docstring) Tokens with indices set to ``-100`` are
+            ignored (masked), the loss is only computed for the tokens with labels n ``[0, ..., config.vocab_size]``
+        past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
+            Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
+            If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`
+            (those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`
+            instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.
+        use_cache (:obj:`bool`, `optional`):
+            If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
+            decoding (see :obj:`past_key_values`).
+        Returns:
+        Example::
+            >>> from transformers import BertTokenizer, BertLMHeadModel, BertConfig
+            >>> import torch
+            >>> tokenizer = BertTokenizer.from_pretrained('bert-base-cased')
+            >>> config = BertConfig.from_pretrained("bert-base-cased")
+            >>> model = BertLMHeadModel.from_pretrained('bert-base-cased', config=config)
+            >>> inputs = tokenizer("Hello, my dog is cute", return_tensors="pt")
+            >>> outputs = model(**inputs)
+            >>> prediction_logits = outputs.logits
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        if labels is not None:
+            use_cache = False
+
+        outputs = self.bert(
+            input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            is_decoder=is_decoder,
+            mode=mode,
+        )
+        
+        sequence_output = outputs[0]
+        prediction_scores = self.cls(sequence_output)
+        
+        if return_logits:
+            return prediction_scores[:, :-1, :].contiguous()  
+
+        lm_loss = None
+        if labels is not None:
+            # we are doing next-token prediction; shift prediction scores and input ids by one
+            shifted_prediction_scores = prediction_scores[:, :-1, :].contiguous()
+            labels = labels[:, 1:].contiguous()
+            loss_fct = CrossEntropyLoss(reduction=reduction, label_smoothing=0.1) 
+            lm_loss = loss_fct(shifted_prediction_scores.view(-1, self.config.vocab_size), labels.view(-1))
+            if reduction=='none':
+                lm_loss = lm_loss.view(prediction_scores.size(0),-1).sum(1)               
+
+        if not return_dict:
+            output = (prediction_scores,) + outputs[2:]
+            return ((lm_loss,) + output) if lm_loss is not None else output
+
+        return CausalLMOutputWithCrossAttentions(
+            loss=lm_loss,
+            logits=prediction_scores,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+            cross_attentions=outputs.cross_attentions,
+        )
+
+    def prepare_inputs_for_generation(self, input_ids, past=None, attention_mask=None, **model_kwargs):
+        input_shape = input_ids.shape
+        # if model is used as a decoder in encoder-decoder model, the decoder attention mask is created on the fly
+        if attention_mask is None:
+            attention_mask = input_ids.new_ones(input_shape)
+
+        # cut decoder_input_ids if past is used
+        if past is not None:
+            input_ids = input_ids[:, -1:]
+
+        return {
+            "input_ids": input_ids, 
+            "attention_mask": attention_mask, 
+            "past_key_values": past,
+            "encoder_hidden_states": model_kwargs.get("encoder_hidden_states", None),
+            "encoder_attention_mask": model_kwargs.get("encoder_attention_mask", None),
+            "is_decoder": True,
+        }
+
+    def _reorder_cache(self, past, beam_idx):
+        reordered_past = ()
+        for layer_past in past:
+            reordered_past += (tuple(past_state.index_select(0, beam_idx) for past_state in layer_past),)
+        return reordered_past

extras/BLIP/models/nlvr_encoder.py

@@ -0,0 +1,843 @@
+import math
+import os
+import warnings
+from dataclasses import dataclass
+from typing import Optional, Tuple
+
+import torch
+from torch import Tensor, device, dtype, nn
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import CrossEntropyLoss
+import torch.nn.functional as F
+
+from transformers.activations import ACT2FN
+from transformers.file_utils import (
+    ModelOutput,
+)
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPastAndCrossAttentions,
+    BaseModelOutputWithPoolingAndCrossAttentions,
+    CausalLMOutputWithCrossAttentions,
+    MaskedLMOutput,
+    MultipleChoiceModelOutput,
+    NextSentencePredictorOutput,
+    QuestionAnsweringModelOutput,
+    SequenceClassifierOutput,
+    TokenClassifierOutput,
+)
+from transformers.modeling_utils import (
+    PreTrainedModel,
+    apply_chunking_to_forward,
+    find_pruneable_heads_and_indices,
+    prune_linear_layer,
+)
+from transformers.utils import logging
+from transformers.models.bert.configuration_bert import BertConfig
+
+
+logger = logging.get_logger(__name__)
+
+
+class BertEmbeddings(nn.Module):
+    """Construct the embeddings from word and position embeddings."""
+
+    def __init__(self, config):
+        super().__init__()
+        self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id)
+        self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size)
+
+        # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
+        # any TensorFlow checkpoint file
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+        # position_ids (1, len position emb) is contiguous in memory and exported when serialized
+        self.register_buffer("position_ids", torch.arange(config.max_position_embeddings).expand((1, -1)))
+        self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
+        
+        self.config = config
+
+    def forward(
+        self, input_ids=None, position_ids=None, inputs_embeds=None, past_key_values_length=0
+    ):
+        if input_ids is not None:
+            input_shape = input_ids.size()
+        else:
+            input_shape = inputs_embeds.size()[:-1]
+
+        seq_length = input_shape[1]
+
+        if position_ids is None:
+            position_ids = self.position_ids[:, past_key_values_length : seq_length + past_key_values_length]
+
+        if inputs_embeds is None:
+            inputs_embeds = self.word_embeddings(input_ids)
+
+        embeddings = inputs_embeds
+
+        if self.position_embedding_type == "absolute":
+            position_embeddings = self.position_embeddings(position_ids)
+            embeddings += position_embeddings
+        embeddings = self.LayerNorm(embeddings)
+        embeddings = self.dropout(embeddings)
+        return embeddings
+
+
+class BertSelfAttention(nn.Module):
+    def __init__(self, config, is_cross_attention):
+        super().__init__()
+        self.config = config
+        if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"):
+            raise ValueError(
+                "The hidden size (%d) is not a multiple of the number of attention "
+                "heads (%d)" % (config.hidden_size, config.num_attention_heads)
+            )
+        
+        self.num_attention_heads = config.num_attention_heads
+        self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
+        self.all_head_size = self.num_attention_heads * self.attention_head_size
+
+        self.query = nn.Linear(config.hidden_size, self.all_head_size)
+        if is_cross_attention:
+            self.key = nn.Linear(config.encoder_width, self.all_head_size)
+            self.value = nn.Linear(config.encoder_width, self.all_head_size)
+        else:
+            self.key = nn.Linear(config.hidden_size, self.all_head_size)
+            self.value = nn.Linear(config.hidden_size, self.all_head_size)
+
+        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
+        self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
+        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
+            self.max_position_embeddings = config.max_position_embeddings
+            self.distance_embedding = nn.Embedding(2 * config.max_position_embeddings - 1, self.attention_head_size)
+        self.save_attention = False   
+            
+    def save_attn_gradients(self, attn_gradients):
+        self.attn_gradients = attn_gradients
+        
+    def get_attn_gradients(self):
+        return self.attn_gradients
+    
+    def save_attention_map(self, attention_map):
+        self.attention_map = attention_map
+        
+    def get_attention_map(self):
+        return self.attention_map
+    
+    def transpose_for_scores(self, x):
+        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
+        x = x.view(*new_x_shape)
+        return x.permute(0, 2, 1, 3)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+    ):
+        mixed_query_layer = self.query(hidden_states)
+
+        # If this is instantiated as a cross-attention module, the keys
+        # and values come from an encoder; the attention mask needs to be
+        # such that the encoder's padding tokens are not attended to.
+        is_cross_attention = encoder_hidden_states is not None
+
+        if is_cross_attention:
+            key_layer = self.transpose_for_scores(self.key(encoder_hidden_states))
+            value_layer = self.transpose_for_scores(self.value(encoder_hidden_states))
+            attention_mask = encoder_attention_mask
+        elif past_key_value is not None:
+            key_layer = self.transpose_for_scores(self.key(hidden_states))
+            value_layer = self.transpose_for_scores(self.value(hidden_states))
+            key_layer = torch.cat([past_key_value[0], key_layer], dim=2)
+            value_layer = torch.cat([past_key_value[1], value_layer], dim=2)
+        else:
+            key_layer = self.transpose_for_scores(self.key(hidden_states))
+            value_layer = self.transpose_for_scores(self.value(hidden_states))
+
+        query_layer = self.transpose_for_scores(mixed_query_layer)
+
+        past_key_value = (key_layer, value_layer)
+
+        # Take the dot product between "query" and "key" to get the raw attention scores.
+        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))
+
+        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
+            seq_length = hidden_states.size()[1]
+            position_ids_l = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(-1, 1)
+            position_ids_r = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(1, -1)
+            distance = position_ids_l - position_ids_r
+            positional_embedding = self.distance_embedding(distance + self.max_position_embeddings - 1)
+            positional_embedding = positional_embedding.to(dtype=query_layer.dtype)  # fp16 compatibility
+
+            if self.position_embedding_type == "relative_key":
+                relative_position_scores = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
+                attention_scores = attention_scores + relative_position_scores
+            elif self.position_embedding_type == "relative_key_query":
+                relative_position_scores_query = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
+                relative_position_scores_key = torch.einsum("bhrd,lrd->bhlr", key_layer, positional_embedding)
+                attention_scores = attention_scores + relative_position_scores_query + relative_position_scores_key
+
+        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
+        if attention_mask is not None:
+            # Apply the attention mask is (precomputed for all layers in BertModel forward() function)
+            attention_scores = attention_scores + attention_mask
+
+        # Normalize the attention scores to probabilities.
+        attention_probs = nn.Softmax(dim=-1)(attention_scores)
+        
+        if is_cross_attention and self.save_attention:
+            self.save_attention_map(attention_probs)
+            attention_probs.register_hook(self.save_attn_gradients)         
+
+        # This is actually dropping out entire tokens to attend to, which might
+        # seem a bit unusual, but is taken from the original Transformer paper.
+        attention_probs_dropped = self.dropout(attention_probs)
+
+        # Mask heads if we want to
+        if head_mask is not None:
+            attention_probs_dropped = attention_probs_dropped * head_mask
+
+        context_layer = torch.matmul(attention_probs_dropped, value_layer)
+
+        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
+        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
+        context_layer = context_layer.view(*new_context_layer_shape)
+
+        outputs = (context_layer, attention_probs) if output_attentions else (context_layer,)
+
+        outputs = outputs + (past_key_value,)
+        return outputs
+
+
+class BertSelfOutput(nn.Module):
+    def __init__(self, config, twin=False, merge=False):     
+        super().__init__()
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)        
+        if twin:
+            self.dense0 = nn.Linear(config.hidden_size, config.hidden_size)
+            self.dense1 = nn.Linear(config.hidden_size, config.hidden_size)         
+        else:
+            self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        if merge:
+            self.act =  ACT2FN[config.hidden_act]
+            self.merge_layer = nn.Linear(config.hidden_size * 2, config.hidden_size)
+            self.merge = True
+        else:
+            self.merge = False
+
+    def forward(self, hidden_states, input_tensor):
+        if type(hidden_states) == list:
+            hidden_states0 = self.dense0(hidden_states[0])
+            hidden_states1 = self.dense1(hidden_states[1])        
+            if self.merge:  
+                #hidden_states = self.merge_layer(self.act(torch.cat([hidden_states0,hidden_states1],dim=-1)))
+                hidden_states = self.merge_layer(torch.cat([hidden_states0,hidden_states1],dim=-1))
+            else:
+                hidden_states = (hidden_states0+hidden_states1)/2
+        else:    
+            hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states + input_tensor)
+        return hidden_states
+
+
+class BertAttention(nn.Module):
+    def __init__(self, config, is_cross_attention=False, layer_num=-1):
+        super().__init__()
+        if is_cross_attention:
+            self.self0 = BertSelfAttention(config, is_cross_attention)
+            self.self1 = BertSelfAttention(config, is_cross_attention)
+        else:    
+            self.self = BertSelfAttention(config, is_cross_attention)
+        self.output = BertSelfOutput(config, twin=is_cross_attention, merge=(is_cross_attention and layer_num>=6))
+        self.pruned_heads = set()
+
+    def prune_heads(self, heads):
+        if len(heads) == 0:
+            return
+        heads, index = find_pruneable_heads_and_indices(
+            heads, self.self.num_attention_heads, self.self.attention_head_size, self.pruned_heads
+        )
+
+        # Prune linear layers
+        self.self.query = prune_linear_layer(self.self.query, index)
+        self.self.key = prune_linear_layer(self.self.key, index)
+        self.self.value = prune_linear_layer(self.self.value, index)
+        self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)
+
+        # Update hyper params and store pruned heads
+        self.self.num_attention_heads = self.self.num_attention_heads - len(heads)
+        self.self.all_head_size = self.self.attention_head_size * self.self.num_attention_heads
+        self.pruned_heads = self.pruned_heads.union(heads)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+    ):        
+        if type(encoder_hidden_states)==list:   
+            self_outputs0 = self.self0(
+                hidden_states,
+                attention_mask,
+                head_mask,
+                encoder_hidden_states[0],
+                encoder_attention_mask[0],
+                past_key_value,
+                output_attentions,
+            )
+            self_outputs1 = self.self1(
+                hidden_states,
+                attention_mask,
+                head_mask,
+                encoder_hidden_states[1],
+                encoder_attention_mask[1],
+                past_key_value,
+                output_attentions,
+            )                        
+            attention_output = self.output([self_outputs0[0],self_outputs1[0]], hidden_states)
+    
+            outputs = (attention_output,) + self_outputs0[1:]  # add attentions if we output them
+        else:        
+            self_outputs = self.self(
+                hidden_states,
+                attention_mask,
+                head_mask,
+                encoder_hidden_states,
+                encoder_attention_mask,
+                past_key_value,
+                output_attentions,
+            )
+            attention_output = self.output(self_outputs[0], hidden_states)
+            outputs = (attention_output,) + self_outputs[1:]  # add attentions if we output them
+        return outputs
+
+
+class BertIntermediate(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
+        if isinstance(config.hidden_act, str):
+            self.intermediate_act_fn = ACT2FN[config.hidden_act]
+        else:
+            self.intermediate_act_fn = config.hidden_act
+
+    def forward(self, hidden_states):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.intermediate_act_fn(hidden_states)
+        return hidden_states
+
+
+class BertOutput(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+    def forward(self, hidden_states, input_tensor):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states + input_tensor)
+        return hidden_states
+
+
+class BertLayer(nn.Module):
+    def __init__(self, config, layer_num):
+        super().__init__()
+        self.config = config
+        self.chunk_size_feed_forward = config.chunk_size_feed_forward
+        self.seq_len_dim = 1
+        self.attention = BertAttention(config)      
+        self.layer_num = layer_num          
+        if self.config.add_cross_attention:
+            self.crossattention = BertAttention(config, is_cross_attention=self.config.add_cross_attention, layer_num=layer_num)
+        self.intermediate = BertIntermediate(config)
+        self.output = BertOutput(config)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+        mode=None,
+    ):
+        # decoder uni-directional self-attention cached key/values tuple is at positions 1,2
+        self_attn_past_key_value = past_key_value[:2] if past_key_value is not None else None
+        self_attention_outputs = self.attention(
+            hidden_states,
+            attention_mask,
+            head_mask,
+            output_attentions=output_attentions,
+            past_key_value=self_attn_past_key_value,
+        )
+        attention_output = self_attention_outputs[0]
+
+        outputs = self_attention_outputs[1:-1]
+        present_key_value = self_attention_outputs[-1]
+
+        if mode=='multimodal':
+            assert encoder_hidden_states is not None, "encoder_hidden_states must be given for cross-attention layers"
+            cross_attention_outputs = self.crossattention(
+                attention_output,
+                attention_mask,
+                head_mask,
+                encoder_hidden_states,
+                encoder_attention_mask,
+                output_attentions=output_attentions,
+            )
+            attention_output = cross_attention_outputs[0]
+            outputs = outputs + cross_attention_outputs[1:-1]  # add cross attentions if we output attention weights                               
+        layer_output = apply_chunking_to_forward(
+            self.feed_forward_chunk, self.chunk_size_feed_forward, self.seq_len_dim, attention_output
+        )
+        outputs = (layer_output,) + outputs
+
+        outputs = outputs + (present_key_value,)
+
+        return outputs
+
+    def feed_forward_chunk(self, attention_output):
+        intermediate_output = self.intermediate(attention_output)
+        layer_output = self.output(intermediate_output, attention_output)
+        return layer_output
+
+
+class BertEncoder(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.layer = nn.ModuleList([BertLayer(config,i) for i in range(config.num_hidden_layers)])
+        self.gradient_checkpointing = False
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_values=None,
+        use_cache=None,
+        output_attentions=False,
+        output_hidden_states=False,
+        return_dict=True,
+        mode='multimodal',
+    ):
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attentions = () if output_attentions else None
+        all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None
+
+        next_decoder_cache = () if use_cache else None
+               
+        for i in range(self.config.num_hidden_layers):
+            layer_module = self.layer[i]
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            layer_head_mask = head_mask[i] if head_mask is not None else None
+            past_key_value = past_key_values[i] if past_key_values is not None else None
+
+            if self.gradient_checkpointing and self.training:
+
+                if use_cache:
+                    logger.warn(
+                        "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                    )
+                    use_cache = False
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        return module(*inputs, past_key_value, output_attentions)
+
+                    return custom_forward
+
+                layer_outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(layer_module),
+                    hidden_states,
+                    attention_mask,
+                    layer_head_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                    mode=mode,
+                )
+            else:
+                layer_outputs = layer_module(
+                    hidden_states,
+                    attention_mask,
+                    layer_head_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                    past_key_value,
+                    output_attentions,
+                    mode=mode,
+                )
+
+            hidden_states = layer_outputs[0]
+            if use_cache:
+                next_decoder_cache += (layer_outputs[-1],)
+            if output_attentions:
+                all_self_attentions = all_self_attentions + (layer_outputs[1],)
+
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(
+                v
+                for v in [
+                    hidden_states,
+                    next_decoder_cache,
+                    all_hidden_states,
+                    all_self_attentions,
+                    all_cross_attentions,
+                ]
+                if v is not None
+            )
+        return BaseModelOutputWithPastAndCrossAttentions(
+            last_hidden_state=hidden_states,
+            past_key_values=next_decoder_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+            cross_attentions=all_cross_attentions,
+        )
+
+
+class BertPooler(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        self.activation = nn.Tanh()
+
+    def forward(self, hidden_states):
+        # We "pool" the model by simply taking the hidden state corresponding
+        # to the first token.
+        first_token_tensor = hidden_states[:, 0]
+        pooled_output = self.dense(first_token_tensor)
+        pooled_output = self.activation(pooled_output)
+        return pooled_output
+
+
+class BertPredictionHeadTransform(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        if isinstance(config.hidden_act, str):
+            self.transform_act_fn = ACT2FN[config.hidden_act]
+        else:
+            self.transform_act_fn = config.hidden_act
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+
+    def forward(self, hidden_states):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.transform_act_fn(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states)
+        return hidden_states
+
+
+class BertLMPredictionHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.transform = BertPredictionHeadTransform(config)
+
+        # The output weights are the same as the input embeddings, but there is
+        # an output-only bias for each token.
+        self.decoder = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        self.bias = nn.Parameter(torch.zeros(config.vocab_size))
+
+        # Need a link between the two variables so that the bias is correctly resized with `resize_token_embeddings`
+        self.decoder.bias = self.bias
+
+    def forward(self, hidden_states):
+        hidden_states = self.transform(hidden_states)
+        hidden_states = self.decoder(hidden_states)
+        return hidden_states
+
+
+class BertOnlyMLMHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.predictions = BertLMPredictionHead(config)
+
+    def forward(self, sequence_output):
+        prediction_scores = self.predictions(sequence_output)
+        return prediction_scores
+
+
+class BertPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = BertConfig
+    base_model_prefix = "bert"
+    _keys_to_ignore_on_load_missing = [r"position_ids"]
+
+    def _init_weights(self, module):
+        """ Initialize the weights """
+        if isinstance(module, (nn.Linear, nn.Embedding)):
+            # Slightly different from the TF version which uses truncated_normal for initialization
+            # cf https://github.com/pytorch/pytorch/pull/5617
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+        elif isinstance(module, nn.LayerNorm):
+            module.bias.data.zero_()
+            module.weight.data.fill_(1.0)
+        if isinstance(module, nn.Linear) and module.bias is not None:
+            module.bias.data.zero_()
+
+
+class BertModel(BertPreTrainedModel):
+    """
+    The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of
+    cross-attention is added between the self-attention layers, following the architecture described in `Attention is
+    all you need <https://arxiv.org/abs/1706.03762>`__ by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit,
+    Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.
+    argument and :obj:`add_cross_attention` set to :obj:`True`; an :obj:`encoder_hidden_states` is then expected as an
+    input to the forward pass.
+    """
+
+    def __init__(self, config, add_pooling_layer=True):
+        super().__init__(config)
+        self.config = config
+
+        self.embeddings = BertEmbeddings(config)
+        
+        self.encoder = BertEncoder(config)
+
+        self.pooler = BertPooler(config) if add_pooling_layer else None
+
+        self.init_weights()
+ 
+
+    def get_input_embeddings(self):
+        return self.embeddings.word_embeddings
+
+    def set_input_embeddings(self, value):
+        self.embeddings.word_embeddings = value
+
+    def _prune_heads(self, heads_to_prune):
+        """
+        Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base
+        class PreTrainedModel
+        """
+        for layer, heads in heads_to_prune.items():
+            self.encoder.layer[layer].attention.prune_heads(heads)
+
+    
+    def get_extended_attention_mask(self, attention_mask: Tensor, input_shape: Tuple[int], device: device, is_decoder: bool) -> Tensor:
+        """
+        Makes broadcastable attention and causal masks so that future and masked tokens are ignored.
+
+        Arguments:
+            attention_mask (:obj:`torch.Tensor`):
+                Mask with ones indicating tokens to attend to, zeros for tokens to ignore.
+            input_shape (:obj:`Tuple[int]`):
+                The shape of the input to the model.
+            device: (:obj:`torch.device`):
+                The device of the input to the model.
+
+        Returns:
+            :obj:`torch.Tensor` The extended attention mask, with a the same dtype as :obj:`attention_mask.dtype`.
+        """
+        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
+        # ourselves in which case we just need to make it broadcastable to all heads.
+        if attention_mask.dim() == 3:
+            extended_attention_mask = attention_mask[:, None, :, :]
+        elif attention_mask.dim() == 2:
+            # Provided a padding mask of dimensions [batch_size, seq_length]
+            # - if the model is a decoder, apply a causal mask in addition to the padding mask
+            # - if the model is an encoder, make the mask broadcastable to [batch_size, num_heads, seq_length, seq_length]
+            if is_decoder:
+                batch_size, seq_length = input_shape
+
+                seq_ids = torch.arange(seq_length, device=device)
+                causal_mask = seq_ids[None, None, :].repeat(batch_size, seq_length, 1) <= seq_ids[None, :, None]
+                # in case past_key_values are used we need to add a prefix ones mask to the causal mask
+                # causal and attention masks must have same type with pytorch version < 1.3
+                causal_mask = causal_mask.to(attention_mask.dtype)
+   
+                if causal_mask.shape[1] < attention_mask.shape[1]:
+                    prefix_seq_len = attention_mask.shape[1] - causal_mask.shape[1]
+                    causal_mask = torch.cat(
+                        [
+                            torch.ones((batch_size, seq_length, prefix_seq_len), device=device, dtype=causal_mask.dtype),
+                            causal_mask,
+                        ],
+                        axis=-1,
+                    )                     
+
+                extended_attention_mask = causal_mask[:, None, :, :] * attention_mask[:, None, None, :]
+            else:
+                extended_attention_mask = attention_mask[:, None, None, :]
+        else:
+            raise ValueError(
+                "Wrong shape for input_ids (shape {}) or attention_mask (shape {})".format(
+                    input_shape, attention_mask.shape
+                )
+            )
+
+        # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
+        # masked positions, this operation will create a tensor which is 0.0 for
+        # positions we want to attend and -10000.0 for masked positions.
+        # Since we are adding it to the raw scores before the softmax, this is
+        # effectively the same as removing these entirely.
+        extended_attention_mask = extended_attention_mask.to(dtype=self.dtype)  # fp16 compatibility
+        extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
+        return extended_attention_mask
+    
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        encoder_embeds=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_values=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+        is_decoder=False,
+        mode='multimodal',
+    ):
+        r"""
+        encoder_hidden_states  (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
+            Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
+            the model is configured as a decoder.
+        encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
+            Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
+            the cross-attention if the model is configured as a decoder. Mask values selected in ``[0, 1]``:
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+        past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
+            Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
+            If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`
+            (those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`
+            instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.
+        use_cache (:obj:`bool`, `optional`):
+            If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
+            decoding (see :obj:`past_key_values`).
+        """
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if is_decoder:
+            use_cache = use_cache if use_cache is not None else self.config.use_cache
+        else:
+            use_cache = False
+
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
+        elif input_ids is not None:
+            input_shape = input_ids.size()
+            batch_size, seq_length = input_shape
+            device = input_ids.device
+        elif inputs_embeds is not None:
+            input_shape = inputs_embeds.size()[:-1]
+            batch_size, seq_length = input_shape
+            device = inputs_embeds.device
+        elif encoder_embeds is not None:    
+            input_shape = encoder_embeds.size()[:-1]
+            batch_size, seq_length = input_shape 
+            device = encoder_embeds.device
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds or encoder_embeds")
+
+        # past_key_values_length
+        past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0
+
+        if attention_mask is None:
+            attention_mask = torch.ones(((batch_size, seq_length + past_key_values_length)), device=device)
+            
+        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
+        # ourselves in which case we just need to make it broadcastable to all heads.
+        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape, 
+                                                                                 device, is_decoder)
+
+        # If a 2D or 3D attention mask is provided for the cross-attention
+        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
+        if encoder_hidden_states is not None:
+            if type(encoder_hidden_states) == list:
+                encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states[0].size()
+            else:
+                encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
+            encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
+            
+            if type(encoder_attention_mask) == list:
+                encoder_extended_attention_mask = [self.invert_attention_mask(mask) for mask in encoder_attention_mask]
+            elif encoder_attention_mask is None:
+                encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
+                encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+            else:    
+                encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+        else:
+            encoder_extended_attention_mask = None
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x n_heads x N x N
+        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
+        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
+        head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)
+        
+        if encoder_embeds is None:
+            embedding_output = self.embeddings(
+                input_ids=input_ids,
+                position_ids=position_ids,
+                inputs_embeds=inputs_embeds,
+                past_key_values_length=past_key_values_length,
+            )
+        else:
+            embedding_output = encoder_embeds
+            
+        encoder_outputs = self.encoder(
+            embedding_output,
+            attention_mask=extended_attention_mask,
+            head_mask=head_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_extended_attention_mask,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            mode=mode,
+        )
+        sequence_output = encoder_outputs[0]
+        pooled_output = self.pooler(sequence_output) if self.pooler is not None else None
+
+        if not return_dict:
+            return (sequence_output, pooled_output) + encoder_outputs[1:]
+
+        return BaseModelOutputWithPoolingAndCrossAttentions(
+            last_hidden_state=sequence_output,
+            pooler_output=pooled_output,
+            past_key_values=encoder_outputs.past_key_values,
+            hidden_states=encoder_outputs.hidden_states,
+            attentions=encoder_outputs.attentions,
+            cross_attentions=encoder_outputs.cross_attentions,
+        )
+

extras/BLIP/models/vit.py

@@ -0,0 +1,308 @@
+'''
+ * Copyright (c) 2022, salesforce.com, inc.
+ * All rights reserved.
+ * SPDX-License-Identifier: BSD-3-Clause
+ * For full license text, see LICENSE.txt file in the repo root or https://opensource.org/licenses/BSD-3-Clause
+ * By Junnan Li
+ * Based on timm code base
+ * https://github.com/rwightman/pytorch-image-models/tree/master/timm
+'''
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from functools import partial
+
+from timm.models.vision_transformer import _cfg, PatchEmbed
+from timm.models.registry import register_model
+from timm.models.layers import trunc_normal_, DropPath
+from timm.models.helpers import named_apply, adapt_input_conv
+
+
+def checkpoint_wrapper(x):
+    return x
+
+
+class Mlp(nn.Module):
+    """ MLP as used in Vision Transformer, MLP-Mixer and related networks
+    """
+    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        self.fc1 = nn.Linear(in_features, hidden_features)
+        self.act = act_layer()
+        self.fc2 = nn.Linear(hidden_features, out_features)
+        self.drop = nn.Dropout(drop)
+
+    def forward(self, x):
+        x = self.fc1(x)
+        x = self.act(x)
+        x = self.drop(x)
+        x = self.fc2(x)
+        x = self.drop(x)
+        return x
+
+
+class Attention(nn.Module):
+    def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.):
+        super().__init__()
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        # NOTE scale factor was wrong in my original version, can set manually to be compat with prev weights
+        self.scale = qk_scale or head_dim ** -0.5
+        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
+        self.attn_drop = nn.Dropout(attn_drop)
+        self.proj = nn.Linear(dim, dim)
+        self.proj_drop = nn.Dropout(proj_drop)
+        self.attn_gradients = None
+        self.attention_map = None
+        
+    def save_attn_gradients(self, attn_gradients):
+        self.attn_gradients = attn_gradients
+        
+    def get_attn_gradients(self):
+        return self.attn_gradients
+    
+    def save_attention_map(self, attention_map):
+        self.attention_map = attention_map
+        
+    def get_attention_map(self):
+        return self.attention_map
+    
+    def forward(self, x, register_hook=False):
+        B, N, C = x.shape
+        qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
+        q, k, v = qkv[0], qkv[1], qkv[2]   # make torchscript happy (cannot use tensor as tuple)
+
+        attn = (q @ k.transpose(-2, -1)) * self.scale
+        attn = attn.softmax(dim=-1)
+        attn = self.attn_drop(attn)
+                
+        if register_hook:
+            self.save_attention_map(attn)
+            attn.register_hook(self.save_attn_gradients)        
+
+        x = (attn @ v).transpose(1, 2).reshape(B, N, C)
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x
+
+
+class Block(nn.Module):
+
+    def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop=0., attn_drop=0.,
+                 drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm, use_grad_checkpointing=False):
+        super().__init__()
+        self.norm1 = norm_layer(dim)
+        self.attn = Attention(
+            dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
+        # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
+        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
+        self.norm2 = norm_layer(dim)
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
+
+        if use_grad_checkpointing:
+            self.attn = checkpoint_wrapper(self.attn)
+            self.mlp = checkpoint_wrapper(self.mlp)
+
+    def forward(self, x, register_hook=False):
+        x = x + self.drop_path(self.attn(self.norm1(x), register_hook=register_hook))
+        x = x + self.drop_path(self.mlp(self.norm2(x)))
+        return x
+
+    
+class VisionTransformer(nn.Module):
+    """ Vision Transformer
+    A PyTorch impl of : `An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale`  -
+        https://arxiv.org/abs/2010.11929
+    """
+    def __init__(self, img_size=224, patch_size=16, in_chans=3, num_classes=1000, embed_dim=768, depth=12,
+                 num_heads=12, mlp_ratio=4., qkv_bias=True, qk_scale=None, representation_size=None,
+                 drop_rate=0., attn_drop_rate=0., drop_path_rate=0., norm_layer=None, 
+                 use_grad_checkpointing=False, ckpt_layer=0):
+        """
+        Args:
+            img_size (int, tuple): input image size
+            patch_size (int, tuple): patch size
+            in_chans (int): number of input channels
+            num_classes (int): number of classes for classification head
+            embed_dim (int): embedding dimension
+            depth (int): depth of transformer
+            num_heads (int): number of attention heads
+            mlp_ratio (int): ratio of mlp hidden dim to embedding dim
+            qkv_bias (bool): enable bias for qkv if True
+            qk_scale (float): override default qk scale of head_dim ** -0.5 if set
+            representation_size (Optional[int]): enable and set representation layer (pre-logits) to this value if set
+            drop_rate (float): dropout rate
+            attn_drop_rate (float): attention dropout rate
+            drop_path_rate (float): stochastic depth rate
+            norm_layer: (nn.Module): normalization layer
+        """
+        super().__init__()
+        self.num_features = self.embed_dim = embed_dim  # num_features for consistency with other models
+        norm_layer = norm_layer or partial(nn.LayerNorm, eps=1e-6)
+
+        self.patch_embed = PatchEmbed(
+            img_size=img_size, patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim)
+
+        num_patches = self.patch_embed.num_patches
+
+        self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim))
+        self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, embed_dim))
+        self.pos_drop = nn.Dropout(p=drop_rate)
+
+        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)]  # stochastic depth decay rule
+        self.blocks = nn.ModuleList([
+            Block(
+                dim=embed_dim, num_heads=num_heads, mlp_ratio=mlp_ratio, qkv_bias=qkv_bias, qk_scale=qk_scale,
+                drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer,
+                use_grad_checkpointing=(use_grad_checkpointing and i>=depth-ckpt_layer)
+            )
+            for i in range(depth)])
+        self.norm = norm_layer(embed_dim)
+
+        trunc_normal_(self.pos_embed, std=.02)
+        trunc_normal_(self.cls_token, std=.02)
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight, std=.02)
+            if isinstance(m, nn.Linear) and m.bias is not None:
+                nn.init.constant_(m.bias, 0)
+        elif isinstance(m, nn.LayerNorm):
+            nn.init.constant_(m.bias, 0)
+            nn.init.constant_(m.weight, 1.0)
+
+    @torch.jit.ignore
+    def no_weight_decay(self):
+        return {'pos_embed', 'cls_token'}
+
+    def forward(self, x, register_blk=-1):
+        B = x.shape[0]
+        x = self.patch_embed(x)
+
+        cls_tokens = self.cls_token.expand(B, -1, -1)  # stole cls_tokens impl from Phil Wang, thanks
+        x = torch.cat((cls_tokens, x), dim=1)
+  
+        x = x + self.pos_embed[:,:x.size(1),:]
+        x = self.pos_drop(x)
+
+        for i,blk in enumerate(self.blocks):
+            x = blk(x, register_blk==i)
+        x = self.norm(x)
+        
+        return x
+
+    @torch.jit.ignore()
+    def load_pretrained(self, checkpoint_path, prefix=''):
+        _load_weights(self, checkpoint_path, prefix)
+        
+
+@torch.no_grad()
+def _load_weights(model: VisionTransformer, checkpoint_path: str, prefix: str = ''):
+    """ Load weights from .npz checkpoints for official Google Brain Flax implementation
+    """
+    import numpy as np
+
+    def _n2p(w, t=True):
+        if w.ndim == 4 and w.shape[0] == w.shape[1] == w.shape[2] == 1:
+            w = w.flatten()
+        if t:
+            if w.ndim == 4:
+                w = w.transpose([3, 2, 0, 1])
+            elif w.ndim == 3:
+                w = w.transpose([2, 0, 1])
+            elif w.ndim == 2:
+                w = w.transpose([1, 0])
+        return torch.from_numpy(w)
+
+    w = np.load(checkpoint_path)
+    if not prefix and 'opt/target/embedding/kernel' in w:
+        prefix = 'opt/target/'
+
+    if hasattr(model.patch_embed, 'backbone'):
+        # hybrid
+        backbone = model.patch_embed.backbone
+        stem_only = not hasattr(backbone, 'stem')
+        stem = backbone if stem_only else backbone.stem
+        stem.conv.weight.copy_(adapt_input_conv(stem.conv.weight.shape[1], _n2p(w[f'{prefix}conv_root/kernel'])))
+        stem.norm.weight.copy_(_n2p(w[f'{prefix}gn_root/scale']))
+        stem.norm.bias.copy_(_n2p(w[f'{prefix}gn_root/bias']))
+        if not stem_only:
+            for i, stage in enumerate(backbone.stages):
+                for j, block in enumerate(stage.blocks):
+                    bp = f'{prefix}block{i + 1}/unit{j + 1}/'
+                    for r in range(3):
+                        getattr(block, f'conv{r + 1}').weight.copy_(_n2p(w[f'{bp}conv{r + 1}/kernel']))
+                        getattr(block, f'norm{r + 1}').weight.copy_(_n2p(w[f'{bp}gn{r + 1}/scale']))
+                        getattr(block, f'norm{r + 1}').bias.copy_(_n2p(w[f'{bp}gn{r + 1}/bias']))
+                    if block.downsample is not None:
+                        block.downsample.conv.weight.copy_(_n2p(w[f'{bp}conv_proj/kernel']))
+                        block.downsample.norm.weight.copy_(_n2p(w[f'{bp}gn_proj/scale']))
+                        block.downsample.norm.bias.copy_(_n2p(w[f'{bp}gn_proj/bias']))
+        embed_conv_w = _n2p(w[f'{prefix}embedding/kernel'])
+    else:
+        embed_conv_w = adapt_input_conv(
+            model.patch_embed.proj.weight.shape[1], _n2p(w[f'{prefix}embedding/kernel']))
+    model.patch_embed.proj.weight.copy_(embed_conv_w)
+    model.patch_embed.proj.bias.copy_(_n2p(w[f'{prefix}embedding/bias']))
+    model.cls_token.copy_(_n2p(w[f'{prefix}cls'], t=False))
+    pos_embed_w = _n2p(w[f'{prefix}Transformer/posembed_input/pos_embedding'], t=False)
+    if pos_embed_w.shape != model.pos_embed.shape:
+        pos_embed_w = resize_pos_embed(  # resize pos embedding when different size from pretrained weights
+            pos_embed_w, model.pos_embed, getattr(model, 'num_tokens', 1), model.patch_embed.grid_size)
+    model.pos_embed.copy_(pos_embed_w)
+    model.norm.weight.copy_(_n2p(w[f'{prefix}Transformer/encoder_norm/scale']))
+    model.norm.bias.copy_(_n2p(w[f'{prefix}Transformer/encoder_norm/bias']))
+#     if isinstance(model.head, nn.Linear) and model.head.bias.shape[0] == w[f'{prefix}head/bias'].shape[-1]:
+#         model.head.weight.copy_(_n2p(w[f'{prefix}head/kernel']))
+#         model.head.bias.copy_(_n2p(w[f'{prefix}head/bias']))
+#     if isinstance(getattr(model.pre_logits, 'fc', None), nn.Linear) and f'{prefix}pre_logits/bias' in w:
+#         model.pre_logits.fc.weight.copy_(_n2p(w[f'{prefix}pre_logits/kernel']))
+#         model.pre_logits.fc.bias.copy_(_n2p(w[f'{prefix}pre_logits/bias']))
+    for i, block in enumerate(model.blocks.children()):
+        block_prefix = f'{prefix}Transformer/encoderblock_{i}/'
+        mha_prefix = block_prefix + 'MultiHeadDotProductAttention_1/'
+        block.norm1.weight.copy_(_n2p(w[f'{block_prefix}LayerNorm_0/scale']))
+        block.norm1.bias.copy_(_n2p(w[f'{block_prefix}LayerNorm_0/bias']))
+        block.attn.qkv.weight.copy_(torch.cat([
+            _n2p(w[f'{mha_prefix}{n}/kernel'], t=False).flatten(1).T for n in ('query', 'key', 'value')]))
+        block.attn.qkv.bias.copy_(torch.cat([
+            _n2p(w[f'{mha_prefix}{n}/bias'], t=False).reshape(-1) for n in ('query', 'key', 'value')]))
+        block.attn.proj.weight.copy_(_n2p(w[f'{mha_prefix}out/kernel']).flatten(1))
+        block.attn.proj.bias.copy_(_n2p(w[f'{mha_prefix}out/bias']))
+        for r in range(2):
+            getattr(block.mlp, f'fc{r + 1}').weight.copy_(_n2p(w[f'{block_prefix}MlpBlock_3/Dense_{r}/kernel']))
+            getattr(block.mlp, f'fc{r + 1}').bias.copy_(_n2p(w[f'{block_prefix}MlpBlock_3/Dense_{r}/bias']))
+        block.norm2.weight.copy_(_n2p(w[f'{block_prefix}LayerNorm_2/scale']))
+        block.norm2.bias.copy_(_n2p(w[f'{block_prefix}LayerNorm_2/bias']))
+
+            
+def interpolate_pos_embed(pos_embed_checkpoint, visual_encoder):        
+    # interpolate position embedding
+    embedding_size = pos_embed_checkpoint.shape[-1]
+    num_patches = visual_encoder.patch_embed.num_patches
+    num_extra_tokens = visual_encoder.pos_embed.shape[-2] - num_patches
+    # height (== width) for the checkpoint position embedding
+    orig_size = int((pos_embed_checkpoint.shape[-2] - num_extra_tokens) ** 0.5)
+    # height (== width) for the new position embedding
+    new_size = int(num_patches ** 0.5)
+
+    if orig_size!=new_size:
+        # class_token and dist_token are kept unchanged
+        extra_tokens = pos_embed_checkpoint[:, :num_extra_tokens]
+        # only the position tokens are interpolated
+        pos_tokens = pos_embed_checkpoint[:, num_extra_tokens:]
+        pos_tokens = pos_tokens.reshape(-1, orig_size, orig_size, embedding_size).permute(0, 3, 1, 2)
+        pos_tokens = torch.nn.functional.interpolate(
+            pos_tokens, size=(new_size, new_size), mode='bicubic', align_corners=False)
+        pos_tokens = pos_tokens.permute(0, 2, 3, 1).flatten(1, 2)
+        new_pos_embed = torch.cat((extra_tokens, pos_tokens), dim=1)
+        print('reshape position embedding from %d to %d'%(orig_size ** 2,new_size ** 2))
+        
+        return new_pos_embed    
+    else:
+        return pos_embed_checkpoint

extras/expansion.py

@@ -0,0 +1,126 @@
+# Fooocus GPT2 Expansion
+# Algorithm created by Lvmin Zhang at 2023, Stanford
+# If used inside Fooocus, any use is permitted.
+# If used outside Fooocus, only non-commercial use is permitted (CC-By NC 4.0).
+# This applies to the word list, vocab, model, and algorithm.
+
+
+import os
+import torch
+import math
+import ldm_patched.modules.model_management as model_management
+
+from transformers.generation.logits_process import LogitsProcessorList
+from transformers import AutoTokenizer, AutoModelForCausalLM, set_seed
+from modules.config import path_fooocus_expansion
+from ldm_patched.modules.model_patcher import ModelPatcher
+
+
+# limitation of np.random.seed(), called from transformers.set_seed()
+SEED_LIMIT_NUMPY = 2**32
+neg_inf = - 8192.0
+
+
+def safe_str(x):
+    x = str(x)
+    for _ in range(16):
+        x = x.replace('  ', ' ')
+    return x.strip(",. \r\n")
+
+
+def remove_pattern(x, pattern):
+    for p in pattern:
+        x = x.replace(p, '')
+    return x
+
+
+class FooocusExpansion:
+    def __init__(self):
+        self.tokenizer = AutoTokenizer.from_pretrained(path_fooocus_expansion)
+
+        positive_words = open(os.path.join(path_fooocus_expansion, 'positive.txt'),
+                              encoding='utf-8').read().splitlines()
+        positive_words = ['Ġ' + x.lower() for x in positive_words if x != '']
+
+        self.logits_bias = torch.zeros((1, len(self.tokenizer.vocab)), dtype=torch.float32) + neg_inf
+
+        debug_list = []
+        for k, v in self.tokenizer.vocab.items():
+            if k in positive_words:
+                self.logits_bias[0, v] = 0
+                debug_list.append(k[1:])
+
+        print(f'Fooocus V2 Expansion: Vocab with {len(debug_list)} words.')
+
+        # debug_list = '\n'.join(sorted(debug_list))
+        # print(debug_list)
+
+        # t11 = self.tokenizer(',', return_tensors="np")
+        # t198 = self.tokenizer('\n', return_tensors="np")
+        # eos = self.tokenizer.eos_token_id
+
+        self.model = AutoModelForCausalLM.from_pretrained(path_fooocus_expansion)
+        self.model.eval()
+
+        load_device = model_management.text_encoder_device()
+        offload_device = model_management.text_encoder_offload_device()
+
+        # MPS hack
+        if model_management.is_device_mps(load_device):
+            load_device = torch.device('cpu')
+            offload_device = torch.device('cpu')
+
+        use_fp16 = model_management.should_use_fp16(device=load_device)
+
+        if use_fp16:
+            self.model.half()
+
+        self.patcher = ModelPatcher(self.model, load_device=load_device, offload_device=offload_device)
+        print(f'Fooocus Expansion engine loaded for {load_device}, use_fp16 = {use_fp16}.')
+
+    @torch.no_grad()
+    @torch.inference_mode()
+    def logits_processor(self, input_ids, scores):
+        assert scores.ndim == 2 and scores.shape[0] == 1
+        self.logits_bias = self.logits_bias.to(scores)
+
+        bias = self.logits_bias.clone()
+        bias[0, input_ids[0].to(bias.device).long()] = neg_inf
+        bias[0, 11] = 0
+
+        return scores + bias
+
+    @torch.no_grad()
+    @torch.inference_mode()
+    def __call__(self, prompt, seed):
+        if prompt == '':
+            return ''
+
+        if self.patcher.current_device != self.patcher.load_device:
+            print('Fooocus Expansion loaded by itself.')
+            model_management.load_model_gpu(self.patcher)
+
+        seed = int(seed) % SEED_LIMIT_NUMPY
+        set_seed(seed)
+        prompt = safe_str(prompt) + ','
+
+        tokenized_kwargs = self.tokenizer(prompt, return_tensors="pt")
+        tokenized_kwargs.data['input_ids'] = tokenized_kwargs.data['input_ids'].to(self.patcher.load_device)
+        tokenized_kwargs.data['attention_mask'] = tokenized_kwargs.data['attention_mask'].to(self.patcher.load_device)
+
+        current_token_length = int(tokenized_kwargs.data['input_ids'].shape[1])
+        max_token_length = 75 * int(math.ceil(float(current_token_length) / 75.0))
+        max_new_tokens = max_token_length - current_token_length
+
+        # https://huggingface.co/blog/introducing-csearch
+        # https://huggingface.co/docs/transformers/generation_strategies
+        features = self.model.generate(**tokenized_kwargs,
+                                       top_k=100,
+                                       max_new_tokens=max_new_tokens,
+                                       do_sample=True,
+                                       logits_processor=LogitsProcessorList([self.logits_processor]))
+
+        response = self.tokenizer.batch_decode(features, skip_special_tokens=True)
+        result = safe_str(response[0])
+
+        return result

extras/face_crop.py

@@ -0,0 +1,50 @@
+import cv2
+import numpy as np
+import modules.config
+
+
+faceRestoreHelper = None
+
+
+def align_warp_face(self, landmark, border_mode='constant'):
+    affine_matrix = cv2.estimateAffinePartial2D(landmark, self.face_template, method=cv2.LMEDS)[0]
+    self.affine_matrices.append(affine_matrix)
+    if border_mode == 'constant':
+        border_mode = cv2.BORDER_CONSTANT
+    elif border_mode == 'reflect101':
+        border_mode = cv2.BORDER_REFLECT101
+    elif border_mode == 'reflect':
+        border_mode = cv2.BORDER_REFLECT
+    input_img = self.input_img
+    cropped_face = cv2.warpAffine(input_img, affine_matrix, self.face_size,
+                                  borderMode=border_mode, borderValue=(135, 133, 132))
+    return cropped_face
+
+
+def crop_image(img_rgb):
+    global faceRestoreHelper
+    
+    if faceRestoreHelper is None:
+        from extras.facexlib.utils.face_restoration_helper import FaceRestoreHelper
+        faceRestoreHelper = FaceRestoreHelper(
+            upscale_factor=1,
+            model_rootpath=modules.config.path_controlnet,
+            device='cpu'  # use cpu is safer since we are out of memory management
+        )
+
+    faceRestoreHelper.clean_all()
+    faceRestoreHelper.read_image(np.ascontiguousarray(img_rgb[:, :, ::-1].copy()))
+    faceRestoreHelper.get_face_landmarks_5()
+
+    landmarks = faceRestoreHelper.all_landmarks_5
+    # landmarks are already sorted with confidence.
+
+    if len(landmarks) == 0:
+        print('No face detected')
+        return img_rgb
+    else:
+        print(f'Detected {len(landmarks)} faces')
+
+    result = align_warp_face(faceRestoreHelper, landmarks[0])
+
+    return np.ascontiguousarray(result[:, :, ::-1].copy())

extras/facexlib/detection/__init__.py

@@ -0,0 +1,31 @@
+import torch
+from copy import deepcopy
+
+from extras.facexlib.utils import load_file_from_url
+from .retinaface import RetinaFace
+
+
+def init_detection_model(model_name, half=False, device='cuda', model_rootpath=None):
+    if model_name == 'retinaface_resnet50':
+        model = RetinaFace(network_name='resnet50', half=half, device=device)
+        model_url = 'https://github.com/xinntao/facexlib/releases/download/v0.1.0/detection_Resnet50_Final.pth'
+    elif model_name == 'retinaface_mobile0.25':
+        model = RetinaFace(network_name='mobile0.25', half=half, device=device)
+        model_url = 'https://github.com/xinntao/facexlib/releases/download/v0.1.0/detection_mobilenet0.25_Final.pth'
+    else:
+        raise NotImplementedError(f'{model_name} is not implemented.')
+
+    model_path = load_file_from_url(
+        url=model_url, model_dir='facexlib/weights', progress=True, file_name=None, save_dir=model_rootpath)
+
+    # TODO: clean pretrained model
+    load_net = torch.load(model_path, map_location=lambda storage, loc: storage)
+    # remove unnecessary 'module.'
+    for k, v in deepcopy(load_net).items():
+        if k.startswith('module.'):
+            load_net[k[7:]] = v
+            load_net.pop(k)
+    model.load_state_dict(load_net, strict=True)
+    model.eval()
+    model = model.to(device)
+    return model

extras/facexlib/detection/align_trans.py

@@ -0,0 +1,219 @@
+import cv2
+import numpy as np
+
+from .matlab_cp2tform import get_similarity_transform_for_cv2
+
+# reference facial points, a list of coordinates (x,y)
+REFERENCE_FACIAL_POINTS = [[30.29459953, 51.69630051], [65.53179932, 51.50139999], [48.02519989, 71.73660278],
+                           [33.54930115, 92.3655014], [62.72990036, 92.20410156]]
+
+DEFAULT_CROP_SIZE = (96, 112)
+
+
+class FaceWarpException(Exception):
+
+    def __str__(self):
+        return 'In File {}:{}'.format(__file__, super.__str__(self))
+
+
+def get_reference_facial_points(output_size=None, inner_padding_factor=0.0, outer_padding=(0, 0), default_square=False):
+    """
+    Function:
+    ----------
+        get reference 5 key points according to crop settings:
+        0. Set default crop_size:
+            if default_square:
+                crop_size = (112, 112)
+            else:
+                crop_size = (96, 112)
+        1. Pad the crop_size by inner_padding_factor in each side;
+        2. Resize crop_size into (output_size - outer_padding*2),
+            pad into output_size with outer_padding;
+        3. Output reference_5point;
+    Parameters:
+    ----------
+        @output_size: (w, h) or None
+            size of aligned face image
+        @inner_padding_factor: (w_factor, h_factor)
+            padding factor for inner (w, h)
+        @outer_padding: (w_pad, h_pad)
+            each row is a pair of coordinates (x, y)
+        @default_square: True or False
+            if True:
+                default crop_size = (112, 112)
+            else:
+                default crop_size = (96, 112);
+        !!! make sure, if output_size is not None:
+                (output_size - outer_padding)
+                = some_scale * (default crop_size * (1.0 +
+                inner_padding_factor))
+    Returns:
+    ----------
+        @reference_5point: 5x2 np.array
+            each row is a pair of transformed coordinates (x, y)
+    """
+
+    tmp_5pts = np.array(REFERENCE_FACIAL_POINTS)
+    tmp_crop_size = np.array(DEFAULT_CROP_SIZE)
+
+    # 0) make the inner region a square
+    if default_square:
+        size_diff = max(tmp_crop_size) - tmp_crop_size
+        tmp_5pts += size_diff / 2
+        tmp_crop_size += size_diff
+
+    if (output_size and output_size[0] == tmp_crop_size[0] and output_size[1] == tmp_crop_size[1]):
+
+        return tmp_5pts
+
+    if (inner_padding_factor == 0 and outer_padding == (0, 0)):
+        if output_size is None:
+            return tmp_5pts
+        else:
+            raise FaceWarpException('No paddings to do, output_size must be None or {}'.format(tmp_crop_size))
+
+    # check output size
+    if not (0 <= inner_padding_factor <= 1.0):
+        raise FaceWarpException('Not (0 <= inner_padding_factor <= 1.0)')
+
+    if ((inner_padding_factor > 0 or outer_padding[0] > 0 or outer_padding[1] > 0) and output_size is None):
+        output_size = tmp_crop_size * \
+            (1 + inner_padding_factor * 2).astype(np.int32)
+        output_size += np.array(outer_padding)
+    if not (outer_padding[0] < output_size[0] and outer_padding[1] < output_size[1]):
+        raise FaceWarpException('Not (outer_padding[0] < output_size[0] and outer_padding[1] < output_size[1])')
+
+    # 1) pad the inner region according inner_padding_factor
+    if inner_padding_factor > 0:
+        size_diff = tmp_crop_size * inner_padding_factor * 2
+        tmp_5pts += size_diff / 2
+        tmp_crop_size += np.round(size_diff).astype(np.int32)
+
+    # 2) resize the padded inner region
+    size_bf_outer_pad = np.array(output_size) - np.array(outer_padding) * 2
+
+    if size_bf_outer_pad[0] * tmp_crop_size[1] != size_bf_outer_pad[1] * tmp_crop_size[0]:
+        raise FaceWarpException('Must have (output_size - outer_padding)'
+                                '= some_scale * (crop_size * (1.0 + inner_padding_factor)')
+
+    scale_factor = size_bf_outer_pad[0].astype(np.float32) / tmp_crop_size[0]
+    tmp_5pts = tmp_5pts * scale_factor
+    #    size_diff = tmp_crop_size * (scale_factor - min(scale_factor))
+    #    tmp_5pts = tmp_5pts + size_diff / 2
+    tmp_crop_size = size_bf_outer_pad
+
+    # 3) add outer_padding to make output_size
+    reference_5point = tmp_5pts + np.array(outer_padding)
+    tmp_crop_size = output_size
+
+    return reference_5point
+
+
+def get_affine_transform_matrix(src_pts, dst_pts):
+    """
+    Function:
+    ----------
+        get affine transform matrix 'tfm' from src_pts to dst_pts
+    Parameters:
+    ----------
+        @src_pts: Kx2 np.array
+            source points matrix, each row is a pair of coordinates (x, y)
+        @dst_pts: Kx2 np.array
+            destination points matrix, each row is a pair of coordinates (x, y)
+    Returns:
+    ----------
+        @tfm: 2x3 np.array
+            transform matrix from src_pts to dst_pts
+    """
+
+    tfm = np.float32([[1, 0, 0], [0, 1, 0]])
+    n_pts = src_pts.shape[0]
+    ones = np.ones((n_pts, 1), src_pts.dtype)
+    src_pts_ = np.hstack([src_pts, ones])
+    dst_pts_ = np.hstack([dst_pts, ones])
+
+    A, res, rank, s = np.linalg.lstsq(src_pts_, dst_pts_)
+
+    if rank == 3:
+        tfm = np.float32([[A[0, 0], A[1, 0], A[2, 0]], [A[0, 1], A[1, 1], A[2, 1]]])
+    elif rank == 2:
+        tfm = np.float32([[A[0, 0], A[1, 0], 0], [A[0, 1], A[1, 1], 0]])
+
+    return tfm
+
+
+def warp_and_crop_face(src_img, facial_pts, reference_pts=None, crop_size=(96, 112), align_type='smilarity'):
+    """
+    Function:
+    ----------
+        apply affine transform 'trans' to uv
+    Parameters:
+    ----------
+        @src_img: 3x3 np.array
+            input image
+        @facial_pts: could be
+            1)a list of K coordinates (x,y)
+        or
+            2) Kx2 or 2xK np.array
+            each row or col is a pair of coordinates (x, y)
+        @reference_pts: could be
+            1) a list of K coordinates (x,y)
+        or
+            2) Kx2 or 2xK np.array
+            each row or col is a pair of coordinates (x, y)
+        or
+            3) None
+            if None, use default reference facial points
+        @crop_size: (w, h)
+            output face image size
+        @align_type: transform type, could be one of
+            1) 'similarity': use similarity transform
+            2) 'cv2_affine': use the first 3 points to do affine transform,
+                    by calling cv2.getAffineTransform()
+            3) 'affine': use all points to do affine transform
+    Returns:
+    ----------
+        @face_img: output face image with size (w, h) = @crop_size
+    """
+
+    if reference_pts is None:
+        if crop_size[0] == 96 and crop_size[1] == 112:
+            reference_pts = REFERENCE_FACIAL_POINTS
+        else:
+            default_square = False
+            inner_padding_factor = 0
+            outer_padding = (0, 0)
+            output_size = crop_size
+
+            reference_pts = get_reference_facial_points(output_size, inner_padding_factor, outer_padding,
+                                                        default_square)
+
+    ref_pts = np.float32(reference_pts)
+    ref_pts_shp = ref_pts.shape
+    if max(ref_pts_shp) < 3 or min(ref_pts_shp) != 2:
+        raise FaceWarpException('reference_pts.shape must be (K,2) or (2,K) and K>2')
+
+    if ref_pts_shp[0] == 2:
+        ref_pts = ref_pts.T
+
+    src_pts = np.float32(facial_pts)
+    src_pts_shp = src_pts.shape
+    if max(src_pts_shp) < 3 or min(src_pts_shp) != 2:
+        raise FaceWarpException('facial_pts.shape must be (K,2) or (2,K) and K>2')
+
+    if src_pts_shp[0] == 2:
+        src_pts = src_pts.T
+
+    if src_pts.shape != ref_pts.shape:
+        raise FaceWarpException('facial_pts and reference_pts must have the same shape')
+
+    if align_type == 'cv2_affine':
+        tfm = cv2.getAffineTransform(src_pts[0:3], ref_pts[0:3])
+    elif align_type == 'affine':
+        tfm = get_affine_transform_matrix(src_pts, ref_pts)
+    else:
+        tfm = get_similarity_transform_for_cv2(src_pts, ref_pts)
+
+    face_img = cv2.warpAffine(src_img, tfm, (crop_size[0], crop_size[1]))
+
+    return face_img

extras/facexlib/detection/matlab_cp2tform.py

@@ -0,0 +1,317 @@
+import numpy as np
+from numpy.linalg import inv, lstsq
+from numpy.linalg import matrix_rank as rank
+from numpy.linalg import norm
+
+
+class MatlabCp2tormException(Exception):
+
+    def __str__(self):
+        return 'In File {}:{}'.format(__file__, super.__str__(self))
+
+
+def tformfwd(trans, uv):
+    """
+    Function:
+    ----------
+        apply affine transform 'trans' to uv
+
+    Parameters:
+    ----------
+        @trans: 3x3 np.array
+            transform matrix
+        @uv: Kx2 np.array
+            each row is a pair of coordinates (x, y)
+
+    Returns:
+    ----------
+        @xy: Kx2 np.array
+            each row is a pair of transformed coordinates (x, y)
+    """
+    uv = np.hstack((uv, np.ones((uv.shape[0], 1))))
+    xy = np.dot(uv, trans)
+    xy = xy[:, 0:-1]
+    return xy
+
+
+def tforminv(trans, uv):
+    """
+    Function:
+    ----------
+        apply the inverse of affine transform 'trans' to uv
+
+    Parameters:
+    ----------
+        @trans: 3x3 np.array
+            transform matrix
+        @uv: Kx2 np.array
+            each row is a pair of coordinates (x, y)
+
+    Returns:
+    ----------
+        @xy: Kx2 np.array
+            each row is a pair of inverse-transformed coordinates (x, y)
+    """
+    Tinv = inv(trans)
+    xy = tformfwd(Tinv, uv)
+    return xy
+
+
+def findNonreflectiveSimilarity(uv, xy, options=None):
+    options = {'K': 2}
+
+    K = options['K']
+    M = xy.shape[0]
+    x = xy[:, 0].reshape((-1, 1))  # use reshape to keep a column vector
+    y = xy[:, 1].reshape((-1, 1))  # use reshape to keep a column vector
+
+    tmp1 = np.hstack((x, y, np.ones((M, 1)), np.zeros((M, 1))))
+    tmp2 = np.hstack((y, -x, np.zeros((M, 1)), np.ones((M, 1))))
+    X = np.vstack((tmp1, tmp2))
+
+    u = uv[:, 0].reshape((-1, 1))  # use reshape to keep a column vector
+    v = uv[:, 1].reshape((-1, 1))  # use reshape to keep a column vector
+    U = np.vstack((u, v))
+
+    # We know that X * r = U
+    if rank(X) >= 2 * K:
+        r, _, _, _ = lstsq(X, U, rcond=-1)
+        r = np.squeeze(r)
+    else:
+        raise Exception('cp2tform:twoUniquePointsReq')
+    sc = r[0]
+    ss = r[1]
+    tx = r[2]
+    ty = r[3]
+
+    Tinv = np.array([[sc, -ss, 0], [ss, sc, 0], [tx, ty, 1]])
+    T = inv(Tinv)
+    T[:, 2] = np.array([0, 0, 1])
+
+    return T, Tinv
+
+
+def findSimilarity(uv, xy, options=None):
+    options = {'K': 2}
+
+    #    uv = np.array(uv)
+    #    xy = np.array(xy)
+
+    # Solve for trans1
+    trans1, trans1_inv = findNonreflectiveSimilarity(uv, xy, options)
+
+    # Solve for trans2
+
+    # manually reflect the xy data across the Y-axis
+    xyR = xy
+    xyR[:, 0] = -1 * xyR[:, 0]
+
+    trans2r, trans2r_inv = findNonreflectiveSimilarity(uv, xyR, options)
+
+    # manually reflect the tform to undo the reflection done on xyR
+    TreflectY = np.array([[-1, 0, 0], [0, 1, 0], [0, 0, 1]])
+
+    trans2 = np.dot(trans2r, TreflectY)
+
+    # Figure out if trans1 or trans2 is better
+    xy1 = tformfwd(trans1, uv)
+    norm1 = norm(xy1 - xy)
+
+    xy2 = tformfwd(trans2, uv)
+    norm2 = norm(xy2 - xy)
+
+    if norm1 <= norm2:
+        return trans1, trans1_inv
+    else:
+        trans2_inv = inv(trans2)
+        return trans2, trans2_inv
+
+
+def get_similarity_transform(src_pts, dst_pts, reflective=True):
+    """
+    Function:
+    ----------
+        Find Similarity Transform Matrix 'trans':
+            u = src_pts[:, 0]
+            v = src_pts[:, 1]
+            x = dst_pts[:, 0]
+            y = dst_pts[:, 1]
+            [x, y, 1] = [u, v, 1] * trans
+
+    Parameters:
+    ----------
+        @src_pts: Kx2 np.array
+            source points, each row is a pair of coordinates (x, y)
+        @dst_pts: Kx2 np.array
+            destination points, each row is a pair of transformed
+            coordinates (x, y)
+        @reflective: True or False
+            if True:
+                use reflective similarity transform
+            else:
+                use non-reflective similarity transform
+
+    Returns:
+    ----------
+       @trans: 3x3 np.array
+            transform matrix from uv to xy
+        trans_inv: 3x3 np.array
+            inverse of trans, transform matrix from xy to uv
+    """
+
+    if reflective:
+        trans, trans_inv = findSimilarity(src_pts, dst_pts)
+    else:
+        trans, trans_inv = findNonreflectiveSimilarity(src_pts, dst_pts)
+
+    return trans, trans_inv
+
+
+def cvt_tform_mat_for_cv2(trans):
+    """
+    Function:
+    ----------
+        Convert Transform Matrix 'trans' into 'cv2_trans' which could be
+        directly used by cv2.warpAffine():
+            u = src_pts[:, 0]
+            v = src_pts[:, 1]
+            x = dst_pts[:, 0]
+            y = dst_pts[:, 1]
+            [x, y].T = cv_trans * [u, v, 1].T
+
+    Parameters:
+    ----------
+        @trans: 3x3 np.array
+            transform matrix from uv to xy
+
+    Returns:
+    ----------
+        @cv2_trans: 2x3 np.array
+            transform matrix from src_pts to dst_pts, could be directly used
+            for cv2.warpAffine()
+    """
+    cv2_trans = trans[:, 0:2].T
+
+    return cv2_trans
+
+
+def get_similarity_transform_for_cv2(src_pts, dst_pts, reflective=True):
+    """
+    Function:
+    ----------
+        Find Similarity Transform Matrix 'cv2_trans' which could be
+        directly used by cv2.warpAffine():
+            u = src_pts[:, 0]
+            v = src_pts[:, 1]
+            x = dst_pts[:, 0]
+            y = dst_pts[:, 1]
+            [x, y].T = cv_trans * [u, v, 1].T
+
+    Parameters:
+    ----------
+        @src_pts: Kx2 np.array
+            source points, each row is a pair of coordinates (x, y)
+        @dst_pts: Kx2 np.array
+            destination points, each row is a pair of transformed
+            coordinates (x, y)
+        reflective: True or False
+            if True:
+                use reflective similarity transform
+            else:
+                use non-reflective similarity transform
+
+    Returns:
+    ----------
+        @cv2_trans: 2x3 np.array
+            transform matrix from src_pts to dst_pts, could be directly used
+            for cv2.warpAffine()
+    """
+    trans, trans_inv = get_similarity_transform(src_pts, dst_pts, reflective)
+    cv2_trans = cvt_tform_mat_for_cv2(trans)
+
+    return cv2_trans
+
+
+if __name__ == '__main__':
+    """
+    u = [0, 6, -2]
+    v = [0, 3, 5]
+    x = [-1, 0, 4]
+    y = [-1, -10, 4]
+
+    # In Matlab, run:
+    #
+    #   uv = [u'; v'];
+    #   xy = [x'; y'];
+    #   tform_sim=cp2tform(uv,xy,'similarity');
+    #
+    #   trans = tform_sim.tdata.T
+    #   ans =
+    #       -0.0764   -1.6190         0
+    #        1.6190   -0.0764         0
+    #       -3.2156    0.0290    1.0000
+    #   trans_inv = tform_sim.tdata.Tinv
+    #    ans =
+    #
+    #       -0.0291    0.6163         0
+    #       -0.6163   -0.0291         0
+    #       -0.0756    1.9826    1.0000
+    #    xy_m=tformfwd(tform_sim, u,v)
+    #
+    #    xy_m =
+    #
+    #       -3.2156    0.0290
+    #        1.1833   -9.9143
+    #        5.0323    2.8853
+    #    uv_m=tforminv(tform_sim, x,y)
+    #
+    #    uv_m =
+    #
+    #        0.5698    1.3953
+    #        6.0872    2.2733
+    #       -2.6570    4.3314
+    """
+    u = [0, 6, -2]
+    v = [0, 3, 5]
+    x = [-1, 0, 4]
+    y = [-1, -10, 4]
+
+    uv = np.array((u, v)).T
+    xy = np.array((x, y)).T
+
+    print('\n--->uv:')
+    print(uv)
+    print('\n--->xy:')
+    print(xy)
+
+    trans, trans_inv = get_similarity_transform(uv, xy)
+
+    print('\n--->trans matrix:')
+    print(trans)
+
+    print('\n--->trans_inv matrix:')
+    print(trans_inv)
+
+    print('\n---> apply transform to uv')
+    print('\nxy_m = uv_augmented * trans')
+    uv_aug = np.hstack((uv, np.ones((uv.shape[0], 1))))
+    xy_m = np.dot(uv_aug, trans)
+    print(xy_m)
+
+    print('\nxy_m = tformfwd(trans, uv)')
+    xy_m = tformfwd(trans, uv)
+    print(xy_m)
+
+    print('\n---> apply inverse transform to xy')
+    print('\nuv_m = xy_augmented * trans_inv')
+    xy_aug = np.hstack((xy, np.ones((xy.shape[0], 1))))
+    uv_m = np.dot(xy_aug, trans_inv)
+    print(uv_m)
+
+    print('\nuv_m = tformfwd(trans_inv, xy)')
+    uv_m = tformfwd(trans_inv, xy)
+    print(uv_m)
+
+    uv_m = tforminv(trans, xy)
+    print('\nuv_m = tforminv(trans, xy)')
+    print(uv_m)

extras/facexlib/detection/retinaface.py

@@ -0,0 +1,366 @@
+import cv2
+import numpy as np
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from PIL import Image
+from torchvision.models._utils import IntermediateLayerGetter as IntermediateLayerGetter
+
+from extras.facexlib.detection.align_trans import get_reference_facial_points, warp_and_crop_face
+from extras.facexlib.detection.retinaface_net import FPN, SSH, MobileNetV1, make_bbox_head, make_class_head, make_landmark_head
+from extras.facexlib.detection.retinaface_utils import (PriorBox, batched_decode, batched_decode_landm, decode, decode_landm,
+                                                 py_cpu_nms)
+
+
+def generate_config(network_name):
+
+    cfg_mnet = {
+        'name': 'mobilenet0.25',
+        'min_sizes': [[16, 32], [64, 128], [256, 512]],
+        'steps': [8, 16, 32],
+        'variance': [0.1, 0.2],
+        'clip': False,
+        'loc_weight': 2.0,
+        'gpu_train': True,
+        'batch_size': 32,
+        'ngpu': 1,
+        'epoch': 250,
+        'decay1': 190,
+        'decay2': 220,
+        'image_size': 640,
+        'return_layers': {
+            'stage1': 1,
+            'stage2': 2,
+            'stage3': 3
+        },
+        'in_channel': 32,
+        'out_channel': 64
+    }
+
+    cfg_re50 = {
+        'name': 'Resnet50',
+        'min_sizes': [[16, 32], [64, 128], [256, 512]],
+        'steps': [8, 16, 32],
+        'variance': [0.1, 0.2],
+        'clip': False,
+        'loc_weight': 2.0,
+        'gpu_train': True,
+        'batch_size': 24,
+        'ngpu': 4,
+        'epoch': 100,
+        'decay1': 70,
+        'decay2': 90,
+        'image_size': 840,
+        'return_layers': {
+            'layer2': 1,
+            'layer3': 2,
+            'layer4': 3
+        },
+        'in_channel': 256,
+        'out_channel': 256
+    }
+
+    if network_name == 'mobile0.25':
+        return cfg_mnet
+    elif network_name == 'resnet50':
+        return cfg_re50
+    else:
+        raise NotImplementedError(f'network_name={network_name}')
+
+
+class RetinaFace(nn.Module):
+
+    def __init__(self, network_name='resnet50', half=False, phase='test', device=None):
+        self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') if device is None else device
+
+        super(RetinaFace, self).__init__()
+        self.half_inference = half
+        cfg = generate_config(network_name)
+        self.backbone = cfg['name']
+
+        self.model_name = f'retinaface_{network_name}'
+        self.cfg = cfg
+        self.phase = phase
+        self.target_size, self.max_size = 1600, 2150
+        self.resize, self.scale, self.scale1 = 1., None, None
+        self.mean_tensor = torch.tensor([[[[104.]], [[117.]], [[123.]]]], device=self.device)
+        self.reference = get_reference_facial_points(default_square=True)
+        # Build network.
+        backbone = None
+        if cfg['name'] == 'mobilenet0.25':
+            backbone = MobileNetV1()
+            self.body = IntermediateLayerGetter(backbone, cfg['return_layers'])
+        elif cfg['name'] == 'Resnet50':
+            import torchvision.models as models
+            backbone = models.resnet50(weights=None)
+            self.body = IntermediateLayerGetter(backbone, cfg['return_layers'])
+
+        in_channels_stage2 = cfg['in_channel']
+        in_channels_list = [
+            in_channels_stage2 * 2,
+            in_channels_stage2 * 4,
+            in_channels_stage2 * 8,
+        ]
+
+        out_channels = cfg['out_channel']
+        self.fpn = FPN(in_channels_list, out_channels)
+        self.ssh1 = SSH(out_channels, out_channels)
+        self.ssh2 = SSH(out_channels, out_channels)
+        self.ssh3 = SSH(out_channels, out_channels)
+
+        self.ClassHead = make_class_head(fpn_num=3, inchannels=cfg['out_channel'])
+        self.BboxHead = make_bbox_head(fpn_num=3, inchannels=cfg['out_channel'])
+        self.LandmarkHead = make_landmark_head(fpn_num=3, inchannels=cfg['out_channel'])
+
+        self.to(self.device)
+        self.eval()
+        if self.half_inference:
+            self.half()
+
+    def forward(self, inputs):
+        out = self.body(inputs)
+
+        if self.backbone == 'mobilenet0.25' or self.backbone == 'Resnet50':
+            out = list(out.values())
+        # FPN
+        fpn = self.fpn(out)
+
+        # SSH
+        feature1 = self.ssh1(fpn[0])
+        feature2 = self.ssh2(fpn[1])
+        feature3 = self.ssh3(fpn[2])
+        features = [feature1, feature2, feature3]
+
+        bbox_regressions = torch.cat([self.BboxHead[i](feature) for i, feature in enumerate(features)], dim=1)
+        classifications = torch.cat([self.ClassHead[i](feature) for i, feature in enumerate(features)], dim=1)
+        tmp = [self.LandmarkHead[i](feature) for i, feature in enumerate(features)]
+        ldm_regressions = (torch.cat(tmp, dim=1))
+
+        if self.phase == 'train':
+            output = (bbox_regressions, classifications, ldm_regressions)
+        else:
+            output = (bbox_regressions, F.softmax(classifications, dim=-1), ldm_regressions)
+        return output
+
+    def __detect_faces(self, inputs):
+        # get scale
+        height, width = inputs.shape[2:]
+        self.scale = torch.tensor([width, height, width, height], dtype=torch.float32, device=self.device)
+        tmp = [width, height, width, height, width, height, width, height, width, height]
+        self.scale1 = torch.tensor(tmp, dtype=torch.float32, device=self.device)
+
+        # forawrd
+        inputs = inputs.to(self.device)
+        if self.half_inference:
+            inputs = inputs.half()
+        loc, conf, landmarks = self(inputs)
+
+        # get priorbox
+        priorbox = PriorBox(self.cfg, image_size=inputs.shape[2:])
+        priors = priorbox.forward().to(self.device)
+
+        return loc, conf, landmarks, priors
+
+    # single image detection
+    def transform(self, image, use_origin_size):
+        # convert to opencv format
+        if isinstance(image, Image.Image):
+            image = cv2.cvtColor(np.asarray(image), cv2.COLOR_RGB2BGR)
+        image = image.astype(np.float32)
+
+        # testing scale
+        im_size_min = np.min(image.shape[0:2])
+        im_size_max = np.max(image.shape[0:2])
+        resize = float(self.target_size) / float(im_size_min)
+
+        # prevent bigger axis from being more than max_size
+        if np.round(resize * im_size_max) > self.max_size:
+            resize = float(self.max_size) / float(im_size_max)
+        resize = 1 if use_origin_size else resize
+
+        # resize
+        if resize != 1:
+            image = cv2.resize(image, None, None, fx=resize, fy=resize, interpolation=cv2.INTER_LINEAR)
+
+        # convert to torch.tensor format
+        # image -= (104, 117, 123)
+        image = image.transpose(2, 0, 1)
+        image = torch.from_numpy(image).unsqueeze(0)
+
+        return image, resize
+
+    def detect_faces(
+        self,
+        image,
+        conf_threshold=0.8,
+        nms_threshold=0.4,
+        use_origin_size=True,
+    ):
+        image, self.resize = self.transform(image, use_origin_size)
+        image = image.to(self.device)
+        if self.half_inference:
+            image = image.half()
+        image = image - self.mean_tensor
+
+        loc, conf, landmarks, priors = self.__detect_faces(image)
+
+        boxes = decode(loc.data.squeeze(0), priors.data, self.cfg['variance'])
+        boxes = boxes * self.scale / self.resize
+        boxes = boxes.cpu().numpy()
+
+        scores = conf.squeeze(0).data.cpu().numpy()[:, 1]
+
+        landmarks = decode_landm(landmarks.squeeze(0), priors, self.cfg['variance'])
+        landmarks = landmarks * self.scale1 / self.resize
+        landmarks = landmarks.cpu().numpy()
+
+        # ignore low scores
+        inds = np.where(scores > conf_threshold)[0]
+        boxes, landmarks, scores = boxes[inds], landmarks[inds], scores[inds]
+
+        # sort
+        order = scores.argsort()[::-1]
+        boxes, landmarks, scores = boxes[order], landmarks[order], scores[order]
+
+        # do NMS
+        bounding_boxes = np.hstack((boxes, scores[:, np.newaxis])).astype(np.float32, copy=False)
+        keep = py_cpu_nms(bounding_boxes, nms_threshold)
+        bounding_boxes, landmarks = bounding_boxes[keep, :], landmarks[keep]
+        # self.t['forward_pass'].toc()
+        # print(self.t['forward_pass'].average_time)
+        # import sys
+        # sys.stdout.flush()
+        return np.concatenate((bounding_boxes, landmarks), axis=1)
+
+    def __align_multi(self, image, boxes, landmarks, limit=None):
+
+        if len(boxes) < 1:
+            return [], []
+
+        if limit:
+            boxes = boxes[:limit]
+            landmarks = landmarks[:limit]
+
+        faces = []
+        for landmark in landmarks:
+            facial5points = [[landmark[2 * j], landmark[2 * j + 1]] for j in range(5)]
+
+            warped_face = warp_and_crop_face(np.array(image), facial5points, self.reference, crop_size=(112, 112))
+            faces.append(warped_face)
+
+        return np.concatenate((boxes, landmarks), axis=1), faces
+
+    def align_multi(self, img, conf_threshold=0.8, limit=None):
+
+        rlt = self.detect_faces(img, conf_threshold=conf_threshold)
+        boxes, landmarks = rlt[:, 0:5], rlt[:, 5:]
+
+        return self.__align_multi(img, boxes, landmarks, limit)
+
+    # batched detection
+    def batched_transform(self, frames, use_origin_size):
+        """
+        Arguments:
+            frames: a list of PIL.Image, or torch.Tensor(shape=[n, h, w, c],
+                type=np.float32, BGR format).
+            use_origin_size: whether to use origin size.
+        """
+        from_PIL = True if isinstance(frames[0], Image.Image) else False
+
+        # convert to opencv format
+        if from_PIL:
+            frames = [cv2.cvtColor(np.asarray(frame), cv2.COLOR_RGB2BGR) for frame in frames]
+            frames = np.asarray(frames, dtype=np.float32)
+
+        # testing scale
+        im_size_min = np.min(frames[0].shape[0:2])
+        im_size_max = np.max(frames[0].shape[0:2])
+        resize = float(self.target_size) / float(im_size_min)
+
+        # prevent bigger axis from being more than max_size
+        if np.round(resize * im_size_max) > self.max_size:
+            resize = float(self.max_size) / float(im_size_max)
+        resize = 1 if use_origin_size else resize
+
+        # resize
+        if resize != 1:
+            if not from_PIL:
+                frames = F.interpolate(frames, scale_factor=resize)
+            else:
+                frames = [
+                    cv2.resize(frame, None, None, fx=resize, fy=resize, interpolation=cv2.INTER_LINEAR)
+                    for frame in frames
+                ]
+
+        # convert to torch.tensor format
+        if not from_PIL:
+            frames = frames.transpose(1, 2).transpose(1, 3).contiguous()
+        else:
+            frames = frames.transpose((0, 3, 1, 2))
+            frames = torch.from_numpy(frames)
+
+        return frames, resize
+
+    def batched_detect_faces(self, frames, conf_threshold=0.8, nms_threshold=0.4, use_origin_size=True):
+        """
+        Arguments:
+            frames: a list of PIL.Image, or np.array(shape=[n, h, w, c],
+                type=np.uint8, BGR format).
+            conf_threshold: confidence threshold.
+            nms_threshold: nms threshold.
+            use_origin_size: whether to use origin size.
+        Returns:
+            final_bounding_boxes: list of np.array ([n_boxes, 5],
+                type=np.float32).
+            final_landmarks: list of np.array ([n_boxes, 10], type=np.float32).
+        """
+        # self.t['forward_pass'].tic()
+        frames, self.resize = self.batched_transform(frames, use_origin_size)
+        frames = frames.to(self.device)
+        frames = frames - self.mean_tensor
+
+        b_loc, b_conf, b_landmarks, priors = self.__detect_faces(frames)
+
+        final_bounding_boxes, final_landmarks = [], []
+
+        # decode
+        priors = priors.unsqueeze(0)
+        b_loc = batched_decode(b_loc, priors, self.cfg['variance']) * self.scale / self.resize
+        b_landmarks = batched_decode_landm(b_landmarks, priors, self.cfg['variance']) * self.scale1 / self.resize
+        b_conf = b_conf[:, :, 1]
+
+        # index for selection
+        b_indice = b_conf > conf_threshold
+
+        # concat
+        b_loc_and_conf = torch.cat((b_loc, b_conf.unsqueeze(-1)), dim=2).float()
+
+        for pred, landm, inds in zip(b_loc_and_conf, b_landmarks, b_indice):
+
+            # ignore low scores
+            pred, landm = pred[inds, :], landm[inds, :]
+            if pred.shape[0] == 0:
+                final_bounding_boxes.append(np.array([], dtype=np.float32))
+                final_landmarks.append(np.array([], dtype=np.float32))
+                continue
+
+            # sort
+            # order = score.argsort(descending=True)
+            # box, landm, score = box[order], landm[order], score[order]
+
+            # to CPU
+            bounding_boxes, landm = pred.cpu().numpy(), landm.cpu().numpy()
+
+            # NMS
+            keep = py_cpu_nms(bounding_boxes, nms_threshold)
+            bounding_boxes, landmarks = bounding_boxes[keep, :], landm[keep]
+
+            # append
+            final_bounding_boxes.append(bounding_boxes)
+            final_landmarks.append(landmarks)
+        # self.t['forward_pass'].toc(average=True)
+        # self.batch_time += self.t['forward_pass'].diff
+        # self.total_frame += len(frames)
+        # print(self.batch_time / self.total_frame)
+
+        return final_bounding_boxes, final_landmarks

extras/facexlib/detection/retinaface_net.py

@@ -0,0 +1,196 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+def conv_bn(inp, oup, stride=1, leaky=0):
+    return nn.Sequential(
+        nn.Conv2d(inp, oup, 3, stride, 1, bias=False), nn.BatchNorm2d(oup),
+        nn.LeakyReLU(negative_slope=leaky, inplace=True))
+
+
+def conv_bn_no_relu(inp, oup, stride):
+    return nn.Sequential(
+        nn.Conv2d(inp, oup, 3, stride, 1, bias=False),
+        nn.BatchNorm2d(oup),
+    )
+
+
+def conv_bn1X1(inp, oup, stride, leaky=0):
+    return nn.Sequential(
+        nn.Conv2d(inp, oup, 1, stride, padding=0, bias=False), nn.BatchNorm2d(oup),
+        nn.LeakyReLU(negative_slope=leaky, inplace=True))
+
+
+def conv_dw(inp, oup, stride, leaky=0.1):
+    return nn.Sequential(
+        nn.Conv2d(inp, inp, 3, stride, 1, groups=inp, bias=False),
+        nn.BatchNorm2d(inp),
+        nn.LeakyReLU(negative_slope=leaky, inplace=True),
+        nn.Conv2d(inp, oup, 1, 1, 0, bias=False),
+        nn.BatchNorm2d(oup),
+        nn.LeakyReLU(negative_slope=leaky, inplace=True),
+    )
+
+
+class SSH(nn.Module):
+
+    def __init__(self, in_channel, out_channel):
+        super(SSH, self).__init__()
+        assert out_channel % 4 == 0
+        leaky = 0
+        if (out_channel <= 64):
+            leaky = 0.1
+        self.conv3X3 = conv_bn_no_relu(in_channel, out_channel // 2, stride=1)
+
+        self.conv5X5_1 = conv_bn(in_channel, out_channel // 4, stride=1, leaky=leaky)
+        self.conv5X5_2 = conv_bn_no_relu(out_channel // 4, out_channel // 4, stride=1)
+
+        self.conv7X7_2 = conv_bn(out_channel // 4, out_channel // 4, stride=1, leaky=leaky)
+        self.conv7x7_3 = conv_bn_no_relu(out_channel // 4, out_channel // 4, stride=1)
+
+    def forward(self, input):
+        conv3X3 = self.conv3X3(input)
+
+        conv5X5_1 = self.conv5X5_1(input)
+        conv5X5 = self.conv5X5_2(conv5X5_1)
+
+        conv7X7_2 = self.conv7X7_2(conv5X5_1)
+        conv7X7 = self.conv7x7_3(conv7X7_2)
+
+        out = torch.cat([conv3X3, conv5X5, conv7X7], dim=1)
+        out = F.relu(out)
+        return out
+
+
+class FPN(nn.Module):
+
+    def __init__(self, in_channels_list, out_channels):
+        super(FPN, self).__init__()
+        leaky = 0
+        if (out_channels <= 64):
+            leaky = 0.1
+        self.output1 = conv_bn1X1(in_channels_list[0], out_channels, stride=1, leaky=leaky)
+        self.output2 = conv_bn1X1(in_channels_list[1], out_channels, stride=1, leaky=leaky)
+        self.output3 = conv_bn1X1(in_channels_list[2], out_channels, stride=1, leaky=leaky)
+
+        self.merge1 = conv_bn(out_channels, out_channels, leaky=leaky)
+        self.merge2 = conv_bn(out_channels, out_channels, leaky=leaky)
+
+    def forward(self, input):
+        # names = list(input.keys())
+        # input = list(input.values())
+
+        output1 = self.output1(input[0])
+        output2 = self.output2(input[1])
+        output3 = self.output3(input[2])
+
+        up3 = F.interpolate(output3, size=[output2.size(2), output2.size(3)], mode='nearest')
+        output2 = output2 + up3
+        output2 = self.merge2(output2)
+
+        up2 = F.interpolate(output2, size=[output1.size(2), output1.size(3)], mode='nearest')
+        output1 = output1 + up2
+        output1 = self.merge1(output1)
+
+        out = [output1, output2, output3]
+        return out
+
+
+class MobileNetV1(nn.Module):
+
+    def __init__(self):
+        super(MobileNetV1, self).__init__()
+        self.stage1 = nn.Sequential(
+            conv_bn(3, 8, 2, leaky=0.1),  # 3
+            conv_dw(8, 16, 1),  # 7
+            conv_dw(16, 32, 2),  # 11
+            conv_dw(32, 32, 1),  # 19
+            conv_dw(32, 64, 2),  # 27
+            conv_dw(64, 64, 1),  # 43
+        )
+        self.stage2 = nn.Sequential(
+            conv_dw(64, 128, 2),  # 43 + 16 = 59
+            conv_dw(128, 128, 1),  # 59 + 32 = 91
+            conv_dw(128, 128, 1),  # 91 + 32 = 123
+            conv_dw(128, 128, 1),  # 123 + 32 = 155
+            conv_dw(128, 128, 1),  # 155 + 32 = 187
+            conv_dw(128, 128, 1),  # 187 + 32 = 219
+        )
+        self.stage3 = nn.Sequential(
+            conv_dw(128, 256, 2),  # 219 +3 2 = 241
+            conv_dw(256, 256, 1),  # 241 + 64 = 301
+        )
+        self.avg = nn.AdaptiveAvgPool2d((1, 1))
+        self.fc = nn.Linear(256, 1000)
+
+    def forward(self, x):
+        x = self.stage1(x)
+        x = self.stage2(x)
+        x = self.stage3(x)
+        x = self.avg(x)
+        # x = self.model(x)
+        x = x.view(-1, 256)
+        x = self.fc(x)
+        return x
+
+
+class ClassHead(nn.Module):
+
+    def __init__(self, inchannels=512, num_anchors=3):
+        super(ClassHead, self).__init__()
+        self.num_anchors = num_anchors
+        self.conv1x1 = nn.Conv2d(inchannels, self.num_anchors * 2, kernel_size=(1, 1), stride=1, padding=0)
+
+    def forward(self, x):
+        out = self.conv1x1(x)
+        out = out.permute(0, 2, 3, 1).contiguous()
+
+        return out.view(out.shape[0], -1, 2)
+
+
+class BboxHead(nn.Module):
+
+    def __init__(self, inchannels=512, num_anchors=3):
+        super(BboxHead, self).__init__()
+        self.conv1x1 = nn.Conv2d(inchannels, num_anchors * 4, kernel_size=(1, 1), stride=1, padding=0)
+
+    def forward(self, x):
+        out = self.conv1x1(x)
+        out = out.permute(0, 2, 3, 1).contiguous()
+
+        return out.view(out.shape[0], -1, 4)
+
+
+class LandmarkHead(nn.Module):
+
+    def __init__(self, inchannels=512, num_anchors=3):
+        super(LandmarkHead, self).__init__()
+        self.conv1x1 = nn.Conv2d(inchannels, num_anchors * 10, kernel_size=(1, 1), stride=1, padding=0)
+
+    def forward(self, x):
+        out = self.conv1x1(x)
+        out = out.permute(0, 2, 3, 1).contiguous()
+
+        return out.view(out.shape[0], -1, 10)
+
+
+def make_class_head(fpn_num=3, inchannels=64, anchor_num=2):
+    classhead = nn.ModuleList()
+    for i in range(fpn_num):
+        classhead.append(ClassHead(inchannels, anchor_num))
+    return classhead
+
+
+def make_bbox_head(fpn_num=3, inchannels=64, anchor_num=2):
+    bboxhead = nn.ModuleList()
+    for i in range(fpn_num):
+        bboxhead.append(BboxHead(inchannels, anchor_num))
+    return bboxhead
+
+
+def make_landmark_head(fpn_num=3, inchannels=64, anchor_num=2):
+    landmarkhead = nn.ModuleList()
+    for i in range(fpn_num):
+        landmarkhead.append(LandmarkHead(inchannels, anchor_num))
+    return landmarkhead

extras/facexlib/detection/retinaface_utils.py

@@ -0,0 +1,421 @@
+import numpy as np
+import torch
+import torchvision
+from itertools import product as product
+from math import ceil
+
+
+class PriorBox(object):
+
+    def __init__(self, cfg, image_size=None, phase='train'):
+        super(PriorBox, self).__init__()
+        self.min_sizes = cfg['min_sizes']
+        self.steps = cfg['steps']
+        self.clip = cfg['clip']
+        self.image_size = image_size
+        self.feature_maps = [[ceil(self.image_size[0] / step), ceil(self.image_size[1] / step)] for step in self.steps]
+        self.name = 's'
+
+    def forward(self):
+        anchors = []
+        for k, f in enumerate(self.feature_maps):
+            min_sizes = self.min_sizes[k]
+            for i, j in product(range(f[0]), range(f[1])):
+                for min_size in min_sizes:
+                    s_kx = min_size / self.image_size[1]
+                    s_ky = min_size / self.image_size[0]
+                    dense_cx = [x * self.steps[k] / self.image_size[1] for x in [j + 0.5]]
+                    dense_cy = [y * self.steps[k] / self.image_size[0] for y in [i + 0.5]]
+                    for cy, cx in product(dense_cy, dense_cx):
+                        anchors += [cx, cy, s_kx, s_ky]
+
+        # back to torch land
+        output = torch.Tensor(anchors).view(-1, 4)
+        if self.clip:
+            output.clamp_(max=1, min=0)
+        return output
+
+
+def py_cpu_nms(dets, thresh):
+    """Pure Python NMS baseline."""
+    keep = torchvision.ops.nms(
+        boxes=torch.Tensor(dets[:, :4]),
+        scores=torch.Tensor(dets[:, 4]),
+        iou_threshold=thresh,
+    )
+
+    return list(keep)
+
+
+def point_form(boxes):
+    """ Convert prior_boxes to (xmin, ymin, xmax, ymax)
+    representation for comparison to point form ground truth data.
+    Args:
+        boxes: (tensor) center-size default boxes from priorbox layers.
+    Return:
+        boxes: (tensor) Converted xmin, ymin, xmax, ymax form of boxes.
+    """
+    return torch.cat(
+        (
+            boxes[:, :2] - boxes[:, 2:] / 2,  # xmin, ymin
+            boxes[:, :2] + boxes[:, 2:] / 2),
+        1)  # xmax, ymax
+
+
+def center_size(boxes):
+    """ Convert prior_boxes to (cx, cy, w, h)
+    representation for comparison to center-size form ground truth data.
+    Args:
+        boxes: (tensor) point_form boxes
+    Return:
+        boxes: (tensor) Converted xmin, ymin, xmax, ymax form of boxes.
+    """
+    return torch.cat(
+        (boxes[:, 2:] + boxes[:, :2]) / 2,  # cx, cy
+        boxes[:, 2:] - boxes[:, :2],
+        1)  # w, h
+
+
+def intersect(box_a, box_b):
+    """ We resize both tensors to [A,B,2] without new malloc:
+    [A,2] -> [A,1,2] -> [A,B,2]
+    [B,2] -> [1,B,2] -> [A,B,2]
+    Then we compute the area of intersect between box_a and box_b.
+    Args:
+      box_a: (tensor) bounding boxes, Shape: [A,4].
+      box_b: (tensor) bounding boxes, Shape: [B,4].
+    Return:
+      (tensor) intersection area, Shape: [A,B].
+    """
+    A = box_a.size(0)
+    B = box_b.size(0)
+    max_xy = torch.min(box_a[:, 2:].unsqueeze(1).expand(A, B, 2), box_b[:, 2:].unsqueeze(0).expand(A, B, 2))
+    min_xy = torch.max(box_a[:, :2].unsqueeze(1).expand(A, B, 2), box_b[:, :2].unsqueeze(0).expand(A, B, 2))
+    inter = torch.clamp((max_xy - min_xy), min=0)
+    return inter[:, :, 0] * inter[:, :, 1]
+
+
+def jaccard(box_a, box_b):
+    """Compute the jaccard overlap of two sets of boxes.  The jaccard overlap
+    is simply the intersection over union of two boxes.  Here we operate on
+    ground truth boxes and default boxes.
+    E.g.:
+        A ∩ B / A ∪ B = A ∩ B / (area(A) + area(B) - A ∩ B)
+    Args:
+        box_a: (tensor) Ground truth bounding boxes, Shape: [num_objects,4]
+        box_b: (tensor) Prior boxes from priorbox layers, Shape: [num_priors,4]
+    Return:
+        jaccard overlap: (tensor) Shape: [box_a.size(0), box_b.size(0)]
+    """
+    inter = intersect(box_a, box_b)
+    area_a = ((box_a[:, 2] - box_a[:, 0]) * (box_a[:, 3] - box_a[:, 1])).unsqueeze(1).expand_as(inter)  # [A,B]
+    area_b = ((box_b[:, 2] - box_b[:, 0]) * (box_b[:, 3] - box_b[:, 1])).unsqueeze(0).expand_as(inter)  # [A,B]
+    union = area_a + area_b - inter
+    return inter / union  # [A,B]
+
+
+def matrix_iou(a, b):
+    """
+    return iou of a and b, numpy version for data augenmentation
+    """
+    lt = np.maximum(a[:, np.newaxis, :2], b[:, :2])
+    rb = np.minimum(a[:, np.newaxis, 2:], b[:, 2:])
+
+    area_i = np.prod(rb - lt, axis=2) * (lt < rb).all(axis=2)
+    area_a = np.prod(a[:, 2:] - a[:, :2], axis=1)
+    area_b = np.prod(b[:, 2:] - b[:, :2], axis=1)
+    return area_i / (area_a[:, np.newaxis] + area_b - area_i)
+
+
+def matrix_iof(a, b):
+    """
+    return iof of a and b, numpy version for data augenmentation
+    """
+    lt = np.maximum(a[:, np.newaxis, :2], b[:, :2])
+    rb = np.minimum(a[:, np.newaxis, 2:], b[:, 2:])
+
+    area_i = np.prod(rb - lt, axis=2) * (lt < rb).all(axis=2)
+    area_a = np.prod(a[:, 2:] - a[:, :2], axis=1)
+    return area_i / np.maximum(area_a[:, np.newaxis], 1)
+
+
+def match(threshold, truths, priors, variances, labels, landms, loc_t, conf_t, landm_t, idx):
+    """Match each prior box with the ground truth box of the highest jaccard
+    overlap, encode the bounding boxes, then return the matched indices
+    corresponding to both confidence and location preds.
+    Args:
+        threshold: (float) The overlap threshold used when matching boxes.
+        truths: (tensor) Ground truth boxes, Shape: [num_obj, 4].
+        priors: (tensor) Prior boxes from priorbox layers, Shape: [n_priors,4].
+        variances: (tensor) Variances corresponding to each prior coord,
+            Shape: [num_priors, 4].
+        labels: (tensor) All the class labels for the image, Shape: [num_obj].
+        landms: (tensor) Ground truth landms, Shape [num_obj, 10].
+        loc_t: (tensor) Tensor to be filled w/ encoded location targets.
+        conf_t: (tensor) Tensor to be filled w/ matched indices for conf preds.
+        landm_t: (tensor) Tensor to be filled w/ encoded landm targets.
+        idx: (int) current batch index
+    Return:
+        The matched indices corresponding to 1)location 2)confidence
+        3)landm preds.
+    """
+    # jaccard index
+    overlaps = jaccard(truths, point_form(priors))
+    # (Bipartite Matching)
+    # [1,num_objects] best prior for each ground truth
+    best_prior_overlap, best_prior_idx = overlaps.max(1, keepdim=True)
+
+    # ignore hard gt
+    valid_gt_idx = best_prior_overlap[:, 0] >= 0.2
+    best_prior_idx_filter = best_prior_idx[valid_gt_idx, :]
+    if best_prior_idx_filter.shape[0] <= 0:
+        loc_t[idx] = 0
+        conf_t[idx] = 0
+        return
+
+    # [1,num_priors] best ground truth for each prior
+    best_truth_overlap, best_truth_idx = overlaps.max(0, keepdim=True)
+    best_truth_idx.squeeze_(0)
+    best_truth_overlap.squeeze_(0)
+    best_prior_idx.squeeze_(1)
+    best_prior_idx_filter.squeeze_(1)
+    best_prior_overlap.squeeze_(1)
+    best_truth_overlap.index_fill_(0, best_prior_idx_filter, 2)  # ensure best prior
+    # TODO refactor: index  best_prior_idx with long tensor
+    # ensure every gt matches with its prior of max overlap
+    for j in range(best_prior_idx.size(0)):  # 判别此anchor是预测哪一个boxes
+        best_truth_idx[best_prior_idx[j]] = j
+    matches = truths[best_truth_idx]  # Shape: [num_priors,4] 此处为每一个anchor对应的bbox取出来
+    conf = labels[best_truth_idx]  # Shape: [num_priors]      此处为每一个anchor对应的label取出来
+    conf[best_truth_overlap < threshold] = 0  # label as background   overlap<0.35的全部作为负样本
+    loc = encode(matches, priors, variances)
+
+    matches_landm = landms[best_truth_idx]
+    landm = encode_landm(matches_landm, priors, variances)
+    loc_t[idx] = loc  # [num_priors,4] encoded offsets to learn
+    conf_t[idx] = conf  # [num_priors] top class label for each prior
+    landm_t[idx] = landm
+
+
+def encode(matched, priors, variances):
+    """Encode the variances from the priorbox layers into the ground truth boxes
+    we have matched (based on jaccard overlap) with the prior boxes.
+    Args:
+        matched: (tensor) Coords of ground truth for each prior in point-form
+            Shape: [num_priors, 4].
+        priors: (tensor) Prior boxes in center-offset form
+            Shape: [num_priors,4].
+        variances: (list[float]) Variances of priorboxes
+    Return:
+        encoded boxes (tensor), Shape: [num_priors, 4]
+    """
+
+    # dist b/t match center and prior's center
+    g_cxcy = (matched[:, :2] + matched[:, 2:]) / 2 - priors[:, :2]
+    # encode variance
+    g_cxcy /= (variances[0] * priors[:, 2:])
+    # match wh / prior wh
+    g_wh = (matched[:, 2:] - matched[:, :2]) / priors[:, 2:]
+    g_wh = torch.log(g_wh) / variances[1]
+    # return target for smooth_l1_loss
+    return torch.cat([g_cxcy, g_wh], 1)  # [num_priors,4]
+
+
+def encode_landm(matched, priors, variances):
+    """Encode the variances from the priorbox layers into the ground truth boxes
+    we have matched (based on jaccard overlap) with the prior boxes.
+    Args:
+        matched: (tensor) Coords of ground truth for each prior in point-form
+            Shape: [num_priors, 10].
+        priors: (tensor) Prior boxes in center-offset form
+            Shape: [num_priors,4].
+        variances: (list[float]) Variances of priorboxes
+    Return:
+        encoded landm (tensor), Shape: [num_priors, 10]
+    """
+
+    # dist b/t match center and prior's center
+    matched = torch.reshape(matched, (matched.size(0), 5, 2))
+    priors_cx = priors[:, 0].unsqueeze(1).expand(matched.size(0), 5).unsqueeze(2)
+    priors_cy = priors[:, 1].unsqueeze(1).expand(matched.size(0), 5).unsqueeze(2)
+    priors_w = priors[:, 2].unsqueeze(1).expand(matched.size(0), 5).unsqueeze(2)
+    priors_h = priors[:, 3].unsqueeze(1).expand(matched.size(0), 5).unsqueeze(2)
+    priors = torch.cat([priors_cx, priors_cy, priors_w, priors_h], dim=2)
+    g_cxcy = matched[:, :, :2] - priors[:, :, :2]
+    # encode variance
+    g_cxcy /= (variances[0] * priors[:, :, 2:])
+    # g_cxcy /= priors[:, :, 2:]
+    g_cxcy = g_cxcy.reshape(g_cxcy.size(0), -1)
+    # return target for smooth_l1_loss
+    return g_cxcy
+
+
+# Adapted from https://github.com/Hakuyume/chainer-ssd
+def decode(loc, priors, variances):
+    """Decode locations from predictions using priors to undo
+    the encoding we did for offset regression at train time.
+    Args:
+        loc (tensor): location predictions for loc layers,
+            Shape: [num_priors,4]
+        priors (tensor): Prior boxes in center-offset form.
+            Shape: [num_priors,4].
+        variances: (list[float]) Variances of priorboxes
+    Return:
+        decoded bounding box predictions
+    """
+
+    boxes = torch.cat((priors[:, :2] + loc[:, :2] * variances[0] * priors[:, 2:],
+                       priors[:, 2:] * torch.exp(loc[:, 2:] * variances[1])), 1)
+    boxes[:, :2] -= boxes[:, 2:] / 2
+    boxes[:, 2:] += boxes[:, :2]
+    return boxes
+
+
+def decode_landm(pre, priors, variances):
+    """Decode landm from predictions using priors to undo
+    the encoding we did for offset regression at train time.
+    Args:
+        pre (tensor): landm predictions for loc layers,
+            Shape: [num_priors,10]
+        priors (tensor): Prior boxes in center-offset form.
+            Shape: [num_priors,4].
+        variances: (list[float]) Variances of priorboxes
+    Return:
+        decoded landm predictions
+    """
+    tmp = (
+        priors[:, :2] + pre[:, :2] * variances[0] * priors[:, 2:],
+        priors[:, :2] + pre[:, 2:4] * variances[0] * priors[:, 2:],
+        priors[:, :2] + pre[:, 4:6] * variances[0] * priors[:, 2:],
+        priors[:, :2] + pre[:, 6:8] * variances[0] * priors[:, 2:],
+        priors[:, :2] + pre[:, 8:10] * variances[0] * priors[:, 2:],
+    )
+    landms = torch.cat(tmp, dim=1)
+    return landms
+
+
+def batched_decode(b_loc, priors, variances):
+    """Decode locations from predictions using priors to undo
+    the encoding we did for offset regression at train time.
+    Args:
+        b_loc (tensor): location predictions for loc layers,
+            Shape: [num_batches,num_priors,4]
+        priors (tensor): Prior boxes in center-offset form.
+            Shape: [1,num_priors,4].
+        variances: (list[float]) Variances of priorboxes
+    Return:
+        decoded bounding box predictions
+    """
+    boxes = (
+        priors[:, :, :2] + b_loc[:, :, :2] * variances[0] * priors[:, :, 2:],
+        priors[:, :, 2:] * torch.exp(b_loc[:, :, 2:] * variances[1]),
+    )
+    boxes = torch.cat(boxes, dim=2)
+
+    boxes[:, :, :2] -= boxes[:, :, 2:] / 2
+    boxes[:, :, 2:] += boxes[:, :, :2]
+    return boxes
+
+
+def batched_decode_landm(pre, priors, variances):
+    """Decode landm from predictions using priors to undo
+    the encoding we did for offset regression at train time.
+    Args:
+        pre (tensor): landm predictions for loc layers,
+            Shape: [num_batches,num_priors,10]
+        priors (tensor): Prior boxes in center-offset form.
+            Shape: [1,num_priors,4].
+        variances: (list[float]) Variances of priorboxes
+    Return:
+        decoded landm predictions
+    """
+    landms = (
+        priors[:, :, :2] + pre[:, :, :2] * variances[0] * priors[:, :, 2:],
+        priors[:, :, :2] + pre[:, :, 2:4] * variances[0] * priors[:, :, 2:],
+        priors[:, :, :2] + pre[:, :, 4:6] * variances[0] * priors[:, :, 2:],
+        priors[:, :, :2] + pre[:, :, 6:8] * variances[0] * priors[:, :, 2:],
+        priors[:, :, :2] + pre[:, :, 8:10] * variances[0] * priors[:, :, 2:],
+    )
+    landms = torch.cat(landms, dim=2)
+    return landms
+
+
+def log_sum_exp(x):
+    """Utility function for computing log_sum_exp while determining
+    This will be used to determine unaveraged confidence loss across
+    all examples in a batch.
+    Args:
+        x (Variable(tensor)): conf_preds from conf layers
+    """
+    x_max = x.data.max()
+    return torch.log(torch.sum(torch.exp(x - x_max), 1, keepdim=True)) + x_max
+
+
+# Original author: Francisco Massa:
+# https://github.com/fmassa/object-detection.torch
+# Ported to PyTorch by Max deGroot (02/01/2017)
+def nms(boxes, scores, overlap=0.5, top_k=200):
+    """Apply non-maximum suppression at test time to avoid detecting too many
+    overlapping bounding boxes for a given object.
+    Args:
+        boxes: (tensor) The location preds for the img, Shape: [num_priors,4].
+        scores: (tensor) The class predscores for the img, Shape:[num_priors].
+        overlap: (float) The overlap thresh for suppressing unnecessary boxes.
+        top_k: (int) The Maximum number of box preds to consider.
+    Return:
+        The indices of the kept boxes with respect to num_priors.
+    """
+
+    keep = torch.Tensor(scores.size(0)).fill_(0).long()
+    if boxes.numel() == 0:
+        return keep
+    x1 = boxes[:, 0]
+    y1 = boxes[:, 1]
+    x2 = boxes[:, 2]
+    y2 = boxes[:, 3]
+    area = torch.mul(x2 - x1, y2 - y1)
+    v, idx = scores.sort(0)  # sort in ascending order
+    # I = I[v >= 0.01]
+    idx = idx[-top_k:]  # indices of the top-k largest vals
+    xx1 = boxes.new()
+    yy1 = boxes.new()
+    xx2 = boxes.new()
+    yy2 = boxes.new()
+    w = boxes.new()
+    h = boxes.new()
+
+    # keep = torch.Tensor()
+    count = 0
+    while idx.numel() > 0:
+        i = idx[-1]  # index of current largest val
+        # keep.append(i)
+        keep[count] = i
+        count += 1
+        if idx.size(0) == 1:
+            break
+        idx = idx[:-1]  # remove kept element from view
+        # load bboxes of next highest vals
+        torch.index_select(x1, 0, idx, out=xx1)
+        torch.index_select(y1, 0, idx, out=yy1)
+        torch.index_select(x2, 0, idx, out=xx2)
+        torch.index_select(y2, 0, idx, out=yy2)
+        # store element-wise max with next highest score
+        xx1 = torch.clamp(xx1, min=x1[i])
+        yy1 = torch.clamp(yy1, min=y1[i])
+        xx2 = torch.clamp(xx2, max=x2[i])
+        yy2 = torch.clamp(yy2, max=y2[i])
+        w.resize_as_(xx2)
+        h.resize_as_(yy2)
+        w = xx2 - xx1
+        h = yy2 - yy1
+        # check sizes of xx1 and xx2.. after each iteration
+        w = torch.clamp(w, min=0.0)
+        h = torch.clamp(h, min=0.0)
+        inter = w * h
+        # IoU = i / (area(a) + area(b) - i)
+        rem_areas = torch.index_select(area, 0, idx)  # load remaining areas)
+        union = (rem_areas - inter) + area[i]
+        IoU = inter / union  # store result in iou
+        # keep only elements with an IoU <= overlap
+        idx = idx[IoU.le(overlap)]
+    return keep, count

extras/facexlib/parsing/__init__.py

@@ -0,0 +1,24 @@
+import torch
+
+from extras.facexlib.utils import load_file_from_url
+from .bisenet import BiSeNet
+from .parsenet import ParseNet
+
+
+def init_parsing_model(model_name='bisenet', half=False, device='cuda', model_rootpath=None):
+    if model_name == 'bisenet':
+        model = BiSeNet(num_class=19)
+        model_url = 'https://github.com/xinntao/facexlib/releases/download/v0.2.0/parsing_bisenet.pth'
+    elif model_name == 'parsenet':
+        model = ParseNet(in_size=512, out_size=512, parsing_ch=19)
+        model_url = 'https://github.com/xinntao/facexlib/releases/download/v0.2.2/parsing_parsenet.pth'
+    else:
+        raise NotImplementedError(f'{model_name} is not implemented.')
+
+    model_path = load_file_from_url(
+        url=model_url, model_dir='facexlib/weights', progress=True, file_name=None, save_dir=model_rootpath)
+    load_net = torch.load(model_path, map_location=lambda storage, loc: storage)
+    model.load_state_dict(load_net, strict=True)
+    model.eval()
+    model = model.to(device)
+    return model

extras/facexlib/parsing/bisenet.py

@@ -0,0 +1,140 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from .resnet import ResNet18
+
+
+class ConvBNReLU(nn.Module):
+
+    def __init__(self, in_chan, out_chan, ks=3, stride=1, padding=1):
+        super(ConvBNReLU, self).__init__()
+        self.conv = nn.Conv2d(in_chan, out_chan, kernel_size=ks, stride=stride, padding=padding, bias=False)
+        self.bn = nn.BatchNorm2d(out_chan)
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = F.relu(self.bn(x))
+        return x
+
+
+class BiSeNetOutput(nn.Module):
+
+    def __init__(self, in_chan, mid_chan, num_class):
+        super(BiSeNetOutput, self).__init__()
+        self.conv = ConvBNReLU(in_chan, mid_chan, ks=3, stride=1, padding=1)
+        self.conv_out = nn.Conv2d(mid_chan, num_class, kernel_size=1, bias=False)
+
+    def forward(self, x):
+        feat = self.conv(x)
+        out = self.conv_out(feat)
+        return out, feat
+
+
+class AttentionRefinementModule(nn.Module):
+
+    def __init__(self, in_chan, out_chan):
+        super(AttentionRefinementModule, self).__init__()
+        self.conv = ConvBNReLU(in_chan, out_chan, ks=3, stride=1, padding=1)
+        self.conv_atten = nn.Conv2d(out_chan, out_chan, kernel_size=1, bias=False)
+        self.bn_atten = nn.BatchNorm2d(out_chan)
+        self.sigmoid_atten = nn.Sigmoid()
+
+    def forward(self, x):
+        feat = self.conv(x)
+        atten = F.avg_pool2d(feat, feat.size()[2:])
+        atten = self.conv_atten(atten)
+        atten = self.bn_atten(atten)
+        atten = self.sigmoid_atten(atten)
+        out = torch.mul(feat, atten)
+        return out
+
+
+class ContextPath(nn.Module):
+
+    def __init__(self):
+        super(ContextPath, self).__init__()
+        self.resnet = ResNet18()
+        self.arm16 = AttentionRefinementModule(256, 128)
+        self.arm32 = AttentionRefinementModule(512, 128)
+        self.conv_head32 = ConvBNReLU(128, 128, ks=3, stride=1, padding=1)
+        self.conv_head16 = ConvBNReLU(128, 128, ks=3, stride=1, padding=1)
+        self.conv_avg = ConvBNReLU(512, 128, ks=1, stride=1, padding=0)
+
+    def forward(self, x):
+        feat8, feat16, feat32 = self.resnet(x)
+        h8, w8 = feat8.size()[2:]
+        h16, w16 = feat16.size()[2:]
+        h32, w32 = feat32.size()[2:]
+
+        avg = F.avg_pool2d(feat32, feat32.size()[2:])
+        avg = self.conv_avg(avg)
+        avg_up = F.interpolate(avg, (h32, w32), mode='nearest')
+
+        feat32_arm = self.arm32(feat32)
+        feat32_sum = feat32_arm + avg_up
+        feat32_up = F.interpolate(feat32_sum, (h16, w16), mode='nearest')
+        feat32_up = self.conv_head32(feat32_up)
+
+        feat16_arm = self.arm16(feat16)
+        feat16_sum = feat16_arm + feat32_up
+        feat16_up = F.interpolate(feat16_sum, (h8, w8), mode='nearest')
+        feat16_up = self.conv_head16(feat16_up)
+
+        return feat8, feat16_up, feat32_up  # x8, x8, x16
+
+
+class FeatureFusionModule(nn.Module):
+
+    def __init__(self, in_chan, out_chan):
+        super(FeatureFusionModule, self).__init__()
+        self.convblk = ConvBNReLU(in_chan, out_chan, ks=1, stride=1, padding=0)
+        self.conv1 = nn.Conv2d(out_chan, out_chan // 4, kernel_size=1, stride=1, padding=0, bias=False)
+        self.conv2 = nn.Conv2d(out_chan // 4, out_chan, kernel_size=1, stride=1, padding=0, bias=False)
+        self.relu = nn.ReLU(inplace=True)
+        self.sigmoid = nn.Sigmoid()
+
+    def forward(self, fsp, fcp):
+        fcat = torch.cat([fsp, fcp], dim=1)
+        feat = self.convblk(fcat)
+        atten = F.avg_pool2d(feat, feat.size()[2:])
+        atten = self.conv1(atten)
+        atten = self.relu(atten)
+        atten = self.conv2(atten)
+        atten = self.sigmoid(atten)
+        feat_atten = torch.mul(feat, atten)
+        feat_out = feat_atten + feat
+        return feat_out
+
+
+class BiSeNet(nn.Module):
+
+    def __init__(self, num_class):
+        super(BiSeNet, self).__init__()
+        self.cp = ContextPath()
+        self.ffm = FeatureFusionModule(256, 256)
+        self.conv_out = BiSeNetOutput(256, 256, num_class)
+        self.conv_out16 = BiSeNetOutput(128, 64, num_class)
+        self.conv_out32 = BiSeNetOutput(128, 64, num_class)
+
+    def forward(self, x, return_feat=False):
+        h, w = x.size()[2:]
+        feat_res8, feat_cp8, feat_cp16 = self.cp(x)  # return res3b1 feature
+        feat_sp = feat_res8  # replace spatial path feature with res3b1 feature
+        feat_fuse = self.ffm(feat_sp, feat_cp8)
+
+        out, feat = self.conv_out(feat_fuse)
+        out16, feat16 = self.conv_out16(feat_cp8)
+        out32, feat32 = self.conv_out32(feat_cp16)
+
+        out = F.interpolate(out, (h, w), mode='bilinear', align_corners=True)
+        out16 = F.interpolate(out16, (h, w), mode='bilinear', align_corners=True)
+        out32 = F.interpolate(out32, (h, w), mode='bilinear', align_corners=True)
+
+        if return_feat:
+            feat = F.interpolate(feat, (h, w), mode='bilinear', align_corners=True)
+            feat16 = F.interpolate(feat16, (h, w), mode='bilinear', align_corners=True)
+            feat32 = F.interpolate(feat32, (h, w), mode='bilinear', align_corners=True)
+            return out, out16, out32, feat, feat16, feat32
+        else:
+            return out, out16, out32

extras/facexlib/parsing/parsenet.py

@@ -0,0 +1,194 @@
+"""Modified from https://github.com/chaofengc/PSFRGAN
+"""
+import numpy as np
+import torch.nn as nn
+from torch.nn import functional as F
+
+
+class NormLayer(nn.Module):
+    """Normalization Layers.
+
+    Args:
+        channels: input channels, for batch norm and instance norm.
+        input_size: input shape without batch size, for layer norm.
+    """
+
+    def __init__(self, channels, normalize_shape=None, norm_type='bn'):
+        super(NormLayer, self).__init__()
+        norm_type = norm_type.lower()
+        self.norm_type = norm_type
+        if norm_type == 'bn':
+            self.norm = nn.BatchNorm2d(channels, affine=True)
+        elif norm_type == 'in':
+            self.norm = nn.InstanceNorm2d(channels, affine=False)
+        elif norm_type == 'gn':
+            self.norm = nn.GroupNorm(32, channels, affine=True)
+        elif norm_type == 'pixel':
+            self.norm = lambda x: F.normalize(x, p=2, dim=1)
+        elif norm_type == 'layer':
+            self.norm = nn.LayerNorm(normalize_shape)
+        elif norm_type == 'none':
+            self.norm = lambda x: x * 1.0
+        else:
+            assert 1 == 0, f'Norm type {norm_type} not support.'
+
+    def forward(self, x, ref=None):
+        if self.norm_type == 'spade':
+            return self.norm(x, ref)
+        else:
+            return self.norm(x)
+
+
+class ReluLayer(nn.Module):
+    """Relu Layer.
+
+    Args:
+        relu type: type of relu layer, candidates are
+            - ReLU
+            - LeakyReLU: default relu slope 0.2
+            - PRelu
+            - SELU
+            - none: direct pass
+    """
+
+    def __init__(self, channels, relu_type='relu'):
+        super(ReluLayer, self).__init__()
+        relu_type = relu_type.lower()
+        if relu_type == 'relu':
+            self.func = nn.ReLU(True)
+        elif relu_type == 'leakyrelu':
+            self.func = nn.LeakyReLU(0.2, inplace=True)
+        elif relu_type == 'prelu':
+            self.func = nn.PReLU(channels)
+        elif relu_type == 'selu':
+            self.func = nn.SELU(True)
+        elif relu_type == 'none':
+            self.func = lambda x: x * 1.0
+        else:
+            assert 1 == 0, f'Relu type {relu_type} not support.'
+
+    def forward(self, x):
+        return self.func(x)
+
+
+class ConvLayer(nn.Module):
+
+    def __init__(self,
+                 in_channels,
+                 out_channels,
+                 kernel_size=3,
+                 scale='none',
+                 norm_type='none',
+                 relu_type='none',
+                 use_pad=True,
+                 bias=True):
+        super(ConvLayer, self).__init__()
+        self.use_pad = use_pad
+        self.norm_type = norm_type
+        if norm_type in ['bn']:
+            bias = False
+
+        stride = 2 if scale == 'down' else 1
+
+        self.scale_func = lambda x: x
+        if scale == 'up':
+            self.scale_func = lambda x: nn.functional.interpolate(x, scale_factor=2, mode='nearest')
+
+        self.reflection_pad = nn.ReflectionPad2d(int(np.ceil((kernel_size - 1.) / 2)))
+        self.conv2d = nn.Conv2d(in_channels, out_channels, kernel_size, stride, bias=bias)
+
+        self.relu = ReluLayer(out_channels, relu_type)
+        self.norm = NormLayer(out_channels, norm_type=norm_type)
+
+    def forward(self, x):
+        out = self.scale_func(x)
+        if self.use_pad:
+            out = self.reflection_pad(out)
+        out = self.conv2d(out)
+        out = self.norm(out)
+        out = self.relu(out)
+        return out
+
+
+class ResidualBlock(nn.Module):
+    """
+    Residual block recommended in: http://torch.ch/blog/2016/02/04/resnets.html
+    """
+
+    def __init__(self, c_in, c_out, relu_type='prelu', norm_type='bn', scale='none'):
+        super(ResidualBlock, self).__init__()
+
+        if scale == 'none' and c_in == c_out:
+            self.shortcut_func = lambda x: x
+        else:
+            self.shortcut_func = ConvLayer(c_in, c_out, 3, scale)
+
+        scale_config_dict = {'down': ['none', 'down'], 'up': ['up', 'none'], 'none': ['none', 'none']}
+        scale_conf = scale_config_dict[scale]
+
+        self.conv1 = ConvLayer(c_in, c_out, 3, scale_conf[0], norm_type=norm_type, relu_type=relu_type)
+        self.conv2 = ConvLayer(c_out, c_out, 3, scale_conf[1], norm_type=norm_type, relu_type='none')
+
+    def forward(self, x):
+        identity = self.shortcut_func(x)
+
+        res = self.conv1(x)
+        res = self.conv2(res)
+        return identity + res
+
+
+class ParseNet(nn.Module):
+
+    def __init__(self,
+                 in_size=128,
+                 out_size=128,
+                 min_feat_size=32,
+                 base_ch=64,
+                 parsing_ch=19,
+                 res_depth=10,
+                 relu_type='LeakyReLU',
+                 norm_type='bn',
+                 ch_range=[32, 256]):
+        super().__init__()
+        self.res_depth = res_depth
+        act_args = {'norm_type': norm_type, 'relu_type': relu_type}
+        min_ch, max_ch = ch_range
+
+        ch_clip = lambda x: max(min_ch, min(x, max_ch))  # noqa: E731
+        min_feat_size = min(in_size, min_feat_size)
+
+        down_steps = int(np.log2(in_size // min_feat_size))
+        up_steps = int(np.log2(out_size // min_feat_size))
+
+        # =============== define encoder-body-decoder ====================
+        self.encoder = []
+        self.encoder.append(ConvLayer(3, base_ch, 3, 1))
+        head_ch = base_ch
+        for i in range(down_steps):
+            cin, cout = ch_clip(head_ch), ch_clip(head_ch * 2)
+            self.encoder.append(ResidualBlock(cin, cout, scale='down', **act_args))
+            head_ch = head_ch * 2
+
+        self.body = []
+        for i in range(res_depth):
+            self.body.append(ResidualBlock(ch_clip(head_ch), ch_clip(head_ch), **act_args))
+
+        self.decoder = []
+        for i in range(up_steps):
+            cin, cout = ch_clip(head_ch), ch_clip(head_ch // 2)
+            self.decoder.append(ResidualBlock(cin, cout, scale='up', **act_args))
+            head_ch = head_ch // 2
+
+        self.encoder = nn.Sequential(*self.encoder)
+        self.body = nn.Sequential(*self.body)
+        self.decoder = nn.Sequential(*self.decoder)
+        self.out_img_conv = ConvLayer(ch_clip(head_ch), 3)
+        self.out_mask_conv = ConvLayer(ch_clip(head_ch), parsing_ch)
+
+    def forward(self, x):
+        feat = self.encoder(x)
+        x = feat + self.body(feat)
+        x = self.decoder(x)
+        out_img = self.out_img_conv(x)
+        out_mask = self.out_mask_conv(x)
+        return out_mask, out_img

extras/facexlib/parsing/resnet.py

@@ -0,0 +1,69 @@
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+def conv3x3(in_planes, out_planes, stride=1):
+    """3x3 convolution with padding"""
+    return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride, padding=1, bias=False)
+
+
+class BasicBlock(nn.Module):
+
+    def __init__(self, in_chan, out_chan, stride=1):
+        super(BasicBlock, self).__init__()
+        self.conv1 = conv3x3(in_chan, out_chan, stride)
+        self.bn1 = nn.BatchNorm2d(out_chan)
+        self.conv2 = conv3x3(out_chan, out_chan)
+        self.bn2 = nn.BatchNorm2d(out_chan)
+        self.relu = nn.ReLU(inplace=True)
+        self.downsample = None
+        if in_chan != out_chan or stride != 1:
+            self.downsample = nn.Sequential(
+                nn.Conv2d(in_chan, out_chan, kernel_size=1, stride=stride, bias=False),
+                nn.BatchNorm2d(out_chan),
+            )
+
+    def forward(self, x):
+        residual = self.conv1(x)
+        residual = F.relu(self.bn1(residual))
+        residual = self.conv2(residual)
+        residual = self.bn2(residual)
+
+        shortcut = x
+        if self.downsample is not None:
+            shortcut = self.downsample(x)
+
+        out = shortcut + residual
+        out = self.relu(out)
+        return out
+
+
+def create_layer_basic(in_chan, out_chan, bnum, stride=1):
+    layers = [BasicBlock(in_chan, out_chan, stride=stride)]
+    for i in range(bnum - 1):
+        layers.append(BasicBlock(out_chan, out_chan, stride=1))
+    return nn.Sequential(*layers)
+
+
+class ResNet18(nn.Module):
+
+    def __init__(self):
+        super(ResNet18, self).__init__()
+        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False)
+        self.bn1 = nn.BatchNorm2d(64)
+        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
+        self.layer1 = create_layer_basic(64, 64, bnum=2, stride=1)
+        self.layer2 = create_layer_basic(64, 128, bnum=2, stride=2)
+        self.layer3 = create_layer_basic(128, 256, bnum=2, stride=2)
+        self.layer4 = create_layer_basic(256, 512, bnum=2, stride=2)
+
+    def forward(self, x):
+        x = self.conv1(x)
+        x = F.relu(self.bn1(x))
+        x = self.maxpool(x)
+
+        x = self.layer1(x)
+        feat8 = self.layer2(x)  # 1/8
+        feat16 = self.layer3(feat8)  # 1/16
+        feat32 = self.layer4(feat16)  # 1/32
+        return feat8, feat16, feat32

extras/facexlib/utils/__init__.py

@@ -0,0 +1,7 @@
+from .face_utils import align_crop_face_landmarks, compute_increased_bbox, get_valid_bboxes, paste_face_back
+from .misc import img2tensor, load_file_from_url, scandir
+
+__all__ = [
+    'align_crop_face_landmarks', 'compute_increased_bbox', 'get_valid_bboxes', 'load_file_from_url', 'paste_face_back',
+    'img2tensor', 'scandir'
+]

extras/facexlib/utils/face_restoration_helper.py

@@ -0,0 +1,374 @@
+import cv2
+import numpy as np
+import os
+import torch
+from torchvision.transforms.functional import normalize
+
+from extras.facexlib.detection import init_detection_model
+from extras.facexlib.parsing import init_parsing_model
+from extras.facexlib.utils.misc import img2tensor, imwrite
+
+
+def get_largest_face(det_faces, h, w):
+
+    def get_location(val, length):
+        if val < 0:
+            return 0
+        elif val > length:
+            return length
+        else:
+            return val
+
+    face_areas = []
+    for det_face in det_faces:
+        left = get_location(det_face[0], w)
+        right = get_location(det_face[2], w)
+        top = get_location(det_face[1], h)
+        bottom = get_location(det_face[3], h)
+        face_area = (right - left) * (bottom - top)
+        face_areas.append(face_area)
+    largest_idx = face_areas.index(max(face_areas))
+    return det_faces[largest_idx], largest_idx
+
+
+def get_center_face(det_faces, h=0, w=0, center=None):
+    if center is not None:
+        center = np.array(center)
+    else:
+        center = np.array([w / 2, h / 2])
+    center_dist = []
+    for det_face in det_faces:
+        face_center = np.array([(det_face[0] + det_face[2]) / 2, (det_face[1] + det_face[3]) / 2])
+        dist = np.linalg.norm(face_center - center)
+        center_dist.append(dist)
+    center_idx = center_dist.index(min(center_dist))
+    return det_faces[center_idx], center_idx
+
+
+class FaceRestoreHelper(object):
+    """Helper for the face restoration pipeline (base class)."""
+
+    def __init__(self,
+                 upscale_factor,
+                 face_size=512,
+                 crop_ratio=(1, 1),
+                 det_model='retinaface_resnet50',
+                 save_ext='png',
+                 template_3points=False,
+                 pad_blur=False,
+                 use_parse=False,
+                 device=None,
+                 model_rootpath=None):
+        self.template_3points = template_3points  # improve robustness
+        self.upscale_factor = upscale_factor
+        # the cropped face ratio based on the square face
+        self.crop_ratio = crop_ratio  # (h, w)
+        assert (self.crop_ratio[0] >= 1 and self.crop_ratio[1] >= 1), 'crop ration only supports >=1'
+        self.face_size = (int(face_size * self.crop_ratio[1]), int(face_size * self.crop_ratio[0]))
+
+        if self.template_3points:
+            self.face_template = np.array([[192, 240], [319, 240], [257, 371]])
+        else:
+            # standard 5 landmarks for FFHQ faces with 512 x 512
+            self.face_template = np.array([[192.98138, 239.94708], [318.90277, 240.1936], [256.63416, 314.01935],
+                                           [201.26117, 371.41043], [313.08905, 371.15118]])
+        self.face_template = self.face_template * (face_size / 512.0)
+        if self.crop_ratio[0] > 1:
+            self.face_template[:, 1] += face_size * (self.crop_ratio[0] - 1) / 2
+        if self.crop_ratio[1] > 1:
+            self.face_template[:, 0] += face_size * (self.crop_ratio[1] - 1) / 2
+        self.save_ext = save_ext
+        self.pad_blur = pad_blur
+        if self.pad_blur is True:
+            self.template_3points = False
+
+        self.all_landmarks_5 = []
+        self.det_faces = []
+        self.affine_matrices = []
+        self.inverse_affine_matrices = []
+        self.cropped_faces = []
+        self.restored_faces = []
+        self.pad_input_imgs = []
+
+        if device is None:
+            self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+        else:
+            self.device = device
+
+        # init face detection model
+        self.face_det = init_detection_model(det_model, half=False, device=self.device, model_rootpath=model_rootpath)
+
+        # init face parsing model
+        self.use_parse = use_parse
+        self.face_parse = init_parsing_model(model_name='parsenet', device=self.device, model_rootpath=model_rootpath)
+
+    def set_upscale_factor(self, upscale_factor):
+        self.upscale_factor = upscale_factor
+
+    def read_image(self, img):
+        """img can be image path or cv2 loaded image."""
+        # self.input_img is Numpy array, (h, w, c), BGR, uint8, [0, 255]
+        if isinstance(img, str):
+            img = cv2.imread(img)
+
+        if np.max(img) > 256:  # 16-bit image
+            img = img / 65535 * 255
+        if len(img.shape) == 2:  # gray image
+            img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
+        elif img.shape[2] == 4:  # RGBA image with alpha channel
+            img = img[:, :, 0:3]
+
+        self.input_img = img
+
+    def get_face_landmarks_5(self,
+                             only_keep_largest=False,
+                             only_center_face=False,
+                             resize=None,
+                             blur_ratio=0.01,
+                             eye_dist_threshold=None):
+        if resize is None:
+            scale = 1
+            input_img = self.input_img
+        else:
+            h, w = self.input_img.shape[0:2]
+            scale = min(h, w) / resize
+            h, w = int(h / scale), int(w / scale)
+            input_img = cv2.resize(self.input_img, (w, h), interpolation=cv2.INTER_LANCZOS4)
+
+        with torch.no_grad():
+            bboxes = self.face_det.detect_faces(input_img, 0.97) * scale
+        for bbox in bboxes:
+            # remove faces with too small eye distance: side faces or too small faces
+            eye_dist = np.linalg.norm([bbox[5] - bbox[7], bbox[6] - bbox[8]])
+            if eye_dist_threshold is not None and (eye_dist < eye_dist_threshold):
+                continue
+
+            if self.template_3points:
+                landmark = np.array([[bbox[i], bbox[i + 1]] for i in range(5, 11, 2)])
+            else:
+                landmark = np.array([[bbox[i], bbox[i + 1]] for i in range(5, 15, 2)])
+            self.all_landmarks_5.append(landmark)
+            self.det_faces.append(bbox[0:5])
+        if len(self.det_faces) == 0:
+            return 0
+        if only_keep_largest:
+            h, w, _ = self.input_img.shape
+            self.det_faces, largest_idx = get_largest_face(self.det_faces, h, w)
+            self.all_landmarks_5 = [self.all_landmarks_5[largest_idx]]
+        elif only_center_face:
+            h, w, _ = self.input_img.shape
+            self.det_faces, center_idx = get_center_face(self.det_faces, h, w)
+            self.all_landmarks_5 = [self.all_landmarks_5[center_idx]]
+
+        # pad blurry images
+        if self.pad_blur:
+            self.pad_input_imgs = []
+            for landmarks in self.all_landmarks_5:
+                # get landmarks
+                eye_left = landmarks[0, :]
+                eye_right = landmarks[1, :]
+                eye_avg = (eye_left + eye_right) * 0.5
+                mouth_avg = (landmarks[3, :] + landmarks[4, :]) * 0.5
+                eye_to_eye = eye_right - eye_left
+                eye_to_mouth = mouth_avg - eye_avg
+
+                # Get the oriented crop rectangle
+                # x: half width of the oriented crop rectangle
+                x = eye_to_eye - np.flipud(eye_to_mouth) * [-1, 1]
+                #  - np.flipud(eye_to_mouth) * [-1, 1]: rotate 90 clockwise
+                # norm with the hypotenuse: get the direction
+                x /= np.hypot(*x)  # get the hypotenuse of a right triangle
+                rect_scale = 1.5
+                x *= max(np.hypot(*eye_to_eye) * 2.0 * rect_scale, np.hypot(*eye_to_mouth) * 1.8 * rect_scale)
+                # y: half height of the oriented crop rectangle
+                y = np.flipud(x) * [-1, 1]
+
+                # c: center
+                c = eye_avg + eye_to_mouth * 0.1
+                # quad: (left_top, left_bottom, right_bottom, right_top)
+                quad = np.stack([c - x - y, c - x + y, c + x + y, c + x - y])
+                # qsize: side length of the square
+                qsize = np.hypot(*x) * 2
+                border = max(int(np.rint(qsize * 0.1)), 3)
+
+                # get pad
+                # pad: (width_left, height_top, width_right, height_bottom)
+                pad = (int(np.floor(min(quad[:, 0]))), int(np.floor(min(quad[:, 1]))), int(np.ceil(max(quad[:, 0]))),
+                       int(np.ceil(max(quad[:, 1]))))
+                pad = [
+                    max(-pad[0] + border, 1),
+                    max(-pad[1] + border, 1),
+                    max(pad[2] - self.input_img.shape[0] + border, 1),
+                    max(pad[3] - self.input_img.shape[1] + border, 1)
+                ]
+
+                if max(pad) > 1:
+                    # pad image
+                    pad_img = np.pad(self.input_img, ((pad[1], pad[3]), (pad[0], pad[2]), (0, 0)), 'reflect')
+                    # modify landmark coords
+                    landmarks[:, 0] += pad[0]
+                    landmarks[:, 1] += pad[1]
+                    # blur pad images
+                    h, w, _ = pad_img.shape
+                    y, x, _ = np.ogrid[:h, :w, :1]
+                    mask = np.maximum(1.0 - np.minimum(np.float32(x) / pad[0],
+                                                       np.float32(w - 1 - x) / pad[2]),
+                                      1.0 - np.minimum(np.float32(y) / pad[1],
+                                                       np.float32(h - 1 - y) / pad[3]))
+                    blur = int(qsize * blur_ratio)
+                    if blur % 2 == 0:
+                        blur += 1
+                    blur_img = cv2.boxFilter(pad_img, 0, ksize=(blur, blur))
+                    # blur_img = cv2.GaussianBlur(pad_img, (blur, blur), 0)
+
+                    pad_img = pad_img.astype('float32')
+                    pad_img += (blur_img - pad_img) * np.clip(mask * 3.0 + 1.0, 0.0, 1.0)
+                    pad_img += (np.median(pad_img, axis=(0, 1)) - pad_img) * np.clip(mask, 0.0, 1.0)
+                    pad_img = np.clip(pad_img, 0, 255)  # float32, [0, 255]
+                    self.pad_input_imgs.append(pad_img)
+                else:
+                    self.pad_input_imgs.append(np.copy(self.input_img))
+
+        return len(self.all_landmarks_5)
+
+    def align_warp_face(self, save_cropped_path=None, border_mode='constant'):
+        """Align and warp faces with face template.
+        """
+        if self.pad_blur:
+            assert len(self.pad_input_imgs) == len(
+                self.all_landmarks_5), f'Mismatched samples: {len(self.pad_input_imgs)} and {len(self.all_landmarks_5)}'
+        for idx, landmark in enumerate(self.all_landmarks_5):
+            # use 5 landmarks to get affine matrix
+            # use cv2.LMEDS method for the equivalence to skimage transform
+            # ref: https://blog.csdn.net/yichxi/article/details/115827338
+            affine_matrix = cv2.estimateAffinePartial2D(landmark, self.face_template, method=cv2.LMEDS)[0]
+            self.affine_matrices.append(affine_matrix)
+            # warp and crop faces
+            if border_mode == 'constant':
+                border_mode = cv2.BORDER_CONSTANT
+            elif border_mode == 'reflect101':
+                border_mode = cv2.BORDER_REFLECT101
+            elif border_mode == 'reflect':
+                border_mode = cv2.BORDER_REFLECT
+            if self.pad_blur:
+                input_img = self.pad_input_imgs[idx]
+            else:
+                input_img = self.input_img
+            cropped_face = cv2.warpAffine(
+                input_img, affine_matrix, self.face_size, borderMode=border_mode, borderValue=(135, 133, 132))  # gray
+            self.cropped_faces.append(cropped_face)
+            # save the cropped face
+            if save_cropped_path is not None:
+                path = os.path.splitext(save_cropped_path)[0]
+                save_path = f'{path}_{idx:02d}.{self.save_ext}'
+                imwrite(cropped_face, save_path)
+
+    def get_inverse_affine(self, save_inverse_affine_path=None):
+        """Get inverse affine matrix."""
+        for idx, affine_matrix in enumerate(self.affine_matrices):
+            inverse_affine = cv2.invertAffineTransform(affine_matrix)
+            inverse_affine *= self.upscale_factor
+            self.inverse_affine_matrices.append(inverse_affine)
+            # save inverse affine matrices
+            if save_inverse_affine_path is not None:
+                path, _ = os.path.splitext(save_inverse_affine_path)
+                save_path = f'{path}_{idx:02d}.pth'
+                torch.save(inverse_affine, save_path)
+
+    def add_restored_face(self, face):
+        self.restored_faces.append(face)
+
+    def paste_faces_to_input_image(self, save_path=None, upsample_img=None):
+        h, w, _ = self.input_img.shape
+        h_up, w_up = int(h * self.upscale_factor), int(w * self.upscale_factor)
+
+        if upsample_img is None:
+            # simply resize the background
+            upsample_img = cv2.resize(self.input_img, (w_up, h_up), interpolation=cv2.INTER_LANCZOS4)
+        else:
+            upsample_img = cv2.resize(upsample_img, (w_up, h_up), interpolation=cv2.INTER_LANCZOS4)
+
+        assert len(self.restored_faces) == len(
+            self.inverse_affine_matrices), ('length of restored_faces and affine_matrices are different.')
+        for restored_face, inverse_affine in zip(self.restored_faces, self.inverse_affine_matrices):
+            # Add an offset to inverse affine matrix, for more precise back alignment
+            if self.upscale_factor > 1:
+                extra_offset = 0.5 * self.upscale_factor
+            else:
+                extra_offset = 0
+            inverse_affine[:, 2] += extra_offset
+            inv_restored = cv2.warpAffine(restored_face, inverse_affine, (w_up, h_up))
+
+            if self.use_parse:
+                # inference
+                face_input = cv2.resize(restored_face, (512, 512), interpolation=cv2.INTER_LINEAR)
+                face_input = img2tensor(face_input.astype('float32') / 255., bgr2rgb=True, float32=True)
+                normalize(face_input, (0.5, 0.5, 0.5), (0.5, 0.5, 0.5), inplace=True)
+                face_input = torch.unsqueeze(face_input, 0).to(self.device)
+                with torch.no_grad():
+                    out = self.face_parse(face_input)[0]
+                out = out.argmax(dim=1).squeeze().cpu().numpy()
+
+                mask = np.zeros(out.shape)
+                MASK_COLORMAP = [0, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 0, 255, 0, 0, 0]
+                for idx, color in enumerate(MASK_COLORMAP):
+                    mask[out == idx] = color
+                #  blur the mask
+                mask = cv2.GaussianBlur(mask, (101, 101), 11)
+                mask = cv2.GaussianBlur(mask, (101, 101), 11)
+                # remove the black borders
+                thres = 10
+                mask[:thres, :] = 0
+                mask[-thres:, :] = 0
+                mask[:, :thres] = 0
+                mask[:, -thres:] = 0
+                mask = mask / 255.
+
+                mask = cv2.resize(mask, restored_face.shape[:2])
+                mask = cv2.warpAffine(mask, inverse_affine, (w_up, h_up), flags=3)
+                inv_soft_mask = mask[:, :, None]
+                pasted_face = inv_restored
+
+            else:  # use square parse maps
+                mask = np.ones(self.face_size, dtype=np.float32)
+                inv_mask = cv2.warpAffine(mask, inverse_affine, (w_up, h_up))
+                # remove the black borders
+                inv_mask_erosion = cv2.erode(
+                    inv_mask, np.ones((int(2 * self.upscale_factor), int(2 * self.upscale_factor)), np.uint8))
+                pasted_face = inv_mask_erosion[:, :, None] * inv_restored
+                total_face_area = np.sum(inv_mask_erosion)  # // 3
+                # compute the fusion edge based on the area of face
+                w_edge = int(total_face_area**0.5) // 20
+                erosion_radius = w_edge * 2
+                inv_mask_center = cv2.erode(inv_mask_erosion, np.ones((erosion_radius, erosion_radius), np.uint8))
+                blur_size = w_edge * 2
+                inv_soft_mask = cv2.GaussianBlur(inv_mask_center, (blur_size + 1, blur_size + 1), 0)
+                if len(upsample_img.shape) == 2:  # upsample_img is gray image
+                    upsample_img = upsample_img[:, :, None]
+                inv_soft_mask = inv_soft_mask[:, :, None]
+
+            if len(upsample_img.shape) == 3 and upsample_img.shape[2] == 4:  # alpha channel
+                alpha = upsample_img[:, :, 3:]
+                upsample_img = inv_soft_mask * pasted_face + (1 - inv_soft_mask) * upsample_img[:, :, 0:3]
+                upsample_img = np.concatenate((upsample_img, alpha), axis=2)
+            else:
+                upsample_img = inv_soft_mask * pasted_face + (1 - inv_soft_mask) * upsample_img
+
+        if np.max(upsample_img) > 256:  # 16-bit image
+            upsample_img = upsample_img.astype(np.uint16)
+        else:
+            upsample_img = upsample_img.astype(np.uint8)
+        if save_path is not None:
+            path = os.path.splitext(save_path)[0]
+            save_path = f'{path}.{self.save_ext}'
+            imwrite(upsample_img, save_path)
+        return upsample_img
+
+    def clean_all(self):
+        self.all_landmarks_5 = []
+        self.restored_faces = []
+        self.affine_matrices = []
+        self.cropped_faces = []
+        self.inverse_affine_matrices = []
+        self.det_faces = []
+        self.pad_input_imgs = []

extras/facexlib/utils/face_utils.py

@@ -0,0 +1,250 @@
+import cv2
+import numpy as np
+import torch
+
+
+def compute_increased_bbox(bbox, increase_area, preserve_aspect=True):
+    left, top, right, bot = bbox
+    width = right - left
+    height = bot - top
+
+    if preserve_aspect:
+        width_increase = max(increase_area, ((1 + 2 * increase_area) * height - width) / (2 * width))
+        height_increase = max(increase_area, ((1 + 2 * increase_area) * width - height) / (2 * height))
+    else:
+        width_increase = height_increase = increase_area
+    left = int(left - width_increase * width)
+    top = int(top - height_increase * height)
+    right = int(right + width_increase * width)
+    bot = int(bot + height_increase * height)
+    return (left, top, right, bot)
+
+
+def get_valid_bboxes(bboxes, h, w):
+    left = max(bboxes[0], 0)
+    top = max(bboxes[1], 0)
+    right = min(bboxes[2], w)
+    bottom = min(bboxes[3], h)
+    return (left, top, right, bottom)
+
+
+def align_crop_face_landmarks(img,
+                              landmarks,
+                              output_size,
+                              transform_size=None,
+                              enable_padding=True,
+                              return_inverse_affine=False,
+                              shrink_ratio=(1, 1)):
+    """Align and crop face with landmarks.
+
+    The output_size and transform_size are based on width. The height is
+    adjusted based on shrink_ratio_h/shring_ration_w.
+
+    Modified from:
+    https://github.com/NVlabs/ffhq-dataset/blob/master/download_ffhq.py
+
+    Args:
+        img (Numpy array): Input image.
+        landmarks (Numpy array): 5 or 68 or 98 landmarks.
+        output_size (int): Output face size.
+        transform_size (ing): Transform size. Usually the four time of
+            output_size.
+        enable_padding (float): Default: True.
+        shrink_ratio (float | tuple[float] | list[float]): Shring the whole
+            face for height and width (crop larger area). Default: (1, 1).
+
+    Returns:
+        (Numpy array): Cropped face.
+    """
+    lm_type = 'retinaface_5'  # Options: dlib_5, retinaface_5
+
+    if isinstance(shrink_ratio, (float, int)):
+        shrink_ratio = (shrink_ratio, shrink_ratio)
+    if transform_size is None:
+        transform_size = output_size * 4
+
+    # Parse landmarks
+    lm = np.array(landmarks)
+    if lm.shape[0] == 5 and lm_type == 'retinaface_5':
+        eye_left = lm[0]
+        eye_right = lm[1]
+        mouth_avg = (lm[3] + lm[4]) * 0.5
+    elif lm.shape[0] == 5 and lm_type == 'dlib_5':
+        lm_eye_left = lm[2:4]
+        lm_eye_right = lm[0:2]
+        eye_left = np.mean(lm_eye_left, axis=0)
+        eye_right = np.mean(lm_eye_right, axis=0)
+        mouth_avg = lm[4]
+    elif lm.shape[0] == 68:
+        lm_eye_left = lm[36:42]
+        lm_eye_right = lm[42:48]
+        eye_left = np.mean(lm_eye_left, axis=0)
+        eye_right = np.mean(lm_eye_right, axis=0)
+        mouth_avg = (lm[48] + lm[54]) * 0.5
+    elif lm.shape[0] == 98:
+        lm_eye_left = lm[60:68]
+        lm_eye_right = lm[68:76]
+        eye_left = np.mean(lm_eye_left, axis=0)
+        eye_right = np.mean(lm_eye_right, axis=0)
+        mouth_avg = (lm[76] + lm[82]) * 0.5
+
+    eye_avg = (eye_left + eye_right) * 0.5
+    eye_to_eye = eye_right - eye_left
+    eye_to_mouth = mouth_avg - eye_avg
+
+    # Get the oriented crop rectangle
+    # x: half width of the oriented crop rectangle
+    x = eye_to_eye - np.flipud(eye_to_mouth) * [-1, 1]
+    #  - np.flipud(eye_to_mouth) * [-1, 1]: rotate 90 clockwise
+    # norm with the hypotenuse: get the direction
+    x /= np.hypot(*x)  # get the hypotenuse of a right triangle
+    rect_scale = 1  # TODO: you can edit it to get larger rect
+    x *= max(np.hypot(*eye_to_eye) * 2.0 * rect_scale, np.hypot(*eye_to_mouth) * 1.8 * rect_scale)
+    # y: half height of the oriented crop rectangle
+    y = np.flipud(x) * [-1, 1]
+
+    x *= shrink_ratio[1]  # width
+    y *= shrink_ratio[0]  # height
+
+    # c: center
+    c = eye_avg + eye_to_mouth * 0.1
+    # quad: (left_top, left_bottom, right_bottom, right_top)
+    quad = np.stack([c - x - y, c - x + y, c + x + y, c + x - y])
+    # qsize: side length of the square
+    qsize = np.hypot(*x) * 2
+
+    quad_ori = np.copy(quad)
+    # Shrink, for large face
+    # TODO: do we really need shrink
+    shrink = int(np.floor(qsize / output_size * 0.5))
+    if shrink > 1:
+        h, w = img.shape[0:2]
+        rsize = (int(np.rint(float(w) / shrink)), int(np.rint(float(h) / shrink)))
+        img = cv2.resize(img, rsize, interpolation=cv2.INTER_AREA)
+        quad /= shrink
+        qsize /= shrink
+
+    # Crop
+    h, w = img.shape[0:2]
+    border = max(int(np.rint(qsize * 0.1)), 3)
+    crop = (int(np.floor(min(quad[:, 0]))), int(np.floor(min(quad[:, 1]))), int(np.ceil(max(quad[:, 0]))),
+            int(np.ceil(max(quad[:, 1]))))
+    crop = (max(crop[0] - border, 0), max(crop[1] - border, 0), min(crop[2] + border, w), min(crop[3] + border, h))
+    if crop[2] - crop[0] < w or crop[3] - crop[1] < h:
+        img = img[crop[1]:crop[3], crop[0]:crop[2], :]
+        quad -= crop[0:2]
+
+    # Pad
+    # pad: (width_left, height_top, width_right, height_bottom)
+    h, w = img.shape[0:2]
+    pad = (int(np.floor(min(quad[:, 0]))), int(np.floor(min(quad[:, 1]))), int(np.ceil(max(quad[:, 0]))),
+           int(np.ceil(max(quad[:, 1]))))
+    pad = (max(-pad[0] + border, 0), max(-pad[1] + border, 0), max(pad[2] - w + border, 0), max(pad[3] - h + border, 0))
+    if enable_padding and max(pad) > border - 4:
+        pad = np.maximum(pad, int(np.rint(qsize * 0.3)))
+        img = np.pad(img, ((pad[1], pad[3]), (pad[0], pad[2]), (0, 0)), 'reflect')
+        h, w = img.shape[0:2]
+        y, x, _ = np.ogrid[:h, :w, :1]
+        mask = np.maximum(1.0 - np.minimum(np.float32(x) / pad[0],
+                                           np.float32(w - 1 - x) / pad[2]),
+                          1.0 - np.minimum(np.float32(y) / pad[1],
+                                           np.float32(h - 1 - y) / pad[3]))
+        blur = int(qsize * 0.02)
+        if blur % 2 == 0:
+            blur += 1
+        blur_img = cv2.boxFilter(img, 0, ksize=(blur, blur))
+
+        img = img.astype('float32')
+        img += (blur_img - img) * np.clip(mask * 3.0 + 1.0, 0.0, 1.0)
+        img += (np.median(img, axis=(0, 1)) - img) * np.clip(mask, 0.0, 1.0)
+        img = np.clip(img, 0, 255)  # float32, [0, 255]
+        quad += pad[:2]
+
+    # Transform use cv2
+    h_ratio = shrink_ratio[0] / shrink_ratio[1]
+    dst_h, dst_w = int(transform_size * h_ratio), transform_size
+    template = np.array([[0, 0], [0, dst_h], [dst_w, dst_h], [dst_w, 0]])
+    # use cv2.LMEDS method for the equivalence to skimage transform
+    # ref: https://blog.csdn.net/yichxi/article/details/115827338
+    affine_matrix = cv2.estimateAffinePartial2D(quad, template, method=cv2.LMEDS)[0]
+    cropped_face = cv2.warpAffine(
+        img, affine_matrix, (dst_w, dst_h), borderMode=cv2.BORDER_CONSTANT, borderValue=(135, 133, 132))  # gray
+
+    if output_size < transform_size:
+        cropped_face = cv2.resize(
+            cropped_face, (output_size, int(output_size * h_ratio)), interpolation=cv2.INTER_LINEAR)
+
+    if return_inverse_affine:
+        dst_h, dst_w = int(output_size * h_ratio), output_size
+        template = np.array([[0, 0], [0, dst_h], [dst_w, dst_h], [dst_w, 0]])
+        # use cv2.LMEDS method for the equivalence to skimage transform
+        # ref: https://blog.csdn.net/yichxi/article/details/115827338
+        affine_matrix = cv2.estimateAffinePartial2D(
+            quad_ori, np.array([[0, 0], [0, output_size], [dst_w, dst_h], [dst_w, 0]]), method=cv2.LMEDS)[0]
+        inverse_affine = cv2.invertAffineTransform(affine_matrix)
+    else:
+        inverse_affine = None
+    return cropped_face, inverse_affine
+
+
+def paste_face_back(img, face, inverse_affine):
+    h, w = img.shape[0:2]
+    face_h, face_w = face.shape[0:2]
+    inv_restored = cv2.warpAffine(face, inverse_affine, (w, h))
+    mask = np.ones((face_h, face_w, 3), dtype=np.float32)
+    inv_mask = cv2.warpAffine(mask, inverse_affine, (w, h))
+    # remove the black borders
+    inv_mask_erosion = cv2.erode(inv_mask, np.ones((2, 2), np.uint8))
+    inv_restored_remove_border = inv_mask_erosion * inv_restored
+    total_face_area = np.sum(inv_mask_erosion) // 3
+    # compute the fusion edge based on the area of face
+    w_edge = int(total_face_area**0.5) // 20
+    erosion_radius = w_edge * 2
+    inv_mask_center = cv2.erode(inv_mask_erosion, np.ones((erosion_radius, erosion_radius), np.uint8))
+    blur_size = w_edge * 2
+    inv_soft_mask = cv2.GaussianBlur(inv_mask_center, (blur_size + 1, blur_size + 1), 0)
+    img = inv_soft_mask * inv_restored_remove_border + (1 - inv_soft_mask) * img
+    # float32, [0, 255]
+    return img
+
+
+if __name__ == '__main__':
+    import os
+
+    from extras.facexlib.detection import init_detection_model
+    from extras.facexlib.utils.face_restoration_helper import get_largest_face
+    from extras.facexlib.visualization import visualize_detection
+
+    img_path = '/home/wxt/datasets/ffhq/ffhq_wild/00009.png'
+    img_name = os.splitext(os.path.basename(img_path))[0]
+
+    # initialize model
+    det_net = init_detection_model('retinaface_resnet50', half=False)
+    img_ori = cv2.imread(img_path)
+    h, w = img_ori.shape[0:2]
+    # if larger than 800, scale it
+    scale = max(h / 800, w / 800)
+    if scale > 1:
+        img = cv2.resize(img_ori, (int(w / scale), int(h / scale)), interpolation=cv2.INTER_LINEAR)
+
+    with torch.no_grad():
+        bboxes = det_net.detect_faces(img, 0.97)
+    if scale > 1:
+        bboxes *= scale  # the score is incorrect
+    bboxes = get_largest_face(bboxes, h, w)[0]
+    visualize_detection(img_ori, [bboxes], f'tmp/{img_name}_det.png')
+
+    landmarks = np.array([[bboxes[i], bboxes[i + 1]] for i in range(5, 15, 2)])
+
+    cropped_face, inverse_affine = align_crop_face_landmarks(
+        img_ori,
+        landmarks,
+        output_size=512,
+        transform_size=None,
+        enable_padding=True,
+        return_inverse_affine=True,
+        shrink_ratio=(1, 1))
+
+    cv2.imwrite(f'tmp/{img_name}_cropeed_face.png', cropped_face)
+    img = paste_face_back(img_ori, cropped_face, inverse_affine)
+    cv2.imwrite(f'tmp/{img_name}_back.png', img)

extras/facexlib/utils/misc.py

@@ -0,0 +1,118 @@
+import cv2
+import os
+import os.path as osp
+import torch
+from torch.hub import download_url_to_file, get_dir
+from urllib.parse import urlparse
+
+ROOT_DIR = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+
+def imwrite(img, file_path, params=None, auto_mkdir=True):
+    """Write image to file.
+
+    Args:
+        img (ndarray): Image array to be written.
+        file_path (str): Image file path.
+        params (None or list): Same as opencv's :func:`imwrite` interface.
+        auto_mkdir (bool): If the parent folder of `file_path` does not exist,
+            whether to create it automatically.
+
+    Returns:
+        bool: Successful or not.
+    """
+    if auto_mkdir:
+        dir_name = os.path.abspath(os.path.dirname(file_path))
+        os.makedirs(dir_name, exist_ok=True)
+    return cv2.imwrite(file_path, img, params)
+
+
+def img2tensor(imgs, bgr2rgb=True, float32=True):
+    """Numpy array to tensor.
+
+    Args:
+        imgs (list[ndarray] | ndarray): Input images.
+        bgr2rgb (bool): Whether to change bgr to rgb.
+        float32 (bool): Whether to change to float32.
+
+    Returns:
+        list[tensor] | tensor: Tensor images. If returned results only have
+            one element, just return tensor.
+    """
+
+    def _totensor(img, bgr2rgb, float32):
+        if img.shape[2] == 3 and bgr2rgb:
+            if img.dtype == 'float64':
+                img = img.astype('float32')
+            img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
+        img = torch.from_numpy(img.transpose(2, 0, 1))
+        if float32:
+            img = img.float()
+        return img
+
+    if isinstance(imgs, list):
+        return [_totensor(img, bgr2rgb, float32) for img in imgs]
+    else:
+        return _totensor(imgs, bgr2rgb, float32)
+
+
+def load_file_from_url(url, model_dir=None, progress=True, file_name=None, save_dir=None):
+    """Ref:https://github.com/1adrianb/face-alignment/blob/master/face_alignment/utils.py
+    """
+    if model_dir is None:
+        hub_dir = get_dir()
+        model_dir = os.path.join(hub_dir, 'checkpoints')
+
+    if save_dir is None:
+        save_dir = os.path.join(ROOT_DIR, model_dir)
+    os.makedirs(save_dir, exist_ok=True)
+
+    parts = urlparse(url)
+    filename = os.path.basename(parts.path)
+    if file_name is not None:
+        filename = file_name
+    cached_file = os.path.abspath(os.path.join(save_dir, filename))
+    if not os.path.exists(cached_file):
+        print(f'Downloading: "{url}" to {cached_file}\n')
+        download_url_to_file(url, cached_file, hash_prefix=None, progress=progress)
+    return cached_file
+
+
+def scandir(dir_path, suffix=None, recursive=False, full_path=False):
+    """Scan a directory to find the interested files.
+    Args:
+        dir_path (str): Path of the directory.
+        suffix (str | tuple(str), optional): File suffix that we are
+            interested in. Default: None.
+        recursive (bool, optional): If set to True, recursively scan the
+            directory. Default: False.
+        full_path (bool, optional): If set to True, include the dir_path.
+            Default: False.
+    Returns:
+        A generator for all the interested files with relative paths.
+    """
+
+    if (suffix is not None) and not isinstance(suffix, (str, tuple)):
+        raise TypeError('"suffix" must be a string or tuple of strings')
+
+    root = dir_path
+
+    def _scandir(dir_path, suffix, recursive):
+        for entry in os.scandir(dir_path):
+            if not entry.name.startswith('.') and entry.is_file():
+                if full_path:
+                    return_path = entry.path
+                else:
+                    return_path = osp.relpath(entry.path, root)
+
+                if suffix is None:
+                    yield return_path
+                elif return_path.endswith(suffix):
+                    yield return_path
+            else:
+                if recursive:
+                    yield from _scandir(entry.path, suffix=suffix, recursive=recursive)
+                else:
+                    continue
+
+    return _scandir(dir_path, suffix=suffix, recursive=recursive)

extras/interrogate.py

@@ -0,0 +1,63 @@
+import os
+import torch
+import ldm_patched.modules.model_management as model_management
+
+from torchvision import transforms
+from torchvision.transforms.functional import InterpolationMode
+from modules.model_loader import load_file_from_url
+from modules.config import path_clip_vision
+from ldm_patched.modules.model_patcher import ModelPatcher
+from extras.BLIP.models.blip import blip_decoder
+
+
+blip_image_eval_size = 384
+blip_repo_root = os.path.join(os.path.dirname(__file__), 'BLIP')
+
+
+class Interrogator:
+    def __init__(self):
+        self.blip_model = None
+        self.load_device = torch.device('cpu')
+        self.offload_device = torch.device('cpu')
+        self.dtype = torch.float32
+
+    @torch.no_grad()
+    @torch.inference_mode()
+    def interrogate(self, img_rgb):
+        if self.blip_model is None:
+            filename = load_file_from_url(
+                url='https://huggingface.co/lllyasviel/misc/resolve/main/model_base_caption_capfilt_large.pth',
+                model_dir=path_clip_vision,
+                file_name='model_base_caption_capfilt_large.pth',
+            )
+
+            model = blip_decoder(pretrained=filename, image_size=blip_image_eval_size, vit='base',
+                                 med_config=os.path.join(blip_repo_root, "configs", "med_config.json"))
+            model.eval()
+
+            self.load_device = model_management.text_encoder_device()
+            self.offload_device = model_management.text_encoder_offload_device()
+            self.dtype = torch.float32
+
+            model.to(self.offload_device)
+
+            if model_management.should_use_fp16(device=self.load_device):
+                model.half()
+                self.dtype = torch.float16
+
+            self.blip_model = ModelPatcher(model, load_device=self.load_device, offload_device=self.offload_device)
+
+        model_management.load_model_gpu(self.blip_model)
+
+        gpu_image = transforms.Compose([
+            transforms.ToTensor(),
+            transforms.Resize((blip_image_eval_size, blip_image_eval_size), interpolation=InterpolationMode.BICUBIC),
+            transforms.Normalize((0.48145466, 0.4578275, 0.40821073), (0.26862954, 0.26130258, 0.27577711))
+        ])(img_rgb).unsqueeze(0).to(device=self.load_device, dtype=self.dtype)
+
+        caption = self.blip_model.model.generate(gpu_image, sample=True, num_beams=1, max_length=75)[0]
+
+        return caption
+
+
+default_interrogator = Interrogator().interrogate

extras/ip_adapter.py

@@ -0,0 +1,281 @@
+import torch
+import ldm_patched.modules.clip_vision
+import safetensors.torch as sf
+import ldm_patched.modules.model_management as model_management
+import contextlib
+import ldm_patched.ldm.modules.attention as attention
+
+from extras.resampler import Resampler
+from ldm_patched.modules.model_patcher import ModelPatcher
+from modules.core import numpy_to_pytorch
+
+
+SD_V12_CHANNELS = [320] * 4 + [640] * 4 + [1280] * 4 + [1280] * 6 + [640] * 6 + [320] * 6 + [1280] * 2
+SD_XL_CHANNELS = [640] * 8 + [1280] * 40 + [1280] * 60 + [640] * 12 + [1280] * 20
+
+
+def sdp(q, k, v, extra_options):
+    return attention.optimized_attention(q, k, v, heads=extra_options["n_heads"], mask=None)
+
+
+class ImageProjModel(torch.nn.Module):
+    def __init__(self, cross_attention_dim=1024, clip_embeddings_dim=1024, clip_extra_context_tokens=4):
+        super().__init__()
+
+        self.cross_attention_dim = cross_attention_dim
+        self.clip_extra_context_tokens = clip_extra_context_tokens
+        self.proj = torch.nn.Linear(clip_embeddings_dim, self.clip_extra_context_tokens * cross_attention_dim)
+        self.norm = torch.nn.LayerNorm(cross_attention_dim)
+
+    def forward(self, image_embeds):
+        embeds = image_embeds
+        clip_extra_context_tokens = self.proj(embeds).reshape(-1, self.clip_extra_context_tokens,
+                                                              self.cross_attention_dim)
+        clip_extra_context_tokens = self.norm(clip_extra_context_tokens)
+        return clip_extra_context_tokens
+
+
+class To_KV(torch.nn.Module):
+    def __init__(self, cross_attention_dim):
+        super().__init__()
+
+        channels = SD_XL_CHANNELS if cross_attention_dim == 2048 else SD_V12_CHANNELS
+        self.to_kvs = torch.nn.ModuleList(
+            [torch.nn.Linear(cross_attention_dim, channel, bias=False) for channel in channels])
+
+    def load_state_dict_ordered(self, sd):
+        state_dict = []
+        for i in range(4096):
+            for k in ['k', 'v']:
+                key = f'{i}.to_{k}_ip.weight'
+                if key in sd:
+                    state_dict.append(sd[key])
+        for i, v in enumerate(state_dict):
+            self.to_kvs[i].weight = torch.nn.Parameter(v, requires_grad=False)
+
+
+class IPAdapterModel(torch.nn.Module):
+    def __init__(self, state_dict, plus, cross_attention_dim=768, clip_embeddings_dim=1024, clip_extra_context_tokens=4,
+                 sdxl_plus=False):
+        super().__init__()
+        self.plus = plus
+        if self.plus:
+            self.image_proj_model = Resampler(
+                dim=1280 if sdxl_plus else cross_attention_dim,
+                depth=4,
+                dim_head=64,
+                heads=20 if sdxl_plus else 12,
+                num_queries=clip_extra_context_tokens,
+                embedding_dim=clip_embeddings_dim,
+                output_dim=cross_attention_dim,
+                ff_mult=4
+            )
+        else:
+            self.image_proj_model = ImageProjModel(
+                cross_attention_dim=cross_attention_dim,
+                clip_embeddings_dim=clip_embeddings_dim,
+                clip_extra_context_tokens=clip_extra_context_tokens
+            )
+
+        self.image_proj_model.load_state_dict(state_dict["image_proj"])
+        self.ip_layers = To_KV(cross_attention_dim)
+        self.ip_layers.load_state_dict_ordered(state_dict["ip_adapter"])
+
+
+clip_vision: ldm_patched.modules.clip_vision.ClipVisionModel = None
+ip_negative: torch.Tensor = None
+ip_adapters: dict = {}
+
+
+def load_ip_adapter(clip_vision_path, ip_negative_path, ip_adapter_path):
+    global clip_vision, ip_negative, ip_adapters
+
+    if clip_vision is None and isinstance(clip_vision_path, str):
+        clip_vision = ldm_patched.modules.clip_vision.load(clip_vision_path)
+
+    if ip_negative is None and isinstance(ip_negative_path, str):
+        ip_negative = sf.load_file(ip_negative_path)['data']
+
+    if not isinstance(ip_adapter_path, str) or ip_adapter_path in ip_adapters:
+        return
+
+    load_device = model_management.get_torch_device()
+    offload_device = torch.device('cpu')
+
+    use_fp16 = model_management.should_use_fp16(device=load_device)
+    ip_state_dict = torch.load(ip_adapter_path, map_location="cpu")
+    plus = "latents" in ip_state_dict["image_proj"]
+    cross_attention_dim = ip_state_dict["ip_adapter"]["1.to_k_ip.weight"].shape[1]
+    sdxl = cross_attention_dim == 2048
+    sdxl_plus = sdxl and plus
+
+    if plus:
+        clip_extra_context_tokens = ip_state_dict["image_proj"]["latents"].shape[1]
+        clip_embeddings_dim = ip_state_dict["image_proj"]["latents"].shape[2]
+    else:
+        clip_extra_context_tokens = ip_state_dict["image_proj"]["proj.weight"].shape[0] // cross_attention_dim
+        clip_embeddings_dim = None
+
+    ip_adapter = IPAdapterModel(
+        ip_state_dict,
+        plus=plus,
+        cross_attention_dim=cross_attention_dim,
+        clip_embeddings_dim=clip_embeddings_dim,
+        clip_extra_context_tokens=clip_extra_context_tokens,
+        sdxl_plus=sdxl_plus
+    )
+    ip_adapter.sdxl = sdxl
+    ip_adapter.load_device = load_device
+    ip_adapter.offload_device = offload_device
+    ip_adapter.dtype = torch.float16 if use_fp16 else torch.float32
+    ip_adapter.to(offload_device, dtype=ip_adapter.dtype)
+
+    image_proj_model = ModelPatcher(model=ip_adapter.image_proj_model, load_device=load_device,
+                                    offload_device=offload_device)
+    ip_layers = ModelPatcher(model=ip_adapter.ip_layers, load_device=load_device,
+                             offload_device=offload_device)
+
+    ip_adapters[ip_adapter_path] = dict(
+        ip_adapter=ip_adapter,
+        image_proj_model=image_proj_model,
+        ip_layers=ip_layers,
+        ip_unconds=None
+    )
+
+    return
+
+
+@torch.no_grad()
+@torch.inference_mode()
+def clip_preprocess(image):
+    mean = torch.tensor([0.48145466, 0.4578275, 0.40821073], device=image.device, dtype=image.dtype).view([1, 3, 1, 1])
+    std = torch.tensor([0.26862954, 0.26130258, 0.27577711], device=image.device, dtype=image.dtype).view([1, 3, 1, 1])
+    image = image.movedim(-1, 1)
+
+    # https://github.com/tencent-ailab/IP-Adapter/blob/d580c50a291566bbf9fc7ac0f760506607297e6d/README.md?plain=1#L75
+    B, C, H, W = image.shape
+    assert H == 224 and W == 224
+
+    return (image - mean) / std
+
+
+@torch.no_grad()
+@torch.inference_mode()
+def preprocess(img, ip_adapter_path):
+    global ip_adapters
+    entry = ip_adapters[ip_adapter_path]
+
+    ldm_patched.modules.model_management.load_model_gpu(clip_vision.patcher)
+    pixel_values = clip_preprocess(numpy_to_pytorch(img).to(clip_vision.load_device))
+    outputs = clip_vision.model(pixel_values=pixel_values, output_hidden_states=True)
+
+    ip_adapter = entry['ip_adapter']
+    ip_layers = entry['ip_layers']
+    image_proj_model = entry['image_proj_model']
+    ip_unconds = entry['ip_unconds']
+
+    if ip_adapter.plus:
+        cond = outputs.hidden_states[-2]
+    else:
+        cond = outputs.image_embeds
+
+    cond = cond.to(device=ip_adapter.load_device, dtype=ip_adapter.dtype)
+
+    ldm_patched.modules.model_management.load_model_gpu(image_proj_model)
+    cond = image_proj_model.model(cond).to(device=ip_adapter.load_device, dtype=ip_adapter.dtype)
+
+    ldm_patched.modules.model_management.load_model_gpu(ip_layers)
+
+    if ip_unconds is None:
+        uncond = ip_negative.to(device=ip_adapter.load_device, dtype=ip_adapter.dtype)
+        ip_unconds = [m(uncond).cpu() for m in ip_layers.model.to_kvs]
+        entry['ip_unconds'] = ip_unconds
+
+    ip_conds = [m(cond).cpu() for m in ip_layers.model.to_kvs]
+
+    return ip_conds, ip_unconds
+
+
+@torch.no_grad()
+@torch.inference_mode()
+def patch_model(model, tasks):
+    new_model = model.clone()
+
+    def make_attn_patcher(ip_index):
+        def patcher(n, context_attn2, value_attn2, extra_options):
+            org_dtype = n.dtype
+            current_step = float(model.model.diffusion_model.current_step.detach().cpu().numpy()[0])
+            cond_or_uncond = extra_options['cond_or_uncond']
+
+            q = n
+            k = [context_attn2]
+            v = [value_attn2]
+            b, _, _ = q.shape
+
+            for (cs, ucs), cn_stop, cn_weight in tasks:
+                if current_step < cn_stop:
+                    ip_k_c = cs[ip_index * 2].to(q)
+                    ip_v_c = cs[ip_index * 2 + 1].to(q)
+                    ip_k_uc = ucs[ip_index * 2].to(q)
+                    ip_v_uc = ucs[ip_index * 2 + 1].to(q)
+
+                    ip_k = torch.cat([(ip_k_c, ip_k_uc)[i] for i in cond_or_uncond], dim=0)
+                    ip_v = torch.cat([(ip_v_c, ip_v_uc)[i] for i in cond_or_uncond], dim=0)
+
+                    # Midjourney's attention formulation of image prompt (non-official reimplementation)
+                    # Written by Lvmin Zhang at Stanford University, 2023 Dec
+                    # For non-commercial use only - if you use this in commercial project then
+                    # probably it has some intellectual property issues.
+                    # Contact lvminzhang@acm.org if you are not sure.
+
+                    # Below is the sensitive part with potential intellectual property issues.
+
+                    ip_v_mean = torch.mean(ip_v, dim=1, keepdim=True)
+                    ip_v_offset = ip_v - ip_v_mean
+
+                    B, F, C = ip_k.shape
+                    channel_penalty = float(C) / 1280.0
+                    weight = cn_weight * channel_penalty
+
+                    ip_k = ip_k * weight
+                    ip_v = ip_v_offset + ip_v_mean * weight
+
+                    k.append(ip_k)
+                    v.append(ip_v)
+
+            k = torch.cat(k, dim=1)
+            v = torch.cat(v, dim=1)
+            out = sdp(q, k, v, extra_options)
+
+
+            return out.to(dtype=org_dtype)
+        return patcher
+
+    def set_model_patch_replace(model, number, key):
+        to = model.model_options["transformer_options"]
+        if "patches_replace" not in to:
+            to["patches_replace"] = {}
+        if "attn2" not in to["patches_replace"]:
+            to["patches_replace"]["attn2"] = {}
+        if key not in to["patches_replace"]["attn2"]:
+            to["patches_replace"]["attn2"][key] = make_attn_patcher(number)
+
+    number = 0
+
+    for id in [4, 5, 7, 8]:
+        block_indices = range(2) if id in [4, 5] else range(10)
+        for index in block_indices:
+            set_model_patch_replace(new_model, number, ("input", id, index))
+            number += 1
+
+    for id in range(6):
+        block_indices = range(2) if id in [3, 4, 5] else range(10)
+        for index in block_indices:
+            set_model_patch_replace(new_model, number, ("output", id, index))
+            number += 1
+
+    for index in range(10):
+        set_model_patch_replace(new_model, number, ("middle", 0, index))
+        number += 1
+
+    return new_model

extras/preprocessors.py

@@ -0,0 +1,82 @@
+import cv2
+import numpy as np
+import modules.advanced_parameters as advanced_parameters
+
+
+def centered_canny(x: np.ndarray):
+    assert isinstance(x, np.ndarray)
+    assert x.ndim == 2 and x.dtype == np.uint8
+
+    y = cv2.Canny(x, int(advanced_parameters.canny_low_threshold), int(advanced_parameters.canny_high_threshold))
+    y = y.astype(np.float32) / 255.0
+    return y
+
+
+def centered_canny_color(x: np.ndarray):
+    assert isinstance(x, np.ndarray)
+    assert x.ndim == 3 and x.shape[2] == 3
+
+    result = [centered_canny(x[..., i]) for i in range(3)]
+    result = np.stack(result, axis=2)
+    return result
+
+
+def pyramid_canny_color(x: np.ndarray):
+    assert isinstance(x, np.ndarray)
+    assert x.ndim == 3 and x.shape[2] == 3
+
+    H, W, C = x.shape
+    acc_edge = None
+
+    for k in [0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0]:
+        Hs, Ws = int(H * k), int(W * k)
+        small = cv2.resize(x, (Ws, Hs), interpolation=cv2.INTER_AREA)
+        edge = centered_canny_color(small)
+        if acc_edge is None:
+            acc_edge = edge
+        else:
+            acc_edge = cv2.resize(acc_edge, (edge.shape[1], edge.shape[0]), interpolation=cv2.INTER_LINEAR)
+            acc_edge = acc_edge * 0.75 + edge * 0.25
+
+    return acc_edge
+
+
+def norm255(x, low=4, high=96):
+    assert isinstance(x, np.ndarray)
+    assert x.ndim == 2 and x.dtype == np.float32
+
+    v_min = np.percentile(x, low)
+    v_max = np.percentile(x, high)
+
+    x -= v_min
+    x /= v_max - v_min
+
+    return x * 255.0
+
+
+def canny_pyramid(x):
+    # For some reasons, SAI's Control-lora Canny seems to be trained on canny maps with non-standard resolutions.
+    # Then we use pyramid to use all resolutions to avoid missing any structure in specific resolutions.
+
+    color_canny = pyramid_canny_color(x)
+    result = np.sum(color_canny, axis=2)
+
+    return norm255(result, low=1, high=99).clip(0, 255).astype(np.uint8)
+
+
+def cpds(x):
+    # cv2.decolor is not "decolor", it is Cewu Lu's method
+    # See http://www.cse.cuhk.edu.hk/leojia/projects/color2gray/index.html
+    # See https://docs.opencv.org/3.0-beta/modules/photo/doc/decolor.html
+
+    raw = cv2.GaussianBlur(x, (0, 0), 0.8)
+    density, boost = cv2.decolor(raw)
+
+    raw = raw.astype(np.float32)
+    density = density.astype(np.float32)
+    boost = boost.astype(np.float32)
+
+    offset = np.sum((raw - boost) ** 2.0, axis=2) ** 0.5
+    result = density + offset
+
+    return norm255(result, low=4, high=96).clip(0, 255).astype(np.uint8)

extras/resampler.py

@@ -0,0 +1,121 @@
+# modified from https://github.com/mlfoundations/open_flamingo/blob/main/open_flamingo/src/helpers.py
+import math
+
+import torch
+import torch.nn as nn
+
+
+# FFN
+def FeedForward(dim, mult=4):
+    inner_dim = int(dim * mult)
+    return nn.Sequential(
+        nn.LayerNorm(dim),
+        nn.Linear(dim, inner_dim, bias=False),
+        nn.GELU(),
+        nn.Linear(inner_dim, dim, bias=False),
+    )
+    
+    
+def reshape_tensor(x, heads):
+    bs, length, width = x.shape
+    #(bs, length, width) --> (bs, length, n_heads, dim_per_head)
+    x = x.view(bs, length, heads, -1)
+    # (bs, length, n_heads, dim_per_head) --> (bs, n_heads, length, dim_per_head)
+    x = x.transpose(1, 2)
+    # (bs, n_heads, length, dim_per_head) --> (bs*n_heads, length, dim_per_head)
+    x = x.reshape(bs, heads, length, -1)
+    return x
+
+
+class PerceiverAttention(nn.Module):
+    def __init__(self, *, dim, dim_head=64, heads=8):
+        super().__init__()
+        self.scale = dim_head**-0.5
+        self.dim_head = dim_head
+        self.heads = heads
+        inner_dim = dim_head * heads
+
+        self.norm1 = nn.LayerNorm(dim)
+        self.norm2 = nn.LayerNorm(dim)
+
+        self.to_q = nn.Linear(dim, inner_dim, bias=False)
+        self.to_kv = nn.Linear(dim, inner_dim * 2, bias=False)
+        self.to_out = nn.Linear(inner_dim, dim, bias=False)
+
+
+    def forward(self, x, latents):
+        """
+        Args:
+            x (torch.Tensor): image features
+                shape (b, n1, D)
+            latent (torch.Tensor): latent features
+                shape (b, n2, D)
+        """
+        x = self.norm1(x)
+        latents = self.norm2(latents)
+        
+        b, l, _ = latents.shape
+
+        q = self.to_q(latents)
+        kv_input = torch.cat((x, latents), dim=-2)
+        k, v = self.to_kv(kv_input).chunk(2, dim=-1)
+        
+        q = reshape_tensor(q, self.heads)
+        k = reshape_tensor(k, self.heads)
+        v = reshape_tensor(v, self.heads)
+
+        # attention
+        scale = 1 / math.sqrt(math.sqrt(self.dim_head))
+        weight = (q * scale) @ (k * scale).transpose(-2, -1) # More stable with f16 than dividing afterwards
+        weight = torch.softmax(weight.float(), dim=-1).type(weight.dtype)
+        out = weight @ v
+        
+        out = out.permute(0, 2, 1, 3).reshape(b, l, -1)
+
+        return self.to_out(out)
+
+
+class Resampler(nn.Module):
+    def __init__(
+        self,
+        dim=1024,
+        depth=8,
+        dim_head=64,
+        heads=16,
+        num_queries=8,
+        embedding_dim=768,
+        output_dim=1024,
+        ff_mult=4,
+    ):
+        super().__init__()
+        
+        self.latents = nn.Parameter(torch.randn(1, num_queries, dim) / dim**0.5)
+        
+        self.proj_in = nn.Linear(embedding_dim, dim)
+
+        self.proj_out = nn.Linear(dim, output_dim)
+        self.norm_out = nn.LayerNorm(output_dim)
+        
+        self.layers = nn.ModuleList([])
+        for _ in range(depth):
+            self.layers.append(
+                nn.ModuleList(
+                    [
+                        PerceiverAttention(dim=dim, dim_head=dim_head, heads=heads),
+                        FeedForward(dim=dim, mult=ff_mult),
+                    ]
+                )
+            )
+
+    def forward(self, x):
+        
+        latents = self.latents.repeat(x.size(0), 1, 1)
+        
+        x = self.proj_in(x)
+        
+        for attn, ff in self.layers:
+            latents = attn(x, latents) + latents
+            latents = ff(latents) + latents
+            
+        latents = self.proj_out(latents)
+        return self.norm_out(latents)

extras/vae_interpose.py

@@ -0,0 +1,93 @@
+# https://github.com/city96/SD-Latent-Interposer/blob/main/interposer.py
+
+import os
+import torch
+import safetensors.torch as sf
+import torch.nn as nn
+import ldm_patched.modules.model_management
+
+from ldm_patched.modules.model_patcher import ModelPatcher
+from modules.config import path_vae_approx
+
+
+class Block(nn.Module):
+    def __init__(self, size):
+        super().__init__()
+        self.join = nn.ReLU()
+        self.long = nn.Sequential(
+            nn.Conv2d(size, size, kernel_size=3, stride=1, padding=1),
+            nn.LeakyReLU(0.1),
+            nn.Conv2d(size, size, kernel_size=3, stride=1, padding=1),
+            nn.LeakyReLU(0.1),
+            nn.Conv2d(size, size, kernel_size=3, stride=1, padding=1),
+        )
+
+    def forward(self, x):
+        y = self.long(x)
+        z = self.join(y + x)
+        return z
+
+
+class Interposer(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.chan = 4
+        self.hid = 128
+
+        self.head_join = nn.ReLU()
+        self.head_short = nn.Conv2d(self.chan, self.hid, kernel_size=3, stride=1, padding=1)
+        self.head_long = nn.Sequential(
+            nn.Conv2d(self.chan, self.hid, kernel_size=3, stride=1, padding=1),
+            nn.LeakyReLU(0.1),
+            nn.Conv2d(self.hid, self.hid, kernel_size=3, stride=1, padding=1),
+            nn.LeakyReLU(0.1),
+            nn.Conv2d(self.hid, self.hid, kernel_size=3, stride=1, padding=1),
+        )
+        self.core = nn.Sequential(
+            Block(self.hid),
+            Block(self.hid),
+            Block(self.hid),
+        )
+        self.tail = nn.Sequential(
+            nn.ReLU(),
+            nn.Conv2d(self.hid, self.chan, kernel_size=3, stride=1, padding=1)
+        )
+
+    def forward(self, x):
+        y = self.head_join(
+            self.head_long(x) +
+            self.head_short(x)
+        )
+        z = self.core(y)
+        return self.tail(z)
+
+
+vae_approx_model = None
+vae_approx_filename = os.path.join(path_vae_approx, 'xl-to-v1_interposer-v3.1.safetensors')
+
+
+def parse(x):
+    global vae_approx_model
+
+    x_origin = x.clone()
+
+    if vae_approx_model is None:
+        model = Interposer()
+        model.eval()
+        sd = sf.load_file(vae_approx_filename)
+        model.load_state_dict(sd)
+        fp16 = ldm_patched.modules.model_management.should_use_fp16()
+        if fp16:
+            model = model.half()
+        vae_approx_model = ModelPatcher(
+            model=model,
+            load_device=ldm_patched.modules.model_management.get_torch_device(),
+            offload_device=torch.device('cpu')
+        )
+        vae_approx_model.dtype = torch.float16 if fp16 else torch.float32
+
+    ldm_patched.modules.model_management.load_model_gpu(vae_approx_model)
+
+    x = x_origin.to(device=vae_approx_model.load_device, dtype=vae_approx_model.dtype)
+    x = vae_approx_model.model(x).to(x_origin)
+    return x

extras/wd14tagger.py

@@ -0,0 +1,98 @@
+# https://huggingface.co/spaces/SmilingWolf/wd-v1-4-tags
+# https://github.com/pythongosssss/ComfyUI-WD14-Tagger/blob/main/wd14tagger.py
+
+# {
+#     "wd-v1-4-moat-tagger-v2": "https://huggingface.co/SmilingWolf/wd-v1-4-moat-tagger-v2",
+#     "wd-v1-4-convnextv2-tagger-v2": "https://huggingface.co/SmilingWolf/wd-v1-4-convnextv2-tagger-v2",
+#     "wd-v1-4-convnext-tagger-v2": "https://huggingface.co/SmilingWolf/wd-v1-4-convnext-tagger-v2",
+#     "wd-v1-4-convnext-tagger": "https://huggingface.co/SmilingWolf/wd-v1-4-convnext-tagger",
+#     "wd-v1-4-vit-tagger-v2": "https://huggingface.co/SmilingWolf/wd-v1-4-vit-tagger-v2"
+# }
+
+
+import numpy as np
+import csv
+import onnxruntime as ort
+
+from PIL import Image
+from onnxruntime import InferenceSession
+from modules.config import path_clip_vision
+from modules.model_loader import load_file_from_url
+
+
+global_model = None
+global_csv = None
+
+
+def default_interrogator(image_rgb, threshold=0.35, character_threshold=0.85, exclude_tags=""):
+    global global_model, global_csv
+
+    model_name = "wd-v1-4-moat-tagger-v2"
+
+    model_onnx_filename = load_file_from_url(
+        url=f'https://huggingface.co/lllyasviel/misc/resolve/main/{model_name}.onnx',
+        model_dir=path_clip_vision,
+        file_name=f'{model_name}.onnx',
+    )
+
+    model_csv_filename = load_file_from_url(
+        url=f'https://huggingface.co/lllyasviel/misc/resolve/main/{model_name}.csv',
+        model_dir=path_clip_vision,
+        file_name=f'{model_name}.csv',
+    )
+
+    if global_model is not None:
+        model = global_model
+    else:
+        model = InferenceSession(model_onnx_filename, providers=ort.get_available_providers())
+        global_model = model
+
+    input = model.get_inputs()[0]
+    height = input.shape[1]
+
+    image = Image.fromarray(image_rgb)  # RGB
+    ratio = float(height)/max(image.size)
+    new_size = tuple([int(x*ratio) for x in image.size])
+    image = image.resize(new_size, Image.LANCZOS)
+    square = Image.new("RGB", (height, height), (255, 255, 255))
+    square.paste(image, ((height-new_size[0])//2, (height-new_size[1])//2))
+
+    image = np.array(square).astype(np.float32)
+    image = image[:, :, ::-1]  # RGB -> BGR
+    image = np.expand_dims(image, 0)
+
+    if global_csv is not None:
+        csv_lines = global_csv
+    else:
+        csv_lines = []
+        with open(model_csv_filename) as f:
+            reader = csv.reader(f)
+            next(reader)
+            for row in reader:
+                csv_lines.append(row)
+        global_csv = csv_lines
+
+    tags = []
+    general_index = None
+    character_index = None
+    for line_num, row in enumerate(csv_lines):
+        if general_index is None and row[2] == "0":
+            general_index = line_num
+        elif character_index is None and row[2] == "4":
+            character_index = line_num
+        tags.append(row[1])
+
+    label_name = model.get_outputs()[0].name
+    probs = model.run([label_name], {input.name: image})[0]
+
+    result = list(zip(tags, probs[0]))
+
+    general = [item for item in result[general_index:character_index] if item[1] > threshold]
+    character = [item for item in result[character_index:] if item[1] > character_threshold]
+
+    all = character + general
+    remove = [s.strip() for s in exclude_tags.lower().split(",")]
+    all = [tag for tag in all if tag[0] not in remove]
+
+    res = ", ".join((item[0].replace("(", "\\(").replace(")", "\\)") for item in all)).replace('_', ' ')
+    return res

fooocus_colab.ipynb

@@ -0,0 +1,35 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "VjYy0F2gZIPR"
+   },
+   "outputs": [],
+   "source": [
+    "!pip install pygit2==1.12.2\n",
+    "%cd /content\n",
+    "!git clone https://github.com/lllyasviel/Fooocus.git\n",
+    "%cd /content/Fooocus\n",
+    "!python entry_with_update.py --share\n"
+   ]
+  }
+ ],
+ "metadata": {
+  "accelerator": "GPU",
+  "colab": {
+   "gpuType": "T4",
+   "provenance": []
+  },
+  "kernelspec": {
+   "display_name": "Python 3",
+   "name": "python3"
+  },
+  "language_info": {
+   "name": "python"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 0
+}

fooocus_version.py

@@ -1,2 +1 @@
-version = '1.0.11'
-
+version = '2.1.851'

javascript/contextMenus.js

@@ -0,0 +1,170 @@
+// based on https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob/v1.6.0/javascript/contextMenus.js
+
+var contextMenuInit = function() {
+    let eventListenerApplied = false;
+    let menuSpecs = new Map();
+
+    const uid = function() {
+        return Date.now().toString(36) + Math.random().toString(36).substring(2);
+    };
+
+    function showContextMenu(event, element, menuEntries) {
+        let posx = event.clientX + document.body.scrollLeft + document.documentElement.scrollLeft;
+        let posy = event.clientY + document.body.scrollTop + document.documentElement.scrollTop;
+
+        let oldMenu = gradioApp().querySelector('#context-menu');
+        if (oldMenu) {
+            oldMenu.remove();
+        }
+
+        let baseStyle = window.getComputedStyle(gradioApp().querySelector('button.selected'));
+
+        const contextMenu = document.createElement('nav');
+        contextMenu.id = "context-menu";
+        contextMenu.style.background = baseStyle.background;
+        contextMenu.style.color = baseStyle.color;
+        contextMenu.style.fontFamily = baseStyle.fontFamily;
+        contextMenu.style.top = posy + 'px';
+        contextMenu.style.left = posx + 'px';
+
+        const contextMenuList = document.createElement('ul');
+        contextMenuList.className = 'context-menu-items';
+        contextMenu.append(contextMenuList);
+
+        menuEntries.forEach(function(entry) {
+            let contextMenuEntry = document.createElement('a');
+            contextMenuEntry.innerHTML = entry['name'];
+            contextMenuEntry.addEventListener("click", function() {
+                entry['func']();
+            });
+            contextMenuList.append(contextMenuEntry);
+
+        });
+
+        gradioApp().appendChild(contextMenu);
+
+        let menuWidth = contextMenu.offsetWidth + 4;
+        let menuHeight = contextMenu.offsetHeight + 4;
+
+        let windowWidth = window.innerWidth;
+        let windowHeight = window.innerHeight;
+
+        if ((windowWidth - posx) < menuWidth) {
+            contextMenu.style.left = windowWidth - menuWidth + "px";
+        }
+
+        if ((windowHeight - posy) < menuHeight) {
+            contextMenu.style.top = windowHeight - menuHeight + "px";
+        }
+
+    }
+
+    function appendContextMenuOption(targetElementSelector, entryName, entryFunction) {
+
+        var currentItems = menuSpecs.get(targetElementSelector);
+
+        if (!currentItems) {
+            currentItems = [];
+            menuSpecs.set(targetElementSelector, currentItems);
+        }
+        let newItem = {
+            id: targetElementSelector + '_' + uid(),
+            name: entryName,
+            func: entryFunction,
+            isNew: true
+        };
+
+        currentItems.push(newItem);
+        return newItem['id'];
+    }
+
+    function removeContextMenuOption(uid) {
+        menuSpecs.forEach(function(v) {
+            let index = -1;
+            v.forEach(function(e, ei) {
+                if (e['id'] == uid) {
+                    index = ei;
+                }
+            });
+            if (index >= 0) {
+                v.splice(index, 1);
+            }
+        });
+    }
+
+    function addContextMenuEventListener() {
+        if (eventListenerApplied) {
+            return;
+        }
+        gradioApp().addEventListener("click", function(e) {
+            if (!e.isTrusted) {
+                return;
+            }
+
+            let oldMenu = gradioApp().querySelector('#context-menu');
+            if (oldMenu) {
+                oldMenu.remove();
+            }
+        });
+        gradioApp().addEventListener("contextmenu", function(e) {
+            let oldMenu = gradioApp().querySelector('#context-menu');
+            if (oldMenu) {
+                oldMenu.remove();
+            }
+            menuSpecs.forEach(function(v, k) {
+                if (e.composedPath()[0].matches(k)) {
+                    showContextMenu(e, e.composedPath()[0], v);
+                    e.preventDefault();
+                }
+            });
+        });
+        eventListenerApplied = true;
+
+    }
+
+    return [appendContextMenuOption, removeContextMenuOption, addContextMenuEventListener];
+};
+
+var initResponse = contextMenuInit();
+var appendContextMenuOption = initResponse[0];
+var removeContextMenuOption = initResponse[1];
+var addContextMenuEventListener = initResponse[2];
+
+let cancelGenerateForever = function() {
+    clearInterval(window.generateOnRepeatInterval);
+};
+
+(function() {
+    //Start example Context Menu Items
+    let generateOnRepeat = function(genbuttonid, interruptbuttonid) {
+        let genbutton = gradioApp().querySelector(genbuttonid);
+        let interruptbutton = gradioApp().querySelector(interruptbuttonid);
+        if (!interruptbutton.offsetParent) {
+            genbutton.click();
+        }
+        clearInterval(window.generateOnRepeatInterval);
+        window.generateOnRepeatInterval = setInterval(function() {
+            if (!interruptbutton.offsetParent) {
+                genbutton.click();
+            }
+        },
+        500);
+    };
+
+    let generateOnRepeatForButtons = function() {
+        generateOnRepeat('#generate_button', '#stop_button');
+    };
+
+    appendContextMenuOption('#generate_button', 'Generate forever', generateOnRepeatForButtons);
+//    appendContextMenuOption('#stop_button', 'Generate forever', generateOnRepeatForButtons);
+
+//    appendContextMenuOption('#stop_button', 'Cancel generate forever', cancelGenerateForever);
+//    appendContextMenuOption('#generate_button', 'Cancel generate forever', cancelGenerateForever);
+})();
+//End example Context Menu Items
+
+document.onreadystatechange = function () {
+    if (document.readyState == "complete") {
+        addContextMenuEventListener();
+    }
+};

javascript/edit-attention.js

@@ -0,0 +1,128 @@
+function updateInput(target) {
+    let e = new Event("input", {bubbles: true});
+    Object.defineProperty(e, "target", {value: target});
+    target.dispatchEvent(e);
+}
+
+function keyupEditAttention(event) {
+    let target = event.originalTarget || event.composedPath()[0];
+    if (!target.matches("*:is([id*='_prompt'], .prompt) textarea")) return;
+    if (!(event.metaKey || event.ctrlKey)) return;
+
+    let isPlus = event.key == "ArrowUp";
+    let isMinus = event.key == "ArrowDown";
+    if (!isPlus && !isMinus) return;
+
+    let selectionStart = target.selectionStart;
+    let selectionEnd = target.selectionEnd;
+    let text = target.value;
+
+    function selectCurrentParenthesisBlock(OPEN, CLOSE) {
+        if (selectionStart !== selectionEnd) return false;
+
+        // Find opening parenthesis around current cursor
+        const before = text.substring(0, selectionStart);
+        let beforeParen = before.lastIndexOf(OPEN);
+        if (beforeParen == -1) return false;
+        let beforeParenClose = before.lastIndexOf(CLOSE);
+        while (beforeParenClose !== -1 && beforeParenClose > beforeParen) {
+            beforeParen = before.lastIndexOf(OPEN, beforeParen - 1);
+            beforeParenClose = before.lastIndexOf(CLOSE, beforeParenClose - 1);
+        }
+
+        // Find closing parenthesis around current cursor
+        const after = text.substring(selectionStart);
+        let afterParen = after.indexOf(CLOSE);
+        if (afterParen == -1) return false;
+        let afterParenOpen = after.indexOf(OPEN);
+        while (afterParenOpen !== -1 && afterParen > afterParenOpen) {
+            afterParen = after.indexOf(CLOSE, afterParen + 1);
+            afterParenOpen = after.indexOf(OPEN, afterParenOpen + 1);
+        }
+        if (beforeParen === -1 || afterParen === -1) return false;
+
+        // Set the selection to the text between the parenthesis
+        const parenContent = text.substring(beforeParen + 1, selectionStart + afterParen);
+        const lastColon = parenContent.lastIndexOf(":");
+        selectionStart = beforeParen + 1;
+        selectionEnd = selectionStart + lastColon;
+        target.setSelectionRange(selectionStart, selectionEnd);
+        return true;
+    }
+
+    function selectCurrentWord() {
+        if (selectionStart !== selectionEnd) return false;
+        const delimiters = ".,\\/!?%^*;:{}=`~() \r\n\t";
+
+        // seek backward until to find beggining
+        while (!delimiters.includes(text[selectionStart - 1]) && selectionStart > 0) {
+            selectionStart--;
+        }
+
+        // seek forward to find end
+        while (!delimiters.includes(text[selectionEnd]) && selectionEnd < text.length) {
+            selectionEnd++;
+        }
+
+        target.setSelectionRange(selectionStart, selectionEnd);
+        return true;
+    }
+
+    // If the user hasn't selected anything, let's select their current parenthesis block or word
+    if (!selectCurrentParenthesisBlock('<', '>') && !selectCurrentParenthesisBlock('(', ')')) {
+        selectCurrentWord();
+    }
+
+    event.preventDefault();
+
+    var closeCharacter = ')';
+    var delta = 0.1;
+
+    if (selectionStart > 0 && text[selectionStart - 1] == '<') {
+        closeCharacter = '>';
+        delta = 0.05;
+    } else if (selectionStart == 0 || text[selectionStart - 1] != "(") {
+
+        // do not include spaces at the end
+        while (selectionEnd > selectionStart && text[selectionEnd - 1] == ' ') {
+            selectionEnd -= 1;
+        }
+        if (selectionStart == selectionEnd) {
+            return;
+        }
+
+        text = text.slice(0, selectionStart) + "(" + text.slice(selectionStart, selectionEnd) + ":1.0)" + text.slice(selectionEnd);
+
+        selectionStart += 1;
+        selectionEnd += 1;
+    }
+
+    var end = text.slice(selectionEnd + 1).indexOf(closeCharacter) + 1;
+    var weight = parseFloat(text.slice(selectionEnd + 1, selectionEnd + 1 + end));
+    if (isNaN(weight)) return;
+
+    weight += isPlus ? delta : -delta;
+    weight = parseFloat(weight.toPrecision(12));
+    if (String(weight).length == 1) weight += ".0";
+
+    if (closeCharacter == ')' && weight == 1) {
+        var endParenPos = text.substring(selectionEnd).indexOf(')');
+        text = text.slice(0, selectionStart - 1) + text.slice(selectionStart, selectionEnd) + text.slice(selectionEnd + endParenPos + 1);
+        selectionStart--;
+        selectionEnd--;
+    } else {
+        text = text.slice(0, selectionEnd + 1) + weight + text.slice(selectionEnd + end);
+    }
+
+    target.focus();
+    target.value = text;
+    target.selectionStart = selectionStart;
+    target.selectionEnd = selectionEnd;
+
+    updateInput(target);
+
+}
+
+addEventListener('keydown', (event) => {
+    keyupEditAttention(event);
+});

javascript/imageviewer.js

@@ -0,0 +1,260 @@
+// From A1111
+
+function closeModal() {
+    gradioApp().getElementById("lightboxModal").style.display = "none";
+}
+
+function showModal(event) {
+    const source = event.target || event.srcElement;
+    const modalImage = gradioApp().getElementById("modalImage");
+    const lb = gradioApp().getElementById("lightboxModal");
+    modalImage.src = source.src;
+    if (modalImage.style.display === 'none') {
+        lb.style.setProperty('background-image', 'url(' + source.src + ')');
+    }
+    lb.style.display = "flex";
+    lb.focus();
+
+    event.stopPropagation();
+}
+
+function negmod(n, m) {
+    return ((n % m) + m) % m;
+}
+
+function updateOnBackgroundChange() {
+    const modalImage = gradioApp().getElementById("modalImage");
+    if (modalImage && modalImage.offsetParent) {
+        let currentButton = selected_gallery_button();
+
+        if (currentButton?.children?.length > 0 && modalImage.src != currentButton.children[0].src) {
+            modalImage.src = currentButton.children[0].src;
+            if (modalImage.style.display === 'none') {
+                const modal = gradioApp().getElementById("lightboxModal");
+                modal.style.setProperty('background-image', `url(${modalImage.src})`);
+            }
+        }
+    }
+}
+
+function all_gallery_buttons() {
+    var allGalleryButtons = gradioApp().querySelectorAll('.image_gallery .thumbnails > .thumbnail-item.thumbnail-small');
+    var visibleGalleryButtons = [];
+    allGalleryButtons.forEach(function(elem) {
+        if (elem.parentElement.offsetParent) {
+            visibleGalleryButtons.push(elem);
+        }
+    });
+    return visibleGalleryButtons;
+}
+
+function selected_gallery_button() {
+    return all_gallery_buttons().find(elem => elem.classList.contains('selected')) ?? null;
+}
+
+function selected_gallery_index() {
+    return all_gallery_buttons().findIndex(elem => elem.classList.contains('selected'));
+}
+
+function modalImageSwitch(offset) {
+    var galleryButtons = all_gallery_buttons();
+
+    if (galleryButtons.length > 1) {
+        var currentButton = selected_gallery_button();
+
+        var result = -1;
+        galleryButtons.forEach(function(v, i) {
+            if (v == currentButton) {
+                result = i;
+            }
+        });
+
+        if (result != -1) {
+            var nextButton = galleryButtons[negmod((result + offset), galleryButtons.length)];
+            nextButton.click();
+            const modalImage = gradioApp().getElementById("modalImage");
+            const modal = gradioApp().getElementById("lightboxModal");
+            modalImage.src = nextButton.children[0].src;
+            if (modalImage.style.display === 'none') {
+                modal.style.setProperty('background-image', `url(${modalImage.src})`);
+            }
+            setTimeout(function() {
+                modal.focus();
+            }, 10);
+        }
+    }
+}
+
+function saveImage() {
+
+}
+
+function modalSaveImage(event) {
+    event.stopPropagation();
+}
+
+function modalNextImage(event) {
+    modalImageSwitch(1);
+    event.stopPropagation();
+}
+
+function modalPrevImage(event) {
+    modalImageSwitch(-1);
+    event.stopPropagation();
+}
+
+function modalKeyHandler(event) {
+    switch (event.key) {
+    case "s":
+        saveImage();
+        break;
+    case "ArrowLeft":
+        modalPrevImage(event);
+        break;
+    case "ArrowRight":
+        modalNextImage(event);
+        break;
+    case "Escape":
+        closeModal();
+        break;
+    }
+}
+
+function setupImageForLightbox(e) {
+    if (e.dataset.modded) {
+        return;
+    }
+
+    e.dataset.modded = true;
+    e.style.cursor = 'pointer';
+    e.style.userSelect = 'none';
+
+    var isFirefox = navigator.userAgent.toLowerCase().indexOf('firefox') > -1;
+
+    // For Firefox, listening on click first switched to next image then shows the lightbox.
+    // If you know how to fix this without switching to mousedown event, please.
+    // For other browsers the event is click to make it possiblr to drag picture.
+    var event = isFirefox ? 'mousedown' : 'click';
+
+    e.addEventListener(event, function(evt) {
+        if (evt.button == 1) {
+            open(evt.target.src);
+            evt.preventDefault();
+            return;
+        }
+        if (evt.button != 0) return;
+
+        modalZoomSet(gradioApp().getElementById('modalImage'), true);
+        evt.preventDefault();
+        showModal(evt);
+    }, true);
+
+}
+
+function modalZoomSet(modalImage, enable) {
+    if (modalImage) modalImage.classList.toggle('modalImageFullscreen', !!enable);
+}
+
+function modalZoomToggle(event) {
+    var modalImage = gradioApp().getElementById("modalImage");
+    modalZoomSet(modalImage, !modalImage.classList.contains('modalImageFullscreen'));
+    event.stopPropagation();
+}
+
+function modalTileImageToggle(event) {
+    const modalImage = gradioApp().getElementById("modalImage");
+    const modal = gradioApp().getElementById("lightboxModal");
+    const isTiling = modalImage.style.display === 'none';
+    if (isTiling) {
+        modalImage.style.display = 'block';
+        modal.style.setProperty('background-image', 'none');
+    } else {
+        modalImage.style.display = 'none';
+        modal.style.setProperty('background-image', `url(${modalImage.src})`);
+    }
+
+    event.stopPropagation();
+}
+
+onAfterUiUpdate(function() {
+    var fullImg_preview = gradioApp().querySelectorAll('.image_gallery > div > img');
+    if (fullImg_preview != null) {
+        fullImg_preview.forEach(setupImageForLightbox);
+    }
+    updateOnBackgroundChange();
+});
+
+document.addEventListener("DOMContentLoaded", function() {
+    //const modalFragment = document.createDocumentFragment();
+    const modal = document.createElement('div');
+    modal.onclick = closeModal;
+    modal.id = "lightboxModal";
+    modal.tabIndex = 0;
+    modal.addEventListener('keydown', modalKeyHandler, true);
+
+    const modalControls = document.createElement('div');
+    modalControls.className = 'modalControls gradio-container';
+    modal.append(modalControls);
+
+    const modalZoom = document.createElement('span');
+    modalZoom.className = 'modalZoom cursor';
+    modalZoom.innerHTML = '⤡';
+    modalZoom.addEventListener('click', modalZoomToggle, true);
+    modalZoom.title = "Toggle zoomed view";
+    modalControls.appendChild(modalZoom);
+
+    // const modalTileImage = document.createElement('span');
+    // modalTileImage.className = 'modalTileImage cursor';
+    // modalTileImage.innerHTML = '⊞';
+    // modalTileImage.addEventListener('click', modalTileImageToggle, true);
+    // modalTileImage.title = "Preview tiling";
+    // modalControls.appendChild(modalTileImage);
+    //
+    // const modalSave = document.createElement("span");
+    // modalSave.className = "modalSave cursor";
+    // modalSave.id = "modal_save";
+    // modalSave.innerHTML = "🖫";
+    // modalSave.addEventListener("click", modalSaveImage, true);
+    // modalSave.title = "Save Image(s)";
+    // modalControls.appendChild(modalSave);
+
+    const modalClose = document.createElement('span');
+    modalClose.className = 'modalClose cursor';
+    modalClose.innerHTML = '×';
+    modalClose.onclick = closeModal;
+    modalClose.title = "Close image viewer";
+    modalControls.appendChild(modalClose);
+
+    const modalImage = document.createElement('img');
+    modalImage.id = 'modalImage';
+    modalImage.onclick = closeModal;
+    modalImage.tabIndex = 0;
+    modalImage.addEventListener('keydown', modalKeyHandler, true);
+    modal.appendChild(modalImage);
+
+    const modalPrev = document.createElement('a');
+    modalPrev.className = 'modalPrev';
+    modalPrev.innerHTML = '❮';
+    modalPrev.tabIndex = 0;
+    modalPrev.addEventListener('click', modalPrevImage, true);
+    modalPrev.addEventListener('keydown', modalKeyHandler, true);
+    modal.appendChild(modalPrev);
+
+    const modalNext = document.createElement('a');
+    modalNext.className = 'modalNext';
+    modalNext.innerHTML = '❯';
+    modalNext.tabIndex = 0;
+    modalNext.addEventListener('click', modalNextImage, true);
+    modalNext.addEventListener('keydown', modalKeyHandler, true);
+
+    modal.appendChild(modalNext);
+
+    try {
+        gradioApp().appendChild(modal);
+    } catch (e) {
+        gradioApp().body.appendChild(modal);
+    }
+
+    document.body.appendChild(modal);
+
+});

javascript/localization.js

@@ -0,0 +1,141 @@
+var re_num = /^[.\d]+$/;
+
+var original_lines = {};
+var translated_lines = {};
+
+function hasLocalization() {
+    return window.localization && Object.keys(window.localization).length > 0;
+}
+
+function textNodesUnder(el) {
+    var n, a = [], walk = document.createTreeWalker(el, NodeFilter.SHOW_TEXT, null, false);
+    while ((n = walk.nextNode())) a.push(n);
+    return a;
+}
+
+function canBeTranslated(node, text) {
+    if (!text) return false;
+    if (!node.parentElement) return false;
+    var parentType = node.parentElement.nodeName;
+    if (parentType == 'SCRIPT' || parentType == 'STYLE' || parentType == 'TEXTAREA') return false;
+    if (re_num.test(text)) return false;
+    return true;
+}
+
+function getTranslation(text) {
+    if (!text) return undefined;
+
+    if (translated_lines[text] === undefined) {
+        original_lines[text] = 1;
+    }
+
+    var tl = localization[text];
+    if (tl !== undefined) {
+        translated_lines[tl] = 1;
+    }
+
+    return tl;
+}
+
+function processTextNode(node) {
+    var text = node.textContent.trim();
+
+    if (!canBeTranslated(node, text)) return;
+
+    var tl = getTranslation(text);
+    if (tl !== undefined) {
+        node.textContent = tl;
+    }
+}
+
+function processNode(node) {
+    if (node.nodeType == 3) {
+        processTextNode(node);
+        return;
+    }
+
+    if (node.title) {
+        let tl = getTranslation(node.title);
+        if (tl !== undefined) {
+            node.title = tl;
+        }
+    }
+
+    if (node.placeholder) {
+        let tl = getTranslation(node.placeholder);
+        if (tl !== undefined) {
+            node.placeholder = tl;
+        }
+    }
+
+    textNodesUnder(node).forEach(function(node) {
+        processTextNode(node);
+    });
+}
+
+function refresh_style_localization() {
+    processNode(document.querySelector('.style_selections'));
+}
+
+function localizeWholePage() {
+    processNode(gradioApp());
+
+    function elem(comp) {
+        var elem_id = comp.props.elem_id ? comp.props.elem_id : "component-" + comp.id;
+        return gradioApp().getElementById(elem_id);
+    }
+
+    for (var comp of window.gradio_config.components) {
+        if (comp.props.webui_tooltip) {
+            let e = elem(comp);
+
+            let tl = e ? getTranslation(e.title) : undefined;
+            if (tl !== undefined) {
+                e.title = tl;
+            }
+        }
+        if (comp.props.placeholder) {
+            let e = elem(comp);
+            let textbox = e ? e.querySelector('[placeholder]') : null;
+
+            let tl = textbox ? getTranslation(textbox.placeholder) : undefined;
+            if (tl !== undefined) {
+                textbox.placeholder = tl;
+            }
+        }
+    }
+}
+
+document.addEventListener("DOMContentLoaded", function() {
+    if (!hasLocalization()) {
+        return;
+    }
+
+    onUiUpdate(function(m) {
+        m.forEach(function(mutation) {
+            mutation.addedNodes.forEach(function(node) {
+                processNode(node);
+            });
+        });
+    });
+
+    localizeWholePage();
+
+    if (localization.rtl) { // if the language is from right to left,
+        (new MutationObserver((mutations, observer) => { // wait for the style to load
+            mutations.forEach(mutation => {
+                mutation.addedNodes.forEach(node => {
+                    if (node.tagName === 'STYLE') {
+                        observer.disconnect();
+
+                        for (const x of node.sheet.rules) { // find all rtl media rules
+                            if (Array.from(x.media || []).includes('rtl')) {
+                                x.media.appendMedium('all'); // enable them
+                            }
+                        }
+                    }
+                });
+            });
+        })).observe(gradioApp(), {childList: true});
+    }
+});

javascript/script.js

@@ -0,0 +1,180 @@
+// based on https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob/v1.6.0/script.js
+function gradioApp() {
+    const elems = document.getElementsByTagName('gradio-app');
+    const elem = elems.length == 0 ? document : elems[0];
+
+    if (elem !== document) {
+        elem.getElementById = function(id) {
+            return document.getElementById(id);
+        };
+    }
+    return elem.shadowRoot ? elem.shadowRoot : elem;
+}
+
+/**
+ * Get the currently selected top-level UI tab button (e.g. the button that says "Extras").
+ */
+function get_uiCurrentTab() {
+    return gradioApp().querySelector('#tabs > .tab-nav > button.selected');
+}
+
+/**
+ * Get the first currently visible top-level UI tab content (e.g. the div hosting the "txt2img" UI).
+ */
+function get_uiCurrentTabContent() {
+    return gradioApp().querySelector('#tabs > .tabitem[id^=tab_]:not([style*="display: none"])');
+}
+
+var uiUpdateCallbacks = [];
+var uiAfterUpdateCallbacks = [];
+var uiLoadedCallbacks = [];
+var uiTabChangeCallbacks = [];
+var optionsChangedCallbacks = [];
+var uiAfterUpdateTimeout = null;
+var uiCurrentTab = null;
+
+/**
+ * Register callback to be called at each UI update.
+ * The callback receives an array of MutationRecords as an argument.
+ */
+function onUiUpdate(callback) {
+    uiUpdateCallbacks.push(callback);
+}
+
+/**
+ * Register callback to be called soon after UI updates.
+ * The callback receives no arguments.
+ *
+ * This is preferred over `onUiUpdate` if you don't need
+ * access to the MutationRecords, as your function will
+ * not be called quite as often.
+ */
+function onAfterUiUpdate(callback) {
+    uiAfterUpdateCallbacks.push(callback);
+}
+
+/**
+ * Register callback to be called when the UI is loaded.
+ * The callback receives no arguments.
+ */
+function onUiLoaded(callback) {
+    uiLoadedCallbacks.push(callback);
+}
+
+/**
+ * Register callback to be called when the UI tab is changed.
+ * The callback receives no arguments.
+ */
+function onUiTabChange(callback) {
+    uiTabChangeCallbacks.push(callback);
+}
+
+/**
+ * Register callback to be called when the options are changed.
+ * The callback receives no arguments.
+ * @param callback
+ */
+function onOptionsChanged(callback) {
+    optionsChangedCallbacks.push(callback);
+}
+
+function executeCallbacks(queue, arg) {
+    for (const callback of queue) {
+        try {
+            callback(arg);
+        } catch (e) {
+            console.error("error running callback", callback, ":", e);
+        }
+    }
+}
+
+/**
+ * Schedule the execution of the callbacks registered with onAfterUiUpdate.
+ * The callbacks are executed after a short while, unless another call to this function
+ * is made before that time. IOW, the callbacks are executed only once, even
+ * when there are multiple mutations observed.
+ */
+function scheduleAfterUiUpdateCallbacks() {
+    clearTimeout(uiAfterUpdateTimeout);
+    uiAfterUpdateTimeout = setTimeout(function() {
+        executeCallbacks(uiAfterUpdateCallbacks);
+    }, 200);
+}
+
+var executedOnLoaded = false;
+
+document.addEventListener("DOMContentLoaded", function() {
+    var mutationObserver = new MutationObserver(function(m) {
+        if (!executedOnLoaded && gradioApp().querySelector('#generate_button')) {
+            executedOnLoaded = true;
+            executeCallbacks(uiLoadedCallbacks);
+        }
+
+        executeCallbacks(uiUpdateCallbacks, m);
+        scheduleAfterUiUpdateCallbacks();
+        const newTab = get_uiCurrentTab();
+        if (newTab && (newTab !== uiCurrentTab)) {
+            uiCurrentTab = newTab;
+            executeCallbacks(uiTabChangeCallbacks);
+        }
+    });
+    mutationObserver.observe(gradioApp(), {childList: true, subtree: true});
+});
+
+/**
+ * Add a ctrl+enter as a shortcut to start a generation
+ */
+document.addEventListener('keydown', function(e) {
+    const isModifierKey = (e.metaKey || e.ctrlKey || e.altKey);
+    const isEnterKey = (e.key == "Enter" || e.keyCode == 13);
+
+    if(isModifierKey && isEnterKey) {
+        const generateButton = gradioApp().querySelector('button:not(.hidden)[id=generate_button]');
+        if (generateButton) {
+            generateButton.click();
+            e.preventDefault();
+            return;
+        }
+
+        const stopButton = gradioApp().querySelector('button:not(.hidden)[id=stop_button]')
+        if(stopButton) {
+            stopButton.click();
+            e.preventDefault();
+            return;
+        }
+    }
+});
+
+/**
+ * checks that a UI element is not in another hidden element or tab content
+ */
+function uiElementIsVisible(el) {
+    if (el === document) {
+        return true;
+    }
+
+    const computedStyle = getComputedStyle(el);
+    const isVisible = computedStyle.display !== 'none';
+
+    if (!isVisible) return false;
+    return uiElementIsVisible(el.parentNode);
+}
+
+function uiElementInSight(el) {
+    const clRect = el.getBoundingClientRect();
+    const windowHeight = window.innerHeight;
+    const isOnScreen = clRect.bottom > 0 && clRect.top < windowHeight;
+
+    return isOnScreen;
+}
+
+function playNotification() {
+    gradioApp().querySelector('#audio_notification audio')?.play();
+}
+
+function set_theme(theme) {
+    var gradioURL = window.location.href;
+    if (!gradioURL.includes('?__theme=')) {
+        window.location.replace(gradioURL + '?__theme=' + theme);
+    }
+}

javascript/viewer.js

@@ -0,0 +1,88 @@
+window.main_viewer_height = 512;
+
+function refresh_grid() {
+    let gridContainer = document.querySelector('#final_gallery .grid-container');
+    let final_gallery = document.getElementById('final_gallery');
+
+    if (gridContainer) if (final_gallery) {
+        let rect = final_gallery.getBoundingClientRect();
+        let cols = Math.ceil((rect.width - 16.0) / rect.height);
+        if (cols < 2) cols = 2;
+        gridContainer.style.setProperty('--grid-cols', cols);
+    }
+}
+
+function refresh_grid_delayed() {
+    refresh_grid();
+    setTimeout(refresh_grid, 100);
+    setTimeout(refresh_grid, 500);
+    setTimeout(refresh_grid, 1000);
+}
+
+function resized() {
+    let windowHeight = window.innerHeight - 260;
+    let elements = document.getElementsByClassName('main_view');
+
+    if (windowHeight > 745) windowHeight = 745;
+
+    for (let i = 0; i < elements.length; i++) {
+        elements[i].style.height = windowHeight + 'px';
+    }
+
+    window.main_viewer_height = windowHeight;
+
+    refresh_grid();
+}
+
+function viewer_to_top(delay = 100) {
+    setTimeout(() => window.scrollTo({top: 0, behavior: 'smooth'}), delay);
+}
+
+function viewer_to_bottom(delay = 100) {
+    let element = document.getElementById('positive_prompt');
+    let yPos = window.main_viewer_height;
+
+    if (element) {
+        yPos = element.getBoundingClientRect().top + window.scrollY;
+    }
+
+    setTimeout(() => window.scrollTo({top: yPos - 8, behavior: 'smooth'}), delay);
+}
+
+window.addEventListener('resize', (e) => {
+    resized();
+});
+
+onUiLoaded(async () => {
+    resized();
+});
+
+function on_style_selection_blur() {
+    let target = document.querySelector("#gradio_receiver_style_selections textarea");
+    target.value = "on_style_selection_blur " + Math.random();
+    let e = new Event("input", {bubbles: true})
+    Object.defineProperty(e, "target", {value: target})
+    target.dispatchEvent(e);
+}
+
+onUiLoaded(async () => {
+    let spans = document.querySelectorAll('.aspect_ratios span');
+
+    spans.forEach(function (span) {
+        span.innerHTML = span.innerHTML.replace(/&lt;/g, '<').replace(/&gt;/g, '>');
+    });
+
+    document.querySelector('.style_selections').addEventListener('focusout', function (event) {
+        setTimeout(() => {
+            if (!this.contains(document.activeElement)) {
+                on_style_selection_blur();
+            }
+        }, 200);
+    });
+
+    let inputs = document.querySelectorAll('.lora_weight input[type="range"]');
+
+    inputs.forEach(function (input) {
+        input.style.marginTop = '12px';
+    });
+});

javascript/zoom.js

@@ -0,0 +1,645 @@
+onUiLoaded(async() => {
+    // Helper functions
+
+    // Detect whether the element has a horizontal scroll bar
+    function hasHorizontalScrollbar(element) {
+        return element.scrollWidth > element.clientWidth;
+    }
+
+    // Function for defining the "Ctrl", "Shift" and "Alt" keys
+    function isModifierKey(event, key) {
+        switch (key) {
+        case "Ctrl":
+            return event.ctrlKey;
+        case "Shift":
+            return event.shiftKey;
+        case "Alt":
+            return event.altKey;
+        default:
+            return false;
+        }
+    }
+
+    // Create hotkey configuration with the provided options
+    function createHotkeyConfig(defaultHotkeysConfig) {
+        const result = {}; // Resulting hotkey configuration
+        for (const key in defaultHotkeysConfig) {
+            result[key] = defaultHotkeysConfig[key];
+        }
+        return result;
+    }
+
+    // Default config
+    const defaultHotkeysConfig = {
+        canvas_hotkey_zoom: "Shift",
+        canvas_hotkey_adjust: "Ctrl",
+        canvas_zoom_undo_extra_key: "Ctrl",
+        canvas_zoom_hotkey_undo: "KeyZ",
+        canvas_hotkey_reset: "KeyR",
+        canvas_hotkey_fullscreen: "KeyS",
+        canvas_hotkey_move: "KeyF",
+        canvas_show_tooltip: true,
+        canvas_auto_expand: true,
+        canvas_blur_prompt: true,
+    };
+
+    // Loading the configuration from opts
+    const hotkeysConfig = createHotkeyConfig(
+        defaultHotkeysConfig
+    );
+
+    let isMoving = false;
+    let activeElement;
+
+    const elemData = {};
+
+    function applyZoomAndPan(elemId) {
+        const targetElement = gradioApp().querySelector(elemId);
+
+        if (!targetElement) {
+            console.log("Element not found");
+            return;
+        }
+
+        targetElement.style.transformOrigin = "0 0";
+
+        elemData[elemId] = {
+            zoom: 1,
+            panX: 0,
+            panY: 0
+        };
+
+        let fullScreenMode = false;
+
+        // Create tooltip
+        function createTooltip() {
+            const toolTipElemnt =
+                targetElement.querySelector(".image-container");
+            const tooltip = document.createElement("div");
+            tooltip.className = "canvas-tooltip";
+
+            // Creating an item of information
+            const info = document.createElement("i");
+            info.className = "canvas-tooltip-info";
+            info.textContent = "";
+
+            // Create a container for the contents of the tooltip
+            const tooltipContent = document.createElement("div");
+            tooltipContent.className = "canvas-tooltip-content";
+
+            // Define an array with hotkey information and their actions
+            const hotkeysInfo = [
+                {
+                    configKey: "canvas_hotkey_zoom",
+                    action: "Zoom canvas",
+                    keySuffix: " + wheel"
+                },
+                {
+                    configKey: "canvas_hotkey_adjust",
+                    action: "Adjust brush size",
+                    keySuffix: " + wheel"
+                },
+                {configKey: "canvas_zoom_hotkey_undo", action: "Undo last action", keyPrefix: `${hotkeysConfig.canvas_zoom_undo_extra_key} + ` },
+                {configKey: "canvas_hotkey_reset", action: "Reset zoom"},
+                {
+                    configKey: "canvas_hotkey_fullscreen",
+                    action: "Fullscreen mode"
+                },
+                {configKey: "canvas_hotkey_move", action: "Move canvas"}
+            ];
+
+            // Create hotkeys array based on the config values
+            const hotkeys = hotkeysInfo.map((info) => {
+                const configValue = hotkeysConfig[info.configKey];
+        
+                let key = configValue.slice(-1);
+        
+                if (info.keySuffix) {
+                  key = `${configValue}${info.keySuffix}`;
+                }
+        
+                if (info.keyPrefix && info.keyPrefix !== "None + ") {
+                  key = `${info.keyPrefix}${configValue[3]}`;
+                }
+        
+                return {
+                  key,
+                  action: info.action,
+                };
+              });
+        
+              hotkeys
+                .forEach(hotkey => {
+                  const p = document.createElement("p");
+                  p.innerHTML = `<b>${hotkey.key}</b> - ${hotkey.action}`;
+                  tooltipContent.appendChild(p);
+                });
+        
+              tooltip.append(info, tooltipContent);
+
+              // Add a hint element to the target element
+              toolTipElemnt.appendChild(tooltip);
+        }
+
+        //Show tool tip if setting enable
+        if (hotkeysConfig.canvas_show_tooltip) {
+            createTooltip();
+        }
+
+        // Reset the zoom level and pan position of the target element to their initial values
+        function resetZoom() {
+            elemData[elemId] = {
+                zoomLevel: 1,
+                panX: 0,
+                panY: 0
+            };
+
+            targetElement.style.overflow = "hidden";
+
+            targetElement.isZoomed = false;
+
+            targetElement.style.transform = `scale(${elemData[elemId].zoomLevel}) translate(${elemData[elemId].panX}px, ${elemData[elemId].panY}px)`;
+
+            const canvas = gradioApp().querySelector(
+                `${elemId} canvas[key="interface"]`
+            );
+
+            toggleOverlap("off");
+            fullScreenMode = false;
+
+            const closeBtn = targetElement.querySelector("button[aria-label='Remove Image']");
+            if (closeBtn) {
+                closeBtn.addEventListener("click", resetZoom);
+            }
+
+            if (canvas) {
+                const parentElement = targetElement.closest('[id^="component-"]');
+                if (
+                    canvas &&
+                    parseFloat(canvas.style.width) > parentElement.offsetWidth &&
+                    parseFloat(targetElement.style.width) > parentElement.offsetWidth
+                ) {
+                    fitToElement();
+                    return;
+                }
+
+            }
+
+            targetElement.style.width = "";
+        }
+
+        // Toggle the zIndex of the target element between two values, allowing it to overlap or be overlapped by other elements
+        function toggleOverlap(forced = "") {
+            const zIndex1 = "0";
+            const zIndex2 = "998";
+
+            targetElement.style.zIndex =
+                targetElement.style.zIndex !== zIndex2 ? zIndex2 : zIndex1;
+
+            if (forced === "off") {
+                targetElement.style.zIndex = zIndex1;
+            } else if (forced === "on") {
+                targetElement.style.zIndex = zIndex2;
+            }
+        }
+
+        // Adjust the brush size based on the deltaY value from a mouse wheel event
+        function adjustBrushSize(
+            elemId,
+            deltaY,
+            withoutValue = false,
+            percentage = 5
+        ) {
+            const input =
+                gradioApp().querySelector(
+                    `${elemId} input[aria-label='Brush radius']`
+                ) ||
+                gradioApp().querySelector(
+                    `${elemId} button[aria-label="Use brush"]`
+                );
+
+            if (input) {
+                input.click();
+                if (!withoutValue) {
+                    const maxValue =
+                        parseFloat(input.getAttribute("max")) || 100;
+                    const changeAmount = maxValue * (percentage / 100);
+                    const newValue =
+                        parseFloat(input.value) +
+                        (deltaY > 0 ? -changeAmount : changeAmount);
+                    input.value = Math.min(Math.max(newValue, 0), maxValue);
+                    input.dispatchEvent(new Event("change"));
+                }
+            }
+        }
+
+        // Reset zoom when uploading a new image
+        const fileInput = gradioApp().querySelector(
+            `${elemId} input[type="file"][accept="image/*"].svelte-116rqfv`
+        );
+        fileInput.addEventListener("click", resetZoom);
+
+        // Update the zoom level and pan position of the target element based on the values of the zoomLevel, panX and panY variables
+        function updateZoom(newZoomLevel, mouseX, mouseY) {
+            newZoomLevel = Math.max(0.1, Math.min(newZoomLevel, 15));
+
+            elemData[elemId].panX +=
+                mouseX - (mouseX * newZoomLevel) / elemData[elemId].zoomLevel;
+            elemData[elemId].panY +=
+                mouseY - (mouseY * newZoomLevel) / elemData[elemId].zoomLevel;
+
+            targetElement.style.transformOrigin = "0 0";
+            targetElement.style.transform = `translate(${elemData[elemId].panX}px, ${elemData[elemId].panY}px) scale(${newZoomLevel})`;
+            targetElement.style.overflow = "visible";
+
+            toggleOverlap("on");
+ 
+            return newZoomLevel;
+        }
+
+        // Change the zoom level based on user interaction
+        function changeZoomLevel(operation, e) {
+            if (isModifierKey(e, hotkeysConfig.canvas_hotkey_zoom)) {
+                e.preventDefault();
+
+                let zoomPosX, zoomPosY;
+                let delta = 0.2;
+
+                if (elemData[elemId].zoomLevel > 7) {
+                    delta = 0.9;
+                } else if (elemData[elemId].zoomLevel > 2) {
+                    delta = 0.6;
+                }
+
+                zoomPosX = e.clientX;
+                zoomPosY = e.clientY;
+
+                fullScreenMode = false;
+                elemData[elemId].zoomLevel = updateZoom(
+                    elemData[elemId].zoomLevel +
+                    (operation === "+" ? delta : -delta),
+                    zoomPosX - targetElement.getBoundingClientRect().left,
+                    zoomPosY - targetElement.getBoundingClientRect().top
+                );
+
+                targetElement.isZoomed = true;
+            }
+        }
+
+        /**
+         * This function fits the target element to the screen by calculating
+         * the required scale and offsets. It also updates the global variables
+         * zoomLevel, panX, and panY to reflect the new state.
+         */
+
+        function fitToElement() {
+            //Reset Zoom
+            targetElement.style.transform = `translate(${0}px, ${0}px) scale(${1})`;
+
+            let parentElement;
+
+            parentElement = targetElement.closest('[id^="component-"]');
+
+            // Get element and screen dimensions
+            const elementWidth = targetElement.offsetWidth;
+            const elementHeight = targetElement.offsetHeight;
+
+            const screenWidth = parentElement.clientWidth - 24;
+            const screenHeight = parentElement.clientHeight;
+
+            // Calculate scale and offsets
+            const scaleX = screenWidth / elementWidth;
+            const scaleY = screenHeight / elementHeight;
+            const scale = Math.min(scaleX, scaleY);
+
+            const offsetX =0;
+            const offsetY =0;
+
+            // Apply scale and offsets to the element
+            targetElement.style.transform = `translate(${offsetX}px, ${offsetY}px) scale(${scale})`;
+
+            // Update global variables
+            elemData[elemId].zoomLevel = scale;
+            elemData[elemId].panX = offsetX;
+            elemData[elemId].panY = offsetY;
+
+            fullScreenMode = false;
+            toggleOverlap("off");
+        }
+
+        // Undo last action
+        function undoLastAction(e) {
+            let isCtrlPressed = isModifierKey(e, hotkeysConfig.canvas_zoom_undo_extra_key)
+            const isAuxButton = e.button >= 3;
+            
+            if (isAuxButton) {
+              isCtrlPressed = true
+            } else {
+              if (!isModifierKey(e, hotkeysConfig.canvas_zoom_undo_extra_key)) return;
+            }
+
+            // Move undoBtn query outside the if statement to avoid unnecessary queries
+            const undoBtn = document.querySelector(`${activeElement} button[aria-label="Undo"]`);
+        
+            if ((isCtrlPressed) && undoBtn ) {
+                e.preventDefault();
+                undoBtn.click();
+            }
+        }
+
+        /**
+         * This function fits the target element to the screen by calculating
+         * the required scale and offsets. It also updates the global variables
+         * zoomLevel, panX, and panY to reflect the new state.
+         */
+
+        // Fullscreen mode
+        function fitToScreen() {
+            const canvas = gradioApp().querySelector(
+                `${elemId} canvas[key="interface"]`
+            );
+
+            if (!canvas) return;
+
+            targetElement.style.width = (canvas.offsetWidth + 2) + "px";
+            targetElement.style.overflow = "visible";
+
+            if (fullScreenMode) {
+                resetZoom();
+                fullScreenMode = false;
+                return;
+            }
+
+            //Reset Zoom
+            targetElement.style.transform = `translate(${0}px, ${0}px) scale(${1})`;
+
+            // Get scrollbar width to right-align the image
+            const scrollbarWidth =
+                window.innerWidth - document.documentElement.clientWidth;
+
+            // Get element and screen dimensions
+            const elementWidth = targetElement.offsetWidth;
+            const elementHeight = targetElement.offsetHeight;
+            const screenWidth = window.innerWidth - scrollbarWidth;
+            const screenHeight = window.innerHeight;
+
+            // Get element's coordinates relative to the page
+            const elementRect = targetElement.getBoundingClientRect();
+            const elementY = elementRect.y;
+            const elementX = elementRect.x;
+
+            // Calculate scale and offsets
+            const scaleX = screenWidth / elementWidth;
+            const scaleY = screenHeight / elementHeight;
+            const scale = Math.min(scaleX, scaleY);
+
+            // Get the current transformOrigin
+            const computedStyle = window.getComputedStyle(targetElement);
+            const transformOrigin = computedStyle.transformOrigin;
+            const [originX, originY] = transformOrigin.split(" ");
+            const originXValue = parseFloat(originX);
+            const originYValue = parseFloat(originY);
+
+            // Calculate offsets with respect to the transformOrigin
+            const offsetX =
+                (screenWidth - elementWidth * scale) / 2 -
+                elementX -
+                originXValue * (1 - scale);
+            const offsetY =
+                (screenHeight - elementHeight * scale) / 2 -
+                elementY -
+                originYValue * (1 - scale);
+
+            // Apply scale and offsets to the element
+            targetElement.style.transform = `translate(${offsetX}px, ${offsetY}px) scale(${scale})`;
+
+            // Update global variables
+            elemData[elemId].zoomLevel = scale;
+            elemData[elemId].panX = offsetX;
+            elemData[elemId].panY = offsetY;
+
+            fullScreenMode = true;
+            toggleOverlap("on");
+        }
+
+        // Handle keydown events
+        function handleKeyDown(event) {
+            // Disable key locks to make pasting from the buffer work correctly
+            if ((event.ctrlKey && event.code === 'KeyV') || (event.ctrlKey && event.code === 'KeyC') || event.code === "F5") {
+                return;
+            }
+
+            // before activating shortcut, ensure user is not actively typing in an input field
+            if (!hotkeysConfig.canvas_blur_prompt) {
+                if (event.target.nodeName === 'TEXTAREA' || event.target.nodeName === 'INPUT') {
+                    return;
+                }
+            }
+
+            const hotkeyActions = {
+                [hotkeysConfig.canvas_hotkey_reset]: resetZoom,
+                [hotkeysConfig.canvas_hotkey_overlap]: toggleOverlap,
+                [hotkeysConfig.canvas_hotkey_fullscreen]: fitToScreen,
+                [hotkeysConfig.canvas_zoom_hotkey_undo]: undoLastAction,
+            };
+
+            const action = hotkeyActions[event.code];
+            if (action) {
+                event.preventDefault();
+                action(event);
+            }
+
+            if (
+                isModifierKey(event, hotkeysConfig.canvas_hotkey_zoom) ||
+                isModifierKey(event, hotkeysConfig.canvas_hotkey_adjust)
+            ) {
+                event.preventDefault();
+            }
+        }
+
+        // Get Mouse position
+        function getMousePosition(e) {
+            mouseX = e.offsetX;
+            mouseY = e.offsetY;
+        }
+
+        // Simulation of the function to put a long image into the screen.
+        // We detect if an image has a scroll bar or not, make a fullscreen to reveal the image, then reduce it to fit into the element.
+        // We hide the image and show it to the user when it is ready.
+
+        targetElement.isExpanded = false;
+        function autoExpand() {
+            const canvas = document.querySelector(`${elemId} canvas[key="interface"]`);
+            if (canvas) {
+                if (hasHorizontalScrollbar(targetElement) && targetElement.isExpanded === false) {
+                    targetElement.style.visibility = "hidden";
+                    setTimeout(() => {
+                        fitToScreen();
+                        resetZoom();
+                        targetElement.style.visibility = "visible";
+                        targetElement.isExpanded = true;
+                    }, 10);
+                }
+            }
+        }
+
+        targetElement.addEventListener("mousemove", getMousePosition);
+        targetElement.addEventListener("auxclick", undoLastAction);
+
+        //observers
+        // Creating an observer with a callback function to handle DOM changes
+        const observer = new MutationObserver((mutationsList, observer) => {
+            for (let mutation of mutationsList) {
+              // If the style attribute of the canvas has changed, by observation it happens only when the picture changes
+              if (mutation.type === 'attributes' && mutation.attributeName === 'style' &&
+                mutation.target.tagName.toLowerCase() === 'canvas') {
+                targetElement.isExpanded = false;
+                setTimeout(resetZoom, 10);
+              }
+            }
+          });
+      
+          // Apply auto expand if enabled
+          if (hotkeysConfig.canvas_auto_expand) {
+            targetElement.addEventListener("mousemove", autoExpand);
+            // Set up an observer to track attribute changes
+            observer.observe(targetElement, { attributes: true, childList: true, subtree: true });
+          }
+
+        // Handle events only inside the targetElement
+        let isKeyDownHandlerAttached = false;
+
+        function handleMouseMove() {
+            if (!isKeyDownHandlerAttached) {
+                document.addEventListener("keydown", handleKeyDown);
+                isKeyDownHandlerAttached = true;
+
+                activeElement = elemId;
+            }
+        }
+
+        function handleMouseLeave() {
+            if (isKeyDownHandlerAttached) {
+                document.removeEventListener("keydown", handleKeyDown);
+                isKeyDownHandlerAttached = false;
+
+                activeElement = null;
+            }
+        }
+
+        // Add mouse event handlers
+        targetElement.addEventListener("mousemove", handleMouseMove);
+        targetElement.addEventListener("mouseleave", handleMouseLeave);
+
+        targetElement.addEventListener("wheel", e => {
+            // change zoom level
+            const operation = e.deltaY > 0 ? "-" : "+";
+            changeZoomLevel(operation, e);
+
+            // Handle brush size adjustment with ctrl key pressed
+            if (isModifierKey(e, hotkeysConfig.canvas_hotkey_adjust)) {
+                e.preventDefault();
+
+                // Increase or decrease brush size based on scroll direction
+                adjustBrushSize(elemId, e.deltaY);
+            }
+        });
+
+        // Handle the move event for pan functionality. Updates the panX and panY variables and applies the new transform to the target element.
+        function handleMoveKeyDown(e) {
+
+            // Disable key locks to make pasting from the buffer work correctly
+            if ((e.ctrlKey && e.code === 'KeyV') || (e.ctrlKey && e.code === 'KeyC') || e.code === "F5") {
+                return;
+            }
+
+            // before activating shortcut, ensure user is not actively typing in an input field
+            if (!hotkeysConfig.canvas_blur_prompt) {
+                if (e.target.nodeName === 'TEXTAREA' || e.target.nodeName === 'INPUT') {
+                    return;
+                }
+            }
+
+
+            if (e.code === hotkeysConfig.canvas_hotkey_move) {
+                if (!e.ctrlKey && !e.metaKey && isKeyDownHandlerAttached) {
+                    e.preventDefault();
+                    document.activeElement.blur();
+                    isMoving = true;
+                }
+            }
+        }
+
+        function handleMoveKeyUp(e) {
+            if (e.code === hotkeysConfig.canvas_hotkey_move) {
+                isMoving = false;
+            }
+        }
+
+        document.addEventListener("keydown", handleMoveKeyDown);
+        document.addEventListener("keyup", handleMoveKeyUp);
+
+        // Detect zoom level and update the pan speed.
+        function updatePanPosition(movementX, movementY) {
+            let panSpeed = 2;
+
+            if (elemData[elemId].zoomLevel > 8) {
+                panSpeed = 3.5;
+            }
+
+            elemData[elemId].panX += movementX * panSpeed;
+            elemData[elemId].panY += movementY * panSpeed;
+
+            // Delayed redraw of an element
+            requestAnimationFrame(() => {
+                targetElement.style.transform = `translate(${elemData[elemId].panX}px, ${elemData[elemId].panY}px) scale(${elemData[elemId].zoomLevel})`;
+                toggleOverlap("on");
+            });
+        }
+
+        function handleMoveByKey(e) {
+            if (isMoving && elemId === activeElement) {
+                updatePanPosition(e.movementX, e.movementY);
+                targetElement.style.pointerEvents = "none";
+                targetElement.style.overflow = "visible";
+            } else {
+                targetElement.style.pointerEvents = "auto";
+            }
+        }
+
+        // Prevents sticking to the mouse
+        window.onblur = function() {
+            isMoving = false;
+        };
+
+        // Checks for extension
+        function checkForOutBox() {
+            const parentElement = targetElement.closest('[id^="component-"]');
+            if (parentElement.offsetWidth < targetElement.offsetWidth && !targetElement.isExpanded) {
+                resetZoom();
+                targetElement.isExpanded = true;
+            }
+
+            if (parentElement.offsetWidth < targetElement.offsetWidth && elemData[elemId].zoomLevel == 1) {
+                resetZoom();
+            }
+
+            if (parentElement.offsetWidth < targetElement.offsetWidth && targetElement.offsetWidth * elemData[elemId].zoomLevel > parentElement.offsetWidth && elemData[elemId].zoomLevel < 1 && !targetElement.isZoomed) {
+                resetZoom();
+            }
+        }
+
+        targetElement.addEventListener("mousemove", checkForOutBox);
+
+        window.addEventListener('resize', (e) => {
+            resetZoom();
+
+            targetElement.isExpanded = false;
+            targetElement.isZoomed = false;
+        });
+
+        gradioApp().addEventListener("mousemove", handleMoveByKey);
+    }
+
+    applyZoomAndPan("#inpaint_canvas");
+});

language/en.json

@@ -0,0 +1,372 @@
+{
+    "Preview": "Preview",
+    "Gallery": "Gallery",
+    "Generate": "Generate",
+    "Skip": "Skip",
+    "Stop": "Stop",
+    "Input Image": "Input Image",
+    "Advanced": "Advanced",
+    "Upscale or Variation": "Upscale or Variation",
+    "Image Prompt": "Image Prompt",
+    "Inpaint or Outpaint (beta)": "Inpaint or Outpaint (beta)",
+    "Drag above image to here": "Drag above image to here",
+    "Upscale or Variation:": "Upscale or Variation:",
+    "Disabled": "Disabled",
+    "Vary (Subtle)": "Vary (Subtle)",
+    "Vary (Strong)": "Vary (Strong)",
+    "Upscale (1.5x)": "Upscale (1.5x)",
+    "Upscale (2x)": "Upscale (2x)",
+    "Upscale (Fast 2x)": "Upscale (Fast 2x)",
+    "\ud83d\udcd4 Document": "\uD83D\uDCD4 Document",
+    "Image": "Image",
+    "Stop At": "Stop At",
+    "Weight": "Weight",
+    "Type": "Type",
+    "PyraCanny": "PyraCanny",
+    "CPDS": "CPDS",
+    "* \"Image Prompt\" is powered by Fooocus Image Mixture Engine (v1.0.1).": "* \"Image Prompt\" is powered by Fooocus Image Mixture Engine (v1.0.1).",
+    "The scaler multiplied to positive ADM (use 1.0 to disable).": "The scaler multiplied to positive ADM (use 1.0 to disable).",
+    "The scaler multiplied to negative ADM (use 1.0 to disable).": "The scaler multiplied to negative ADM (use 1.0 to disable).",
+    "When to end the guidance from positive/negative ADM.": "When to end the guidance from positive/negative ADM.",
+    "Similar to the Control Mode in A1111 (use 0.0 to disable).": "Similar to the Control Mode in A1111 (use 0.0 to disable).",
+    "Outpaint Expansion (": "Outpaint Expansion (",
+    "Outpaint": "Outpaint",
+    "Left": "Left",
+    "Right": "Right",
+    "Top": "Top",
+    "Bottom": "Bottom",
+    "* \"Inpaint or Outpaint\" is powered by the sampler \"DPMPP Fooocus Seamless 2M SDE Karras Inpaint Sampler\" (beta)": "* \"Inpaint or Outpaint\" is powered by the sampler \"DPMPP Fooocus Seamless 2M SDE Karras Inpaint Sampler\" (beta)",
+    "Setting": "Setting",
+    "Style": "Style",
+    "Performance": "Performance",
+    "Speed": "Speed",
+    "Quality": "Quality",
+    "Aspect Ratios": "Aspect Ratios",
+    "width \u00d7 height": "width \u00d7 height",
+    "Image Number": "Image Number",
+    "Negative Prompt": "Negative Prompt",
+    "Describing what you do not want to see.": "Describing what you do not want to see.",
+    "Random": "Random",
+    "Seed": "Seed",
+    "\ud83d\udcda History Log": "\uD83D\uDCDA History Log",
+    "Image Style": "Image Style",
+    "Fooocus V2": "Fooocus V2",
+    "Default (Slightly Cinematic)": "Default (Slightly Cinematic)",
+    "Fooocus Masterpiece": "Fooocus Masterpiece",
+    "Fooocus Photograph": "Fooocus Photograph",
+    "Fooocus Negative": "Fooocus Negative",
+    "SAI 3D Model": "SAI 3D Model",
+    "SAI Analog Film": "SAI Analog Film",
+    "SAI Anime": "SAI Anime",
+    "SAI Cinematic": "SAI Cinematic",
+    "SAI Comic Book": "SAI Comic Book",
+    "SAI Craft Clay": "SAI Craft Clay",
+    "SAI Digital Art": "SAI Digital Art",
+    "SAI Enhance": "SAI Enhance",
+    "SAI Fantasy Art": "SAI Fantasy Art",
+    "SAI Isometric": "SAI Isometric",
+    "SAI Line Art": "SAI Line Art",
+    "SAI Lowpoly": "SAI Lowpoly",
+    "SAI Neonpunk": "SAI Neonpunk",
+    "SAI Origami": "SAI Origami",
+    "SAI Photographic": "SAI Photographic",
+    "SAI Pixel Art": "SAI Pixel Art",
+    "SAI Texture": "SAI Texture",
+    "MRE Cinematic Dynamic": "MRE Cinematic Dynamic",
+    "MRE Spontaneous Picture": "MRE Spontaneous Picture",
+    "MRE Artistic Vision": "MRE Artistic Vision",
+    "MRE Dark Dream": "MRE Dark Dream",
+    "MRE Gloomy Art": "MRE Gloomy Art",
+    "MRE Bad Dream": "MRE Bad Dream",
+    "MRE Underground": "MRE Underground",
+    "MRE Surreal Painting": "MRE Surreal Painting",
+    "MRE Dynamic Illustration": "MRE Dynamic Illustration",
+    "MRE Undead Art": "MRE Undead Art",
+    "MRE Elemental Art": "MRE Elemental Art",
+    "MRE Space Art": "MRE Space Art",
+    "MRE Ancient Illustration": "MRE Ancient Illustration",
+    "MRE Brave Art": "MRE Brave Art",
+    "MRE Heroic Fantasy": "MRE Heroic Fantasy",
+    "MRE Dark Cyberpunk": "MRE Dark Cyberpunk",
+    "MRE Lyrical Geometry": "MRE Lyrical Geometry",
+    "MRE Sumi E Symbolic": "MRE Sumi E Symbolic",
+    "MRE Sumi E Detailed": "MRE Sumi E Detailed",
+    "MRE Manga": "MRE Manga",
+    "MRE Anime": "MRE Anime",
+    "MRE Comic": "MRE Comic",
+    "Ads Advertising": "Ads Advertising",
+    "Ads Automotive": "Ads Automotive",
+    "Ads Corporate": "Ads Corporate",
+    "Ads Fashion Editorial": "Ads Fashion Editorial",
+    "Ads Food Photography": "Ads Food Photography",
+    "Ads Gourmet Food Photography": "Ads Gourmet Food Photography",
+    "Ads Luxury": "Ads Luxury",
+    "Ads Real Estate": "Ads Real Estate",
+    "Ads Retail": "Ads Retail",
+    "Artstyle Abstract": "Artstyle Abstract",
+    "Artstyle Abstract Expressionism": "Artstyle Abstract Expressionism",
+    "Artstyle Art Deco": "Artstyle Art Deco",
+    "Artstyle Art Nouveau": "Artstyle Art Nouveau",
+    "Artstyle Constructivist": "Artstyle Constructivist",
+    "Artstyle Cubist": "Artstyle Cubist",
+    "Artstyle Expressionist": "Artstyle Expressionist",
+    "Artstyle Graffiti": "Artstyle Graffiti",
+    "Artstyle Hyperrealism": "Artstyle Hyperrealism",
+    "Artstyle Impressionist": "Artstyle Impressionist",
+    "Artstyle Pointillism": "Artstyle Pointillism",
+    "Artstyle Pop Art": "Artstyle Pop Art",
+    "Artstyle Psychedelic": "Artstyle Psychedelic",
+    "Artstyle Renaissance": "Artstyle Renaissance",
+    "Artstyle Steampunk": "Artstyle Steampunk",
+    "Artstyle Surrealist": "Artstyle Surrealist",
+    "Artstyle Typography": "Artstyle Typography",
+    "Artstyle Watercolor": "Artstyle Watercolor",
+    "Futuristic Biomechanical": "Futuristic Biomechanical",
+    "Futuristic Biomechanical Cyberpunk": "Futuristic Biomechanical Cyberpunk",
+    "Futuristic Cybernetic": "Futuristic Cybernetic",
+    "Futuristic Cybernetic Robot": "Futuristic Cybernetic Robot",
+    "Futuristic Cyberpunk Cityscape": "Futuristic Cyberpunk Cityscape",
+    "Futuristic Futuristic": "Futuristic Futuristic",
+    "Futuristic Retro Cyberpunk": "Futuristic Retro Cyberpunk",
+    "Futuristic Retro Futurism": "Futuristic Retro Futurism",
+    "Futuristic Sci Fi": "Futuristic Sci Fi",
+    "Futuristic Vaporwave": "Futuristic Vaporwave",
+    "Game Bubble Bobble": "Game Bubble Bobble",
+    "Game Cyberpunk Game": "Game Cyberpunk Game",
+    "Game Fighting Game": "Game Fighting Game",
+    "Game Gta": "Game Gta",
+    "Game Mario": "Game Mario",
+    "Game Minecraft": "Game Minecraft",
+    "Game Pokemon": "Game Pokemon",
+    "Game Retro Arcade": "Game Retro Arcade",
+    "Game Retro Game": "Game Retro Game",
+    "Game Rpg Fantasy Game": "Game Rpg Fantasy Game",
+    "Game Strategy Game": "Game Strategy Game",
+    "Game Streetfighter": "Game Streetfighter",
+    "Game Zelda": "Game Zelda",
+    "Misc Architectural": "Misc Architectural",
+    "Misc Disco": "Misc Disco",
+    "Misc Dreamscape": "Misc Dreamscape",
+    "Misc Dystopian": "Misc Dystopian",
+    "Misc Fairy Tale": "Misc Fairy Tale",
+    "Misc Gothic": "Misc Gothic",
+    "Misc Grunge": "Misc Grunge",
+    "Misc Horror": "Misc Horror",
+    "Misc Kawaii": "Misc Kawaii",
+    "Misc Lovecraftian": "Misc Lovecraftian",
+    "Misc Macabre": "Misc Macabre",
+    "Misc Manga": "Misc Manga",
+    "Misc Metropolis": "Misc Metropolis",
+    "Misc Minimalist": "Misc Minimalist",
+    "Misc Monochrome": "Misc Monochrome",
+    "Misc Nautical": "Misc Nautical",
+    "Misc Space": "Misc Space",
+    "Misc Stained Glass": "Misc Stained Glass",
+    "Misc Techwear Fashion": "Misc Techwear Fashion",
+    "Misc Tribal": "Misc Tribal",
+    "Misc Zentangle": "Misc Zentangle",
+    "Papercraft Collage": "Papercraft Collage",
+    "Papercraft Flat Papercut": "Papercraft Flat Papercut",
+    "Papercraft Kirigami": "Papercraft Kirigami",
+    "Papercraft Paper Mache": "Papercraft Paper Mache",
+    "Papercraft Paper Quilling": "Papercraft Paper Quilling",
+    "Papercraft Papercut Collage": "Papercraft Papercut Collage",
+    "Papercraft Papercut Shadow Box": "Papercraft Papercut Shadow Box",
+    "Papercraft Stacked Papercut": "Papercraft Stacked Papercut",
+    "Papercraft Thick Layered Papercut": "Papercraft Thick Layered Papercut",
+    "Photo Alien": "Photo Alien",
+    "Photo Film Noir": "Photo Film Noir",
+    "Photo Glamour": "Photo Glamour",
+    "Photo Hdr": "Photo Hdr",
+    "Photo Iphone Photographic": "Photo Iphone Photographic",
+    "Photo Long Exposure": "Photo Long Exposure",
+    "Photo Neon Noir": "Photo Neon Noir",
+    "Photo Silhouette": "Photo Silhouette",
+    "Photo Tilt Shift": "Photo Tilt Shift",
+    "Cinematic Diva": "Cinematic Diva",
+    "Abstract Expressionism": "Abstract Expressionism",
+    "Academia": "Academia",
+    "Action Figure": "Action Figure",
+    "Adorable 3D Character": "Adorable 3D Character",
+    "Adorable Kawaii": "Adorable Kawaii",
+    "Art Deco": "Art Deco",
+    "Art Nouveau": "Art Nouveau",
+    "Astral Aura": "Astral Aura",
+    "Avant Garde": "Avant Garde",
+    "Baroque": "Baroque",
+    "Bauhaus Style Poster": "Bauhaus Style Poster",
+    "Blueprint Schematic Drawing": "Blueprint Schematic Drawing",
+    "Caricature": "Caricature",
+    "Cel Shaded Art": "Cel Shaded Art",
+    "Character Design Sheet": "Character Design Sheet",
+    "Classicism Art": "Classicism Art",
+    "Color Field Painting": "Color Field Painting",
+    "Colored Pencil Art": "Colored Pencil Art",
+    "Conceptual Art": "Conceptual Art",
+    "Constructivism": "Constructivism",
+    "Cubism": "Cubism",
+    "Dadaism": "Dadaism",
+    "Dark Fantasy": "Dark Fantasy",
+    "Dark Moody Atmosphere": "Dark Moody Atmosphere",
+    "Dmt Art Style": "Dmt Art Style",
+    "Doodle Art": "Doodle Art",
+    "Double Exposure": "Double Exposure",
+    "Dripping Paint Splatter Art": "Dripping Paint Splatter Art",
+    "Expressionism": "Expressionism",
+    "Faded Polaroid Photo": "Faded Polaroid Photo",
+    "Fauvism": "Fauvism",
+    "Flat 2d Art": "Flat 2d Art",
+    "Fortnite Art Style": "Fortnite Art Style",
+    "Futurism": "Futurism",
+    "Glitchcore": "Glitchcore",
+    "Glo Fi": "Glo Fi",
+    "Googie Art Style": "Googie Art Style",
+    "Graffiti Art": "Graffiti Art",
+    "Harlem Renaissance Art": "Harlem Renaissance Art",
+    "High Fashion": "High Fashion",
+    "Idyllic": "Idyllic",
+    "Impressionism": "Impressionism",
+    "Infographic Drawing": "Infographic Drawing",
+    "Ink Dripping Drawing": "Ink Dripping Drawing",
+    "Japanese Ink Drawing": "Japanese Ink Drawing",
+    "Knolling Photography": "Knolling Photography",
+    "Light Cheery Atmosphere": "Light Cheery Atmosphere",
+    "Logo Design": "Logo Design",
+    "Luxurious Elegance": "Luxurious Elegance",
+    "Macro Photography": "Macro Photography",
+    "Mandola Art": "Mandola Art",
+    "Marker Drawing": "Marker Drawing",
+    "Medievalism": "Medievalism",
+    "Minimalism": "Minimalism",
+    "Neo Baroque": "Neo Baroque",
+    "Neo Byzantine": "Neo Byzantine",
+    "Neo Futurism": "Neo Futurism",
+    "Neo Impressionism": "Neo Impressionism",
+    "Neo Rococo": "Neo Rococo",
+    "Neoclassicism": "Neoclassicism",
+    "Op Art": "Op Art",
+    "Ornate And Intricate": "Ornate And Intricate",
+    "Pencil Sketch Drawing": "Pencil Sketch Drawing",
+    "Pop Art 2": "Pop Art 2",
+    "Rococo": "Rococo",
+    "Silhouette Art": "Silhouette Art",
+    "Simple Vector Art": "Simple Vector Art",
+    "Sketchup": "Sketchup",
+    "Steampunk 2": "Steampunk 2",
+    "Surrealism": "Surrealism",
+    "Suprematism": "Suprematism",
+    "Terragen": "Terragen",
+    "Tranquil Relaxing Atmosphere": "Tranquil Relaxing Atmosphere",
+    "Sticker Designs": "Sticker Designs",
+    "Vibrant Rim Light": "Vibrant Rim Light",
+    "Volumetric Lighting": "Volumetric Lighting",
+    "Watercolor 2": "Watercolor 2",
+    "Whimsical And Playful": "Whimsical And Playful",
+    "Model": "Model",
+    "Base Model (SDXL only)": "Base Model (SDXL only)",
+    "sd_xl_base_1.0_0.9vae.safetensors": "sd_xl_base_1.0_0.9vae.safetensors",
+    "bluePencilXL_v009.safetensors": "bluePencilXL_v009.safetensors",
+    "bluePencilXL_v050.safetensors": "bluePencilXL_v050.safetensors",
+    "DreamShaper_8_pruned.safetensors": "DreamShaper_8_pruned.safetensors",
+    "realisticStockPhoto_v10.safetensors": "realisticStockPhoto_v10.safetensors",
+    "realisticVisionV51_v51VAE.safetensors": "realisticVisionV51_v51VAE.safetensors",
+    "sd_xl_refiner_1.0_0.9vae.safetensors": "sd_xl_refiner_1.0_0.9vae.safetensors",
+    "Refiner (SDXL or SD 1.5)": "Refiner (SDXL or SD 1.5)",
+    "None": "None",
+    "LoRAs": "LoRAs",
+    "SDXL LoRA 1": "SDXL LoRA 1",
+    "sd_xl_offset_example-lora_1.0.safetensors": "sd_xl_offset_example-lora_1.0.safetensors",
+    "3d_render_style_xl.safetensors": "3d_render_style_xl.safetensors",
+    "Bloodstained-XL-V1.safetensors": "Bloodstained-XL-V1.safetensors",
+    "SDXL_FILM_PHOTOGRAPHY_STYLE_BetaV0.4.safetensors": "SDXL_FILM_PHOTOGRAPHY_STYLE_BetaV0.4.safetensors",
+    "SDXL LoRA 2": "SDXL LoRA 2",
+    "SDXL LoRA 3": "SDXL LoRA 3",
+    "SDXL LoRA 4": "SDXL LoRA 4",
+    "SDXL LoRA 5": "SDXL LoRA 5",
+    "Refresh": "Refresh",
+    "\ud83d\udd04 Refresh All Files": "\ud83d\udd04 Refresh All Files",
+    "Sampling Sharpness": "Sampling Sharpness",
+    "Higher value means image and texture are sharper.": "Higher value means image and texture are sharper.",
+    "Guidance Scale": "Guidance Scale",
+    "Higher value means style is cleaner, vivider, and more artistic.": "Higher value means style is cleaner, vivider, and more artistic.",
+    "Developer Debug Mode": "Developer Debug Mode",
+    "Developer Debug Tools": "Developer Debug Tools",
+    "Positive ADM Guidance Scaler": "Positive ADM Guidance Scaler",
+    "The scaler multiplied to positive ADM (use 1.0 to disable). ": "The scaler multiplied to positive ADM (use 1.0 to disable). ",
+    "Negative ADM Guidance Scaler": "Negative ADM Guidance Scaler",
+    "The scaler multiplied to negative ADM (use 1.0 to disable). ": "The scaler multiplied to negative ADM (use 1.0 to disable). ",
+    "ADM Guidance End At Step": "ADM Guidance End At Step",
+    "When to end the guidance from positive/negative ADM. ": "When to end the guidance from positive/negative ADM. ",
+    "Refiner swap method": "Refiner swap method",
+    "joint": "joint",
+    "separate": "separate",
+    "vae": "vae",
+    "CFG Mimicking from TSNR": "CFG Mimicking from TSNR",
+    "Enabling Fooocus's implementation of CFG mimicking for TSNR (effective when real CFG > mimicked CFG).": "Enabling Fooocus's implementation of CFG mimicking for TSNR (effective when real CFG > mimicked CFG).",
+    "Sampler": "Sampler",
+    "dpmpp_2m_sde_gpu": "dpmpp_2m_sde_gpu",
+    "Only effective in non-inpaint mode.": "Only effective in non-inpaint mode.",
+    "euler": "euler",
+    "euler_ancestral": "euler_ancestral",
+    "heun": "heun",
+    "dpm_2": "dpm_2",
+    "dpm_2_ancestral": "dpm_2_ancestral",
+    "lms": "lms",
+    "dpm_fast": "dpm_fast",
+    "dpm_adaptive": "dpm_adaptive",
+    "dpmpp_2s_ancestral": "dpmpp_2s_ancestral",
+    "dpmpp_sde": "dpmpp_sde",
+    "dpmpp_sde_gpu": "dpmpp_sde_gpu",
+    "dpmpp_2m": "dpmpp_2m",
+    "dpmpp_2m_sde": "dpmpp_2m_sde",
+    "dpmpp_3m_sde": "dpmpp_3m_sde",
+    "dpmpp_3m_sde_gpu": "dpmpp_3m_sde_gpu",
+    "ddpm": "ddpm",
+    "ddim": "ddim",
+    "uni_pc": "uni_pc",
+    "uni_pc_bh2": "uni_pc_bh2",
+    "Scheduler": "Scheduler",
+    "karras": "karras",
+    "Scheduler of Sampler.": "Scheduler of Sampler.",
+    "normal": "normal",
+    "exponential": "exponential",
+    "sgm_uniform": "sgm_uniform",
+    "simple": "simple",
+    "ddim_uniform": "ddim_uniform",
+    "Forced Overwrite of Sampling Step": "Forced Overwrite of Sampling Step",
+    "Set as -1 to disable. For developer debugging.": "Set as -1 to disable. For developer debugging.",
+    "Forced Overwrite of Refiner Switch Step": "Forced Overwrite of Refiner Switch Step",
+    "Forced Overwrite of Generating Width": "Forced Overwrite of Generating Width",
+    "Set as -1 to disable. For developer debugging. Results will be worse for non-standard numbers that SDXL is not trained on.": "Set as -1 to disable. For developer debugging. Results will be worse for non-standard numbers that SDXL is not trained on.",
+    "Forced Overwrite of Generating Height": "Forced Overwrite of Generating Height",
+    "Forced Overwrite of Denoising Strength of \"Vary\"": "Forced Overwrite of Denoising Strength of \"Vary\"",
+    "Set as negative number to disable. For developer debugging.": "Set as negative number to disable. For developer debugging.",
+    "Forced Overwrite of Denoising Strength of \"Upscale\"": "Forced Overwrite of Denoising Strength of \"Upscale\"",
+    "Inpaint Engine": "Inpaint Engine",
+    "v1": "v1",
+    "Version of Fooocus inpaint model": "Version of Fooocus inpaint model",
+    "v2.5": "v2.5",
+    "Control Debug": "Control Debug",
+    "Debug Preprocessors": "Debug Preprocessors",
+    "Mixing Image Prompt and Vary/Upscale": "Mixing Image Prompt and Vary/Upscale",
+    "Mixing Image Prompt and Inpaint": "Mixing Image Prompt and Inpaint",
+    "Softness of ControlNet": "Softness of ControlNet",
+    "Similar to the Control Mode in A1111 (use 0.0 to disable). ": "Similar to the Control Mode in A1111 (use 0.0 to disable). ",
+    "Canny": "Canny",
+    "Canny Low Threshold": "Canny Low Threshold",
+    "Canny High Threshold": "Canny High Threshold",
+    "FreeU": "FreeU",
+    "Enabled": "Enabled",
+    "B1": "B1",
+    "B2": "B2",
+    "S1": "S1",
+    "S2": "S2",
+    "Extreme Speed": "Extreme Speed",
+    "\uD83D\uDD0E Type here to search styles ...": "\uD83D\uDD0E Type here to search styles ...",
+    "Type prompt here.": "Type prompt here.",
+    "Outpaint Expansion Direction:": "Outpaint Expansion Direction:",
+    "* Powered by Fooocus Inpaint Engine (beta)": "* Powered by Fooocus Inpaint Engine (beta)",
+    "Fooocus Enhance": "Fooocus Enhance",
+    "Fooocus Cinematic": "Fooocus Cinematic",
+    "Fooocus Sharp": "Fooocus Sharp"
+}

language/example.json

@@ -0,0 +1,6 @@
+{
+  "Generate": "生成",
+  "Input Image": "入力画像",
+  "Advanced": "고급",
+  "SAI 3D Model": "SAI 3D Modèle"
+}

launch.py

@@ -1,49 +1,60 @@
 import os
 import sys
+import ssl
+
+print('[System ARGV] ' + str(sys.argv))
+
+root = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(root)
+os.chdir(root)
+
+os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] = "1"
+os.environ["PYTORCH_MPS_HIGH_WATERMARK_RATIO"] = "0.0"
+os.environ["GRADIO_SERVER_PORT"] = "7865"
+
+ssl._create_default_https_context = ssl._create_unverified_context
+
+
 import platform
 import fooocus_version
 
-from modules.launch_util import is_installed, run, python, \
-    run_pip, repo_dir, git_clone, requirements_met, script_path, dir_repos
+from build_launcher import build_launcher
+from modules.launch_util import is_installed, run, python, run_pip, requirements_met
 from modules.model_loader import load_file_from_url
-from modules.path import modelfile_path, lorafile_path
+from modules.config import path_checkpoints, path_loras, path_vae_approx, path_fooocus_expansion, \
+    checkpoint_downloads, path_embeddings, embeddings_downloads, lora_downloads
+
 
 REINSTALL_ALL = False
+TRY_INSTALL_XFORMERS = False
 
 
 def prepare_environment():
-    torch_index_url = os.environ.get('TORCH_INDEX_URL', "https://download.pytorch.org/whl/cu118")
+    torch_index_url = os.environ.get('TORCH_INDEX_URL', "https://download.pytorch.org/whl/cu121")
     torch_command = os.environ.get('TORCH_COMMAND',
-                                   f"pip install torch==2.0.1 torchvision==0.15.2 --extra-index-url {torch_index_url}")
+                                   f"pip install torch==2.1.0 torchvision==0.16.0 --extra-index-url {torch_index_url}")
     requirements_file = os.environ.get('REQS_FILE', "requirements_versions.txt")
 
-    xformers_package = os.environ.get('XFORMERS_PACKAGE', 'xformers==0.0.20')
-
-    comfy_repo = os.environ.get('COMFY_REPO', "https://github.com/comfyanonymous/ComfyUI")
-    comfy_commit_hash = os.environ.get('COMFY_COMMIT_HASH', "2bc12d3d22efb5c63ae3a7fc342bb2dd16b31735")
-
     print(f"Python {sys.version}")
     print(f"Fooocus version: {fooocus_version.version}")
 
-    comfyui_name = 'ComfyUI-from-StabilityAI-Official'
-    git_clone(comfy_repo, repo_dir(comfyui_name), "Inference Engine", comfy_commit_hash)
-    sys.path.append(os.path.join(script_path, dir_repos, comfyui_name))
-
     if REINSTALL_ALL or not is_installed("torch") or not is_installed("torchvision"):
         run(f'"{python}" -m {torch_command}', "Installing torch and torchvision", "Couldn't install torch", live=True)
 
-    if REINSTALL_ALL or not is_installed("xformers"):
-        if platform.system() == "Windows":
-            if platform.python_version().startswith("3.10"):
-                run_pip(f"install -U -I --no-deps {xformers_package}", "xformers", live=True)
-            else:
-                print("Installation of xformers is not supported in this version of Python.")
-                print(
-                    "You can also check this and build manually: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Xformers#building-xformers-on-windows-by-duckness")
-                if not is_installed("xformers"):
-                    exit(0)
-        elif platform.system() == "Linux":
-            run_pip(f"install -U -I --no-deps {xformers_package}", "xformers")
+    if TRY_INSTALL_XFORMERS:
+        if REINSTALL_ALL or not is_installed("xformers"):
+            xformers_package = os.environ.get('XFORMERS_PACKAGE', 'xformers==0.0.20')
+            if platform.system() == "Windows":
+                if platform.python_version().startswith("3.10"):
+                    run_pip(f"install -U -I --no-deps {xformers_package}", "xformers", live=True)
+                else:
+                    print("Installation of xformers is not supported in this version of Python.")
+                    print(
+                        "You can also check this and build manually: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Xformers#building-xformers-on-windows-by-duckness")
+                    if not is_installed("xformers"):
+                        exit(0)
+            elif platform.system() == "Linux":
+                run_pip(f"install -U -I --no-deps {xformers_package}", "xformers")
 
     if REINSTALL_ALL or not requirements_met(requirements_file):
         run_pip(f"install -r \"{requirements_file}\"", "requirements")
@@ -51,30 +62,47 @@ def prepare_environment():
     return
 
 
-model_filenames = [
-    ('sd_xl_base_1.0.safetensors',
-     'https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0.safetensors'),
-    ('sd_xl_refiner_1.0.safetensors',
-     'https://huggingface.co/stabilityai/stable-diffusion-xl-refiner-1.0/resolve/main/sd_xl_refiner_1.0.safetensors')
-]
-
-lora_filenames = [
-    ('sd_xl_offset_example-lora_1.0.safetensors',
-     'https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_offset_example-lora_1.0.safetensors')
+vae_approx_filenames = [
+    ('xlvaeapp.pth', 'https://huggingface.co/lllyasviel/misc/resolve/main/xlvaeapp.pth'),
+    ('vaeapp_sd15.pth', 'https://huggingface.co/lllyasviel/misc/resolve/main/vaeapp_sd15.pt'),
+    ('xl-to-v1_interposer-v3.1.safetensors',
+     'https://huggingface.co/lllyasviel/misc/resolve/main/xl-to-v1_interposer-v3.1.safetensors')
 ]
 
 
 def download_models():
-    for file_name, url in model_filenames:
-        load_file_from_url(url=url, model_dir=modelfile_path, file_name=file_name)
-    for file_name, url in lora_filenames:
-        load_file_from_url(url=url, model_dir=lorafile_path, file_name=file_name)
+    for file_name, url in checkpoint_downloads.items():
+        load_file_from_url(url=url, model_dir=path_checkpoints, file_name=file_name)
+    for file_name, url in embeddings_downloads.items():
+        load_file_from_url(url=url, model_dir=path_embeddings, file_name=file_name)
+    for file_name, url in lora_downloads.items():
+        load_file_from_url(url=url, model_dir=path_loras, file_name=file_name)
+    for file_name, url in vae_approx_filenames:
+        load_file_from_url(url=url, model_dir=path_vae_approx, file_name=file_name)
+
+    load_file_from_url(
+        url='https://huggingface.co/lllyasviel/misc/resolve/main/fooocus_expansion.bin',
+        model_dir=path_fooocus_expansion,
+        file_name='pytorch_model.bin'
+    )
+
     return
 
 
+def ini_args():
+    from args_manager import args
+    return args
+
+
 prepare_environment()
+build_launcher()
+args = ini_args()
+
+
+if args.gpu_device_id is not None:
+    os.environ['CUDA_VISIBLE_DEVICES'] = str(args.gpu_device_id)
+    print("Set device to:", args.gpu_device_id)
 
-import cuda_malloc
 
 download_models()
 

ldm_patched/contrib/external_canny.py

@@ -0,0 +1,301 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+#From https://github.com/kornia/kornia
+import math
+
+import torch
+import torch.nn.functional as F
+import ldm_patched.modules.model_management
+
+def get_canny_nms_kernel(device=None, dtype=None):
+    """Utility function that returns 3x3 kernels for the Canny Non-maximal suppression."""
+    return torch.tensor(
+        [
+            [[[0.0, 0.0, 0.0], [0.0, 1.0, -1.0], [0.0, 0.0, 0.0]]],
+            [[[0.0, 0.0, 0.0], [0.0, 1.0, 0.0], [0.0, 0.0, -1.0]]],
+            [[[0.0, 0.0, 0.0], [0.0, 1.0, 0.0], [0.0, -1.0, 0.0]]],
+            [[[0.0, 0.0, 0.0], [0.0, 1.0, 0.0], [-1.0, 0.0, 0.0]]],
+            [[[0.0, 0.0, 0.0], [-1.0, 1.0, 0.0], [0.0, 0.0, 0.0]]],
+            [[[-1.0, 0.0, 0.0], [0.0, 1.0, 0.0], [0.0, 0.0, 0.0]]],
+            [[[0.0, -1.0, 0.0], [0.0, 1.0, 0.0], [0.0, 0.0, 0.0]]],
+            [[[0.0, 0.0, -1.0], [0.0, 1.0, 0.0], [0.0, 0.0, 0.0]]],
+        ],
+        device=device,
+        dtype=dtype,
+    )
+
+
+def get_hysteresis_kernel(device=None, dtype=None):
+    """Utility function that returns the 3x3 kernels for the Canny hysteresis."""
+    return torch.tensor(
+        [
+            [[[0.0, 0.0, 0.0], [0.0, 0.0, 1.0], [0.0, 0.0, 0.0]]],
+            [[[0.0, 0.0, 0.0], [0.0, 0.0, 0.0], [0.0, 0.0, 1.0]]],
+            [[[0.0, 0.0, 0.0], [0.0, 0.0, 0.0], [0.0, 1.0, 0.0]]],
+            [[[0.0, 0.0, 0.0], [0.0, 0.0, 0.0], [1.0, 0.0, 0.0]]],
+            [[[0.0, 0.0, 0.0], [1.0, 0.0, 0.0], [0.0, 0.0, 0.0]]],
+            [[[1.0, 0.0, 0.0], [0.0, 0.0, 0.0], [0.0, 0.0, 0.0]]],
+            [[[0.0, 1.0, 0.0], [0.0, 0.0, 0.0], [0.0, 0.0, 0.0]]],
+            [[[0.0, 0.0, 1.0], [0.0, 0.0, 0.0], [0.0, 0.0, 0.0]]],
+        ],
+        device=device,
+        dtype=dtype,
+    )
+
+def gaussian_blur_2d(img, kernel_size, sigma):
+    ksize_half = (kernel_size - 1) * 0.5
+
+    x = torch.linspace(-ksize_half, ksize_half, steps=kernel_size)
+
+    pdf = torch.exp(-0.5 * (x / sigma).pow(2))
+
+    x_kernel = pdf / pdf.sum()
+    x_kernel = x_kernel.to(device=img.device, dtype=img.dtype)
+
+    kernel2d = torch.mm(x_kernel[:, None], x_kernel[None, :])
+    kernel2d = kernel2d.expand(img.shape[-3], 1, kernel2d.shape[0], kernel2d.shape[1])
+
+    padding = [kernel_size // 2, kernel_size // 2, kernel_size // 2, kernel_size // 2]
+
+    img = torch.nn.functional.pad(img, padding, mode="reflect")
+    img = torch.nn.functional.conv2d(img, kernel2d, groups=img.shape[-3])
+
+    return img
+
+def get_sobel_kernel2d(device=None, dtype=None):
+    kernel_x = torch.tensor([[-1.0, 0.0, 1.0], [-2.0, 0.0, 2.0], [-1.0, 0.0, 1.0]], device=device, dtype=dtype)
+    kernel_y = kernel_x.transpose(0, 1)
+    return torch.stack([kernel_x, kernel_y])
+
+def spatial_gradient(input, normalized: bool = True):
+    r"""Compute the first order image derivative in both x and y using a Sobel operator.
+    .. image:: _static/img/spatial_gradient.png
+    Args:
+        input: input image tensor with shape :math:`(B, C, H, W)`.
+        mode: derivatives modality, can be: `sobel` or `diff`.
+        order: the order of the derivatives.
+        normalized: whether the output is normalized.
+    Return:
+        the derivatives of the input feature map. with shape :math:`(B, C, 2, H, W)`.
+    .. note::
+       See a working example `here <https://kornia-tutorials.readthedocs.io/en/latest/
+       filtering_edges.html>`__.
+    Examples:
+        >>> input = torch.rand(1, 3, 4, 4)
+        >>> output = spatial_gradient(input)  # 1x3x2x4x4
+        >>> output.shape
+        torch.Size([1, 3, 2, 4, 4])
+    """
+    # KORNIA_CHECK_IS_TENSOR(input)
+    # KORNIA_CHECK_SHAPE(input, ['B', 'C', 'H', 'W'])
+
+    # allocate kernel
+    kernel = get_sobel_kernel2d(device=input.device, dtype=input.dtype)
+    if normalized:
+        kernel = normalize_kernel2d(kernel)
+
+    # prepare kernel
+    b, c, h, w = input.shape
+    tmp_kernel = kernel[:, None, ...]
+
+    # Pad with "replicate for spatial dims, but with zeros for channel
+    spatial_pad = [kernel.size(1) // 2, kernel.size(1) // 2, kernel.size(2) // 2, kernel.size(2) // 2]
+    out_channels: int = 2
+    padded_inp = torch.nn.functional.pad(input.reshape(b * c, 1, h, w), spatial_pad, 'replicate')
+    out = F.conv2d(padded_inp, tmp_kernel, groups=1, padding=0, stride=1)
+    return out.reshape(b, c, out_channels, h, w)
+
+def rgb_to_grayscale(image, rgb_weights = None):
+    r"""Convert a RGB image to grayscale version of image.
+
+    .. image:: _static/img/rgb_to_grayscale.png
+
+    The image data is assumed to be in the range of (0, 1).
+
+    Args:
+        image: RGB image to be converted to grayscale with shape :math:`(*,3,H,W)`.
+        rgb_weights: Weights that will be applied on each channel (RGB).
+            The sum of the weights should add up to one.
+    Returns:
+        grayscale version of the image with shape :math:`(*,1,H,W)`.
+
+    .. note::
+       See a working example `here <https://kornia-tutorials.readthedocs.io/en/latest/
+       color_conversions.html>`__.
+
+    Example:
+        >>> input = torch.rand(2, 3, 4, 5)
+        >>> gray = rgb_to_grayscale(input) # 2x1x4x5
+    """
+
+    if len(image.shape) < 3 or image.shape[-3] != 3:
+        raise ValueError(f"Input size must have a shape of (*, 3, H, W). Got {image.shape}")
+
+    if rgb_weights is None:
+        # 8 bit images
+        if image.dtype == torch.uint8:
+            rgb_weights = torch.tensor([76, 150, 29], device=image.device, dtype=torch.uint8)
+        # floating point images
+        elif image.dtype in (torch.float16, torch.float32, torch.float64):
+            rgb_weights = torch.tensor([0.299, 0.587, 0.114], device=image.device, dtype=image.dtype)
+        else:
+            raise TypeError(f"Unknown data type: {image.dtype}")
+    else:
+        # is tensor that we make sure is in the same device/dtype
+        rgb_weights = rgb_weights.to(image)
+
+    # unpack the color image channels with RGB order
+    r: Tensor = image[..., 0:1, :, :]
+    g: Tensor = image[..., 1:2, :, :]
+    b: Tensor = image[..., 2:3, :, :]
+
+    w_r, w_g, w_b = rgb_weights.unbind()
+    return w_r * r + w_g * g + w_b * b
+
+def canny(
+    input,
+    low_threshold = 0.1,
+    high_threshold = 0.2,
+    kernel_size  = 5,
+    sigma = 1,
+    hysteresis = True,
+    eps = 1e-6,
+):
+    r"""Find edges of the input image and filters them using the Canny algorithm.
+    .. image:: _static/img/canny.png
+    Args:
+        input: input image tensor with shape :math:`(B,C,H,W)`.
+        low_threshold: lower threshold for the hysteresis procedure.
+        high_threshold: upper threshold for the hysteresis procedure.
+        kernel_size: the size of the kernel for the gaussian blur.
+        sigma: the standard deviation of the kernel for the gaussian blur.
+        hysteresis: if True, applies the hysteresis edge tracking.
+            Otherwise, the edges are divided between weak (0.5) and strong (1) edges.
+        eps: regularization number to avoid NaN during backprop.
+    Returns:
+        - the canny edge magnitudes map, shape of :math:`(B,1,H,W)`.
+        - the canny edge detection filtered by thresholds and hysteresis, shape of :math:`(B,1,H,W)`.
+    .. note::
+       See a working example `here <https://kornia-tutorials.readthedocs.io/en/latest/
+       canny.html>`__.
+    Example:
+        >>> input = torch.rand(5, 3, 4, 4)
+        >>> magnitude, edges = canny(input)  # 5x3x4x4
+        >>> magnitude.shape
+        torch.Size([5, 1, 4, 4])
+        >>> edges.shape
+        torch.Size([5, 1, 4, 4])
+    """
+    # KORNIA_CHECK_IS_TENSOR(input)
+    # KORNIA_CHECK_SHAPE(input, ['B', 'C', 'H', 'W'])
+    # KORNIA_CHECK(
+    #     low_threshold <= high_threshold,
+    #     "Invalid input thresholds. low_threshold should be smaller than the high_threshold. Got: "
+    #     f"{low_threshold}>{high_threshold}",
+    # )
+    # KORNIA_CHECK(0 < low_threshold < 1, f'Invalid low threshold. Should be in range (0, 1). Got: {low_threshold}')
+    # KORNIA_CHECK(0 < high_threshold < 1, f'Invalid high threshold. Should be in range (0, 1). Got: {high_threshold}')
+
+    device = input.device
+    dtype = input.dtype
+
+    # To Grayscale
+    if input.shape[1] == 3:
+        input = rgb_to_grayscale(input)
+
+    # Gaussian filter
+    blurred: Tensor = gaussian_blur_2d(input, kernel_size, sigma)
+
+    # Compute the gradients
+    gradients: Tensor = spatial_gradient(blurred, normalized=False)
+
+    # Unpack the edges
+    gx: Tensor = gradients[:, :, 0]
+    gy: Tensor = gradients[:, :, 1]
+
+    # Compute gradient magnitude and angle
+    magnitude: Tensor = torch.sqrt(gx * gx + gy * gy + eps)
+    angle: Tensor = torch.atan2(gy, gx)
+
+    # Radians to Degrees
+    angle = 180.0 * angle / math.pi
+
+    # Round angle to the nearest 45 degree
+    angle = torch.round(angle / 45) * 45
+
+    # Non-maximal suppression
+    nms_kernels: Tensor = get_canny_nms_kernel(device, dtype)
+    nms_magnitude: Tensor = F.conv2d(magnitude, nms_kernels, padding=nms_kernels.shape[-1] // 2)
+
+    # Get the indices for both directions
+    positive_idx: Tensor = (angle / 45) % 8
+    positive_idx = positive_idx.long()
+
+    negative_idx: Tensor = ((angle / 45) + 4) % 8
+    negative_idx = negative_idx.long()
+
+    # Apply the non-maximum suppression to the different directions
+    channel_select_filtered_positive: Tensor = torch.gather(nms_magnitude, 1, positive_idx)
+    channel_select_filtered_negative: Tensor = torch.gather(nms_magnitude, 1, negative_idx)
+
+    channel_select_filtered: Tensor = torch.stack(
+        [channel_select_filtered_positive, channel_select_filtered_negative], 1
+    )
+
+    is_max: Tensor = channel_select_filtered.min(dim=1)[0] > 0.0
+
+    magnitude = magnitude * is_max
+
+    # Threshold
+    edges: Tensor = F.threshold(magnitude, low_threshold, 0.0)
+
+    low: Tensor = magnitude > low_threshold
+    high: Tensor = magnitude > high_threshold
+
+    edges = low * 0.5 + high * 0.5
+    edges = edges.to(dtype)
+
+    # Hysteresis
+    if hysteresis:
+        edges_old: Tensor = -torch.ones(edges.shape, device=edges.device, dtype=dtype)
+        hysteresis_kernels: Tensor = get_hysteresis_kernel(device, dtype)
+
+        while ((edges_old - edges).abs() != 0).any():
+            weak: Tensor = (edges == 0.5).float()
+            strong: Tensor = (edges == 1).float()
+
+            hysteresis_magnitude: Tensor = F.conv2d(
+                edges, hysteresis_kernels, padding=hysteresis_kernels.shape[-1] // 2
+            )
+            hysteresis_magnitude = (hysteresis_magnitude == 1).any(1, keepdim=True).to(dtype)
+            hysteresis_magnitude = hysteresis_magnitude * weak + strong
+
+            edges_old = edges.clone()
+            edges = hysteresis_magnitude + (hysteresis_magnitude == 0) * weak * 0.5
+
+        edges = hysteresis_magnitude
+
+    return magnitude, edges
+
+
+class Canny:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": {"image": ("IMAGE",),
+                                "low_threshold": ("FLOAT", {"default": 0.4, "min": 0.01, "max": 0.99, "step": 0.01}),
+                                "high_threshold": ("FLOAT", {"default": 0.8, "min": 0.01, "max": 0.99, "step": 0.01})
+                                }}
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "detect_edge"
+
+    CATEGORY = "image/preprocessors"
+
+    def detect_edge(self, image, low_threshold, high_threshold):
+        output = canny(image.to(ldm_patched.modules.model_management.get_torch_device()).movedim(-1, 1), low_threshold, high_threshold)
+        img_out = output[1].to(ldm_patched.modules.model_management.intermediate_device()).repeat(1, 3, 1, 1).movedim(1, -1)
+        return (img_out,)
+
+NODE_CLASS_MAPPINGS = {
+    "Canny": Canny,
+}

ldm_patched/contrib/external_clip_sdxl.py

@@ -0,0 +1,58 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+import torch
+from ldm_patched.contrib.external import MAX_RESOLUTION
+
+class CLIPTextEncodeSDXLRefiner:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": {
+            "ascore": ("FLOAT", {"default": 6.0, "min": 0.0, "max": 1000.0, "step": 0.01}),
+            "width": ("INT", {"default": 1024.0, "min": 0, "max": MAX_RESOLUTION}),
+            "height": ("INT", {"default": 1024.0, "min": 0, "max": MAX_RESOLUTION}),
+            "text": ("STRING", {"multiline": True}), "clip": ("CLIP", ),
+            }}
+    RETURN_TYPES = ("CONDITIONING",)
+    FUNCTION = "encode"
+
+    CATEGORY = "advanced/conditioning"
+
+    def encode(self, clip, ascore, width, height, text):
+        tokens = clip.tokenize(text)
+        cond, pooled = clip.encode_from_tokens(tokens, return_pooled=True)
+        return ([[cond, {"pooled_output": pooled, "aesthetic_score": ascore, "width": width,"height": height}]], )
+
+class CLIPTextEncodeSDXL:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": {
+            "width": ("INT", {"default": 1024.0, "min": 0, "max": MAX_RESOLUTION}),
+            "height": ("INT", {"default": 1024.0, "min": 0, "max": MAX_RESOLUTION}),
+            "crop_w": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION}),
+            "crop_h": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION}),
+            "target_width": ("INT", {"default": 1024.0, "min": 0, "max": MAX_RESOLUTION}),
+            "target_height": ("INT", {"default": 1024.0, "min": 0, "max": MAX_RESOLUTION}),
+            "text_g": ("STRING", {"multiline": True, "default": "CLIP_G"}), "clip": ("CLIP", ),
+            "text_l": ("STRING", {"multiline": True, "default": "CLIP_L"}), "clip": ("CLIP", ),
+            }}
+    RETURN_TYPES = ("CONDITIONING",)
+    FUNCTION = "encode"
+
+    CATEGORY = "advanced/conditioning"
+
+    def encode(self, clip, width, height, crop_w, crop_h, target_width, target_height, text_g, text_l):
+        tokens = clip.tokenize(text_g)
+        tokens["l"] = clip.tokenize(text_l)["l"]
+        if len(tokens["l"]) != len(tokens["g"]):
+            empty = clip.tokenize("")
+            while len(tokens["l"]) < len(tokens["g"]):
+                tokens["l"] += empty["l"]
+            while len(tokens["l"]) > len(tokens["g"]):
+                tokens["g"] += empty["g"]
+        cond, pooled = clip.encode_from_tokens(tokens, return_pooled=True)
+        return ([[cond, {"pooled_output": pooled, "width": width, "height": height, "crop_w": crop_w, "crop_h": crop_h, "target_width": target_width, "target_height": target_height}]], )
+
+NODE_CLASS_MAPPINGS = {
+    "CLIPTextEncodeSDXLRefiner": CLIPTextEncodeSDXLRefiner,
+    "CLIPTextEncodeSDXL": CLIPTextEncodeSDXL,
+}

ldm_patched/contrib/external_compositing.py

@@ -0,0 +1,204 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+import numpy as np
+import torch
+import ldm_patched.modules.utils
+from enum import Enum
+
+def resize_mask(mask, shape):
+    return torch.nn.functional.interpolate(mask.reshape((-1, 1, mask.shape[-2], mask.shape[-1])), size=(shape[0], shape[1]), mode="bilinear").squeeze(1)
+
+class PorterDuffMode(Enum):
+    ADD = 0
+    CLEAR = 1
+    DARKEN = 2
+    DST = 3
+    DST_ATOP = 4
+    DST_IN = 5
+    DST_OUT = 6
+    DST_OVER = 7
+    LIGHTEN = 8
+    MULTIPLY = 9
+    OVERLAY = 10
+    SCREEN = 11
+    SRC = 12
+    SRC_ATOP = 13
+    SRC_IN = 14
+    SRC_OUT = 15
+    SRC_OVER = 16
+    XOR = 17
+
+
+def porter_duff_composite(src_image: torch.Tensor, src_alpha: torch.Tensor, dst_image: torch.Tensor, dst_alpha: torch.Tensor, mode: PorterDuffMode):
+    if mode == PorterDuffMode.ADD:
+        out_alpha = torch.clamp(src_alpha + dst_alpha, 0, 1)
+        out_image = torch.clamp(src_image + dst_image, 0, 1)
+    elif mode == PorterDuffMode.CLEAR:
+        out_alpha = torch.zeros_like(dst_alpha)
+        out_image = torch.zeros_like(dst_image)
+    elif mode == PorterDuffMode.DARKEN:
+        out_alpha = src_alpha + dst_alpha  - src_alpha * dst_alpha
+        out_image = (1 - dst_alpha) * src_image + (1 - src_alpha) * dst_image + torch.min(src_image, dst_image)
+    elif mode == PorterDuffMode.DST:
+        out_alpha = dst_alpha
+        out_image = dst_image
+    elif mode == PorterDuffMode.DST_ATOP:
+        out_alpha = src_alpha
+        out_image = src_alpha * dst_image + (1 - dst_alpha) * src_image
+    elif mode == PorterDuffMode.DST_IN:
+        out_alpha = src_alpha * dst_alpha
+        out_image = dst_image * src_alpha
+    elif mode == PorterDuffMode.DST_OUT:
+        out_alpha = (1 - src_alpha) * dst_alpha
+        out_image = (1 - src_alpha) * dst_image
+    elif mode == PorterDuffMode.DST_OVER:
+        out_alpha = dst_alpha + (1 - dst_alpha) * src_alpha
+        out_image = dst_image + (1 - dst_alpha) * src_image
+    elif mode == PorterDuffMode.LIGHTEN:
+        out_alpha = src_alpha + dst_alpha - src_alpha * dst_alpha
+        out_image = (1 - dst_alpha) * src_image + (1 - src_alpha) * dst_image + torch.max(src_image, dst_image)
+    elif mode == PorterDuffMode.MULTIPLY:
+        out_alpha = src_alpha * dst_alpha
+        out_image = src_image * dst_image
+    elif mode == PorterDuffMode.OVERLAY:
+        out_alpha = src_alpha + dst_alpha - src_alpha * dst_alpha
+        out_image = torch.where(2 * dst_image < dst_alpha, 2 * src_image * dst_image,
+            src_alpha * dst_alpha - 2 * (dst_alpha - src_image) * (src_alpha - dst_image))
+    elif mode == PorterDuffMode.SCREEN:
+        out_alpha = src_alpha + dst_alpha - src_alpha * dst_alpha
+        out_image = src_image + dst_image - src_image * dst_image
+    elif mode == PorterDuffMode.SRC:
+        out_alpha = src_alpha
+        out_image = src_image
+    elif mode == PorterDuffMode.SRC_ATOP:
+        out_alpha = dst_alpha
+        out_image = dst_alpha * src_image + (1 - src_alpha) * dst_image
+    elif mode == PorterDuffMode.SRC_IN:
+        out_alpha = src_alpha * dst_alpha
+        out_image = src_image * dst_alpha
+    elif mode == PorterDuffMode.SRC_OUT:
+        out_alpha = (1 - dst_alpha) * src_alpha
+        out_image = (1 - dst_alpha) * src_image
+    elif mode == PorterDuffMode.SRC_OVER:
+        out_alpha = src_alpha + (1 - src_alpha) * dst_alpha
+        out_image = src_image + (1 - src_alpha) * dst_image
+    elif mode == PorterDuffMode.XOR:
+        out_alpha = (1 - dst_alpha) * src_alpha + (1 - src_alpha) * dst_alpha
+        out_image = (1 - dst_alpha) * src_image + (1 - src_alpha) * dst_image
+    else:
+        out_alpha = None
+        out_image = None
+    return out_image, out_alpha
+
+
+class PorterDuffImageComposite:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+                "source": ("IMAGE",),
+                "source_alpha": ("MASK",),
+                "destination": ("IMAGE",),
+                "destination_alpha": ("MASK",),
+                "mode": ([mode.name for mode in PorterDuffMode], {"default": PorterDuffMode.DST.name}),
+            },
+        }
+
+    RETURN_TYPES = ("IMAGE", "MASK")
+    FUNCTION = "composite"
+    CATEGORY = "mask/compositing"
+
+    def composite(self, source: torch.Tensor, source_alpha: torch.Tensor, destination: torch.Tensor, destination_alpha: torch.Tensor, mode):
+        batch_size = min(len(source), len(source_alpha), len(destination), len(destination_alpha))
+        out_images = []
+        out_alphas = []
+
+        for i in range(batch_size):
+            src_image = source[i]
+            dst_image = destination[i]
+
+            assert src_image.shape[2] == dst_image.shape[2] # inputs need to have same number of channels
+
+            src_alpha = source_alpha[i].unsqueeze(2)
+            dst_alpha = destination_alpha[i].unsqueeze(2)
+
+            if dst_alpha.shape[:2] != dst_image.shape[:2]:
+                upscale_input = dst_alpha.unsqueeze(0).permute(0, 3, 1, 2)
+                upscale_output = ldm_patched.modules.utils.common_upscale(upscale_input, dst_image.shape[1], dst_image.shape[0], upscale_method='bicubic', crop='center')
+                dst_alpha = upscale_output.permute(0, 2, 3, 1).squeeze(0)
+            if src_image.shape != dst_image.shape:
+                upscale_input = src_image.unsqueeze(0).permute(0, 3, 1, 2)
+                upscale_output = ldm_patched.modules.utils.common_upscale(upscale_input, dst_image.shape[1], dst_image.shape[0], upscale_method='bicubic', crop='center')
+                src_image = upscale_output.permute(0, 2, 3, 1).squeeze(0)
+            if src_alpha.shape != dst_alpha.shape:
+                upscale_input = src_alpha.unsqueeze(0).permute(0, 3, 1, 2)
+                upscale_output = ldm_patched.modules.utils.common_upscale(upscale_input, dst_alpha.shape[1], dst_alpha.shape[0], upscale_method='bicubic', crop='center')
+                src_alpha = upscale_output.permute(0, 2, 3, 1).squeeze(0)
+
+            out_image, out_alpha = porter_duff_composite(src_image, src_alpha, dst_image, dst_alpha, PorterDuffMode[mode])
+
+            out_images.append(out_image)
+            out_alphas.append(out_alpha.squeeze(2))
+
+        result = (torch.stack(out_images), torch.stack(out_alphas))
+        return result
+
+
+class SplitImageWithAlpha:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+                "required": {
+                    "image": ("IMAGE",),
+                }
+        }
+
+    CATEGORY = "mask/compositing"
+    RETURN_TYPES = ("IMAGE", "MASK")
+    FUNCTION = "split_image_with_alpha"
+
+    def split_image_with_alpha(self, image: torch.Tensor):
+        out_images = [i[:,:,:3] for i in image]
+        out_alphas = [i[:,:,3] if i.shape[2] > 3 else torch.ones_like(i[:,:,0]) for i in image]
+        result = (torch.stack(out_images), 1.0 - torch.stack(out_alphas))
+        return result
+
+
+class JoinImageWithAlpha:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+                "required": {
+                    "image": ("IMAGE",),
+                    "alpha": ("MASK",),
+                }
+        }
+
+    CATEGORY = "mask/compositing"
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "join_image_with_alpha"
+
+    def join_image_with_alpha(self, image: torch.Tensor, alpha: torch.Tensor):
+        batch_size = min(len(image), len(alpha))
+        out_images = []
+
+        alpha = 1.0 - resize_mask(alpha, image.shape[1:])
+        for i in range(batch_size):
+           out_images.append(torch.cat((image[i][:,:,:3], alpha[i].unsqueeze(2)), dim=2))
+
+        result = (torch.stack(out_images),)
+        return result
+
+
+NODE_CLASS_MAPPINGS = {
+    "PorterDuffImageComposite": PorterDuffImageComposite,
+    "SplitImageWithAlpha": SplitImageWithAlpha,
+    "JoinImageWithAlpha": JoinImageWithAlpha,
+}
+
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "PorterDuffImageComposite": "Porter-Duff Image Composite",
+    "SplitImageWithAlpha": "Split Image with Alpha",
+    "JoinImageWithAlpha": "Join Image with Alpha",
+}

ldm_patched/contrib/external_custom_sampler.py

@@ -0,0 +1,287 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+import ldm_patched.modules.samplers
+import ldm_patched.modules.sample
+from ldm_patched.k_diffusion import sampling as k_diffusion_sampling
+import ldm_patched.utils.latent_visualization
+import torch
+import ldm_patched.modules.utils
+
+
+class BasicScheduler:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required":
+                    {"model": ("MODEL",),
+                     "scheduler": (ldm_patched.modules.samplers.SCHEDULER_NAMES, ),
+                     "steps": ("INT", {"default": 20, "min": 1, "max": 10000}),
+                      }
+               }
+    RETURN_TYPES = ("SIGMAS",)
+    CATEGORY = "sampling/custom_sampling/schedulers"
+
+    FUNCTION = "get_sigmas"
+
+    def get_sigmas(self, model, scheduler, steps):
+        sigmas = ldm_patched.modules.samplers.calculate_sigmas_scheduler(model.model, scheduler, steps).cpu()
+        return (sigmas, )
+
+
+class KarrasScheduler:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required":
+                    {"steps": ("INT", {"default": 20, "min": 1, "max": 10000}),
+                     "sigma_max": ("FLOAT", {"default": 14.614642, "min": 0.0, "max": 1000.0, "step":0.01, "round": False}),
+                     "sigma_min": ("FLOAT", {"default": 0.0291675, "min": 0.0, "max": 1000.0, "step":0.01, "round": False}),
+                     "rho": ("FLOAT", {"default": 7.0, "min": 0.0, "max": 100.0, "step":0.01, "round": False}),
+                    }
+               }
+    RETURN_TYPES = ("SIGMAS",)
+    CATEGORY = "sampling/custom_sampling/schedulers"
+
+    FUNCTION = "get_sigmas"
+
+    def get_sigmas(self, steps, sigma_max, sigma_min, rho):
+        sigmas = k_diffusion_sampling.get_sigmas_karras(n=steps, sigma_min=sigma_min, sigma_max=sigma_max, rho=rho)
+        return (sigmas, )
+
+class ExponentialScheduler:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required":
+                    {"steps": ("INT", {"default": 20, "min": 1, "max": 10000}),
+                     "sigma_max": ("FLOAT", {"default": 14.614642, "min": 0.0, "max": 1000.0, "step":0.01, "round": False}),
+                     "sigma_min": ("FLOAT", {"default": 0.0291675, "min": 0.0, "max": 1000.0, "step":0.01, "round": False}),
+                    }
+               }
+    RETURN_TYPES = ("SIGMAS",)
+    CATEGORY = "sampling/custom_sampling/schedulers"
+
+    FUNCTION = "get_sigmas"
+
+    def get_sigmas(self, steps, sigma_max, sigma_min):
+        sigmas = k_diffusion_sampling.get_sigmas_exponential(n=steps, sigma_min=sigma_min, sigma_max=sigma_max)
+        return (sigmas, )
+
+class PolyexponentialScheduler:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required":
+                    {"steps": ("INT", {"default": 20, "min": 1, "max": 10000}),
+                     "sigma_max": ("FLOAT", {"default": 14.614642, "min": 0.0, "max": 1000.0, "step":0.01, "round": False}),
+                     "sigma_min": ("FLOAT", {"default": 0.0291675, "min": 0.0, "max": 1000.0, "step":0.01, "round": False}),
+                     "rho": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0, "step":0.01, "round": False}),
+                    }
+               }
+    RETURN_TYPES = ("SIGMAS",)
+    CATEGORY = "sampling/custom_sampling/schedulers"
+
+    FUNCTION = "get_sigmas"
+
+    def get_sigmas(self, steps, sigma_max, sigma_min, rho):
+        sigmas = k_diffusion_sampling.get_sigmas_polyexponential(n=steps, sigma_min=sigma_min, sigma_max=sigma_max, rho=rho)
+        return (sigmas, )
+
+class SDTurboScheduler:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required":
+                    {"model": ("MODEL",),
+                     "steps": ("INT", {"default": 1, "min": 1, "max": 10}),
+                      }
+               }
+    RETURN_TYPES = ("SIGMAS",)
+    CATEGORY = "sampling/custom_sampling/schedulers"
+
+    FUNCTION = "get_sigmas"
+
+    def get_sigmas(self, model, steps):
+        timesteps = torch.flip(torch.arange(1, 11) * 100 - 1, (0,))[:steps]
+        sigmas = model.model.model_sampling.sigma(timesteps)
+        sigmas = torch.cat([sigmas, sigmas.new_zeros([1])])
+        return (sigmas, )
+
+class VPScheduler:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required":
+                    {"steps": ("INT", {"default": 20, "min": 1, "max": 10000}),
+                     "beta_d": ("FLOAT", {"default": 19.9, "min": 0.0, "max": 1000.0, "step":0.01, "round": False}), #TODO: fix default values
+                     "beta_min": ("FLOAT", {"default": 0.1, "min": 0.0, "max": 1000.0, "step":0.01, "round": False}),
+                     "eps_s": ("FLOAT", {"default": 0.001, "min": 0.0, "max": 1.0, "step":0.0001, "round": False}),
+                    }
+               }
+    RETURN_TYPES = ("SIGMAS",)
+    CATEGORY = "sampling/custom_sampling/schedulers"
+
+    FUNCTION = "get_sigmas"
+
+    def get_sigmas(self, steps, beta_d, beta_min, eps_s):
+        sigmas = k_diffusion_sampling.get_sigmas_vp(n=steps, beta_d=beta_d, beta_min=beta_min, eps_s=eps_s)
+        return (sigmas, )
+
+class SplitSigmas:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required":
+                    {"sigmas": ("SIGMAS", ),
+                    "step": ("INT", {"default": 0, "min": 0, "max": 10000}),
+                     }
+                }
+    RETURN_TYPES = ("SIGMAS","SIGMAS")
+    CATEGORY = "sampling/custom_sampling/sigmas"
+
+    FUNCTION = "get_sigmas"
+
+    def get_sigmas(self, sigmas, step):
+        sigmas1 = sigmas[:step + 1]
+        sigmas2 = sigmas[step:]
+        return (sigmas1, sigmas2)
+
+class FlipSigmas:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required":
+                    {"sigmas": ("SIGMAS", ),
+                     }
+                }
+    RETURN_TYPES = ("SIGMAS",)
+    CATEGORY = "sampling/custom_sampling/sigmas"
+
+    FUNCTION = "get_sigmas"
+
+    def get_sigmas(self, sigmas):
+        sigmas = sigmas.flip(0)
+        if sigmas[0] == 0:
+            sigmas[0] = 0.0001
+        return (sigmas,)
+
+class KSamplerSelect:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required":
+                    {"sampler_name": (ldm_patched.modules.samplers.SAMPLER_NAMES, ),
+                      }
+               }
+    RETURN_TYPES = ("SAMPLER",)
+    CATEGORY = "sampling/custom_sampling/samplers"
+
+    FUNCTION = "get_sampler"
+
+    def get_sampler(self, sampler_name):
+        sampler = ldm_patched.modules.samplers.sampler_object(sampler_name)
+        return (sampler, )
+
+class SamplerDPMPP_2M_SDE:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required":
+                    {"solver_type": (['midpoint', 'heun'], ),
+                     "eta": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0, "step":0.01, "round": False}),
+                     "s_noise": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0, "step":0.01, "round": False}),
+                     "noise_device": (['gpu', 'cpu'], ),
+                      }
+               }
+    RETURN_TYPES = ("SAMPLER",)
+    CATEGORY = "sampling/custom_sampling/samplers"
+
+    FUNCTION = "get_sampler"
+
+    def get_sampler(self, solver_type, eta, s_noise, noise_device):
+        if noise_device == 'cpu':
+            sampler_name = "dpmpp_2m_sde"
+        else:
+            sampler_name = "dpmpp_2m_sde_gpu"
+        sampler = ldm_patched.modules.samplers.ksampler(sampler_name, {"eta": eta, "s_noise": s_noise, "solver_type": solver_type})
+        return (sampler, )
+
+
+class SamplerDPMPP_SDE:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required":
+                    {"eta": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0, "step":0.01, "round": False}),
+                     "s_noise": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0, "step":0.01, "round": False}),
+                     "r": ("FLOAT", {"default": 0.5, "min": 0.0, "max": 100.0, "step":0.01, "round": False}),
+                     "noise_device": (['gpu', 'cpu'], ),
+                      }
+               }
+    RETURN_TYPES = ("SAMPLER",)
+    CATEGORY = "sampling/custom_sampling/samplers"
+
+    FUNCTION = "get_sampler"
+
+    def get_sampler(self, eta, s_noise, r, noise_device):
+        if noise_device == 'cpu':
+            sampler_name = "dpmpp_sde"
+        else:
+            sampler_name = "dpmpp_sde_gpu"
+        sampler = ldm_patched.modules.samplers.ksampler(sampler_name, {"eta": eta, "s_noise": s_noise, "r": r})
+        return (sampler, )
+
+class SamplerCustom:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required":
+                    {"model": ("MODEL",),
+                    "add_noise": ("BOOLEAN", {"default": True}),
+                    "noise_seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff}),
+                    "cfg": ("FLOAT", {"default": 8.0, "min": 0.0, "max": 100.0, "step":0.1, "round": 0.01}),
+                    "positive": ("CONDITIONING", ),
+                    "negative": ("CONDITIONING", ),
+                    "sampler": ("SAMPLER", ),
+                    "sigmas": ("SIGMAS", ),
+                    "latent_image": ("LATENT", ),
+                     }
+                }
+
+    RETURN_TYPES = ("LATENT","LATENT")
+    RETURN_NAMES = ("output", "denoised_output")
+
+    FUNCTION = "sample"
+
+    CATEGORY = "sampling/custom_sampling"
+
+    def sample(self, model, add_noise, noise_seed, cfg, positive, negative, sampler, sigmas, latent_image):
+        latent = latent_image
+        latent_image = latent["samples"]
+        if not add_noise:
+            noise = torch.zeros(latent_image.size(), dtype=latent_image.dtype, layout=latent_image.layout, device="cpu")
+        else:
+            batch_inds = latent["batch_index"] if "batch_index" in latent else None
+            noise = ldm_patched.modules.sample.prepare_noise(latent_image, noise_seed, batch_inds)
+
+        noise_mask = None
+        if "noise_mask" in latent:
+            noise_mask = latent["noise_mask"]
+
+        x0_output = {}
+        callback = ldm_patched.utils.latent_visualization.prepare_callback(model, sigmas.shape[-1] - 1, x0_output)
+
+        disable_pbar = not ldm_patched.modules.utils.PROGRESS_BAR_ENABLED
+        samples = ldm_patched.modules.sample.sample_custom(model, noise, cfg, sampler, sigmas, positive, negative, latent_image, noise_mask=noise_mask, callback=callback, disable_pbar=disable_pbar, seed=noise_seed)
+
+        out = latent.copy()
+        out["samples"] = samples
+        if "x0" in x0_output:
+            out_denoised = latent.copy()
+            out_denoised["samples"] = model.model.process_latent_out(x0_output["x0"].cpu())
+        else:
+            out_denoised = out
+        return (out, out_denoised)
+
+NODE_CLASS_MAPPINGS = {
+    "SamplerCustom": SamplerCustom,
+    "BasicScheduler": BasicScheduler,
+    "KarrasScheduler": KarrasScheduler,
+    "ExponentialScheduler": ExponentialScheduler,
+    "PolyexponentialScheduler": PolyexponentialScheduler,
+    "VPScheduler": VPScheduler,
+    "SDTurboScheduler": SDTurboScheduler,
+    "KSamplerSelect": KSamplerSelect,
+    "SamplerDPMPP_2M_SDE": SamplerDPMPP_2M_SDE,
+    "SamplerDPMPP_SDE": SamplerDPMPP_SDE,
+    "SplitSigmas": SplitSigmas,
+    "FlipSigmas": FlipSigmas,
+}

ldm_patched/contrib/external_freelunch.py

@@ -0,0 +1,115 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+#code originally taken from: https://github.com/ChenyangSi/FreeU (under MIT License)
+
+import torch
+
+
+def Fourier_filter(x, threshold, scale):
+    # FFT
+    x_freq = torch.fft.fftn(x.float(), dim=(-2, -1))
+    x_freq = torch.fft.fftshift(x_freq, dim=(-2, -1))
+
+    B, C, H, W = x_freq.shape
+    mask = torch.ones((B, C, H, W), device=x.device)
+
+    crow, ccol = H // 2, W //2
+    mask[..., crow - threshold:crow + threshold, ccol - threshold:ccol + threshold] = scale
+    x_freq = x_freq * mask
+
+    # IFFT
+    x_freq = torch.fft.ifftshift(x_freq, dim=(-2, -1))
+    x_filtered = torch.fft.ifftn(x_freq, dim=(-2, -1)).real
+
+    return x_filtered.to(x.dtype)
+
+
+class FreeU:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model": ("MODEL",),
+                             "b1": ("FLOAT", {"default": 1.1, "min": 0.0, "max": 10.0, "step": 0.01}),
+                             "b2": ("FLOAT", {"default": 1.2, "min": 0.0, "max": 10.0, "step": 0.01}),
+                             "s1": ("FLOAT", {"default": 0.9, "min": 0.0, "max": 10.0, "step": 0.01}),
+                             "s2": ("FLOAT", {"default": 0.2, "min": 0.0, "max": 10.0, "step": 0.01}),
+                              }}
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "patch"
+
+    CATEGORY = "_for_testing"
+
+    def patch(self, model, b1, b2, s1, s2):
+        model_channels = model.model.model_config.unet_config["model_channels"]
+        scale_dict = {model_channels * 4: (b1, s1), model_channels * 2: (b2, s2)}
+        on_cpu_devices = {}
+
+        def output_block_patch(h, hsp, transformer_options):
+            scale = scale_dict.get(h.shape[1], None)
+            if scale is not None:
+                h[:,:h.shape[1] // 2] = h[:,:h.shape[1] // 2] * scale[0]
+                if hsp.device not in on_cpu_devices:
+                    try:
+                        hsp = Fourier_filter(hsp, threshold=1, scale=scale[1])
+                    except:
+                        print("Device", hsp.device, "does not support the torch.fft functions used in the FreeU node, switching to CPU.")
+                        on_cpu_devices[hsp.device] = True
+                        hsp = Fourier_filter(hsp.cpu(), threshold=1, scale=scale[1]).to(hsp.device)
+                else:
+                    hsp = Fourier_filter(hsp.cpu(), threshold=1, scale=scale[1]).to(hsp.device)
+
+            return h, hsp
+
+        m = model.clone()
+        m.set_model_output_block_patch(output_block_patch)
+        return (m, )
+
+class FreeU_V2:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model": ("MODEL",),
+                             "b1": ("FLOAT", {"default": 1.3, "min": 0.0, "max": 10.0, "step": 0.01}),
+                             "b2": ("FLOAT", {"default": 1.4, "min": 0.0, "max": 10.0, "step": 0.01}),
+                             "s1": ("FLOAT", {"default": 0.9, "min": 0.0, "max": 10.0, "step": 0.01}),
+                             "s2": ("FLOAT", {"default": 0.2, "min": 0.0, "max": 10.0, "step": 0.01}),
+                              }}
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "patch"
+
+    CATEGORY = "_for_testing"
+
+    def patch(self, model, b1, b2, s1, s2):
+        model_channels = model.model.model_config.unet_config["model_channels"]
+        scale_dict = {model_channels * 4: (b1, s1), model_channels * 2: (b2, s2)}
+        on_cpu_devices = {}
+
+        def output_block_patch(h, hsp, transformer_options):
+            scale = scale_dict.get(h.shape[1], None)
+            if scale is not None:
+                hidden_mean = h.mean(1).unsqueeze(1)
+                B = hidden_mean.shape[0]
+                hidden_max, _ = torch.max(hidden_mean.view(B, -1), dim=-1, keepdim=True)
+                hidden_min, _ = torch.min(hidden_mean.view(B, -1), dim=-1, keepdim=True)
+                hidden_mean = (hidden_mean - hidden_min.unsqueeze(2).unsqueeze(3)) / (hidden_max - hidden_min).unsqueeze(2).unsqueeze(3)
+
+                h[:,:h.shape[1] // 2] = h[:,:h.shape[1] // 2] * ((scale[0] - 1 ) * hidden_mean + 1)
+
+                if hsp.device not in on_cpu_devices:
+                    try:
+                        hsp = Fourier_filter(hsp, threshold=1, scale=scale[1])
+                    except:
+                        print("Device", hsp.device, "does not support the torch.fft functions used in the FreeU node, switching to CPU.")
+                        on_cpu_devices[hsp.device] = True
+                        hsp = Fourier_filter(hsp.cpu(), threshold=1, scale=scale[1]).to(hsp.device)
+                else:
+                    hsp = Fourier_filter(hsp.cpu(), threshold=1, scale=scale[1]).to(hsp.device)
+
+            return h, hsp
+
+        m = model.clone()
+        m.set_model_output_block_patch(output_block_patch)
+        return (m, )
+
+NODE_CLASS_MAPPINGS = {
+    "FreeU": FreeU,
+    "FreeU_V2": FreeU_V2,
+}

ldm_patched/contrib/external_hypernetwork.py

@@ -0,0 +1,121 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+import ldm_patched.modules.utils
+import ldm_patched.utils.path_utils
+import torch
+
+def load_hypernetwork_patch(path, strength):
+    sd = ldm_patched.modules.utils.load_torch_file(path, safe_load=True)
+    activation_func = sd.get('activation_func', 'linear')
+    is_layer_norm = sd.get('is_layer_norm', False)
+    use_dropout = sd.get('use_dropout', False)
+    activate_output = sd.get('activate_output', False)
+    last_layer_dropout = sd.get('last_layer_dropout', False)
+
+    valid_activation = {
+        "linear": torch.nn.Identity,
+        "relu": torch.nn.ReLU,
+        "leakyrelu": torch.nn.LeakyReLU,
+        "elu": torch.nn.ELU,
+        "swish": torch.nn.Hardswish,
+        "tanh": torch.nn.Tanh,
+        "sigmoid": torch.nn.Sigmoid,
+        "softsign": torch.nn.Softsign,
+        "mish": torch.nn.Mish,
+    }
+
+    if activation_func not in valid_activation:
+        print("Unsupported Hypernetwork format, if you report it I might implement it.", path, " ", activation_func, is_layer_norm, use_dropout, activate_output, last_layer_dropout)
+        return None
+
+    out = {}
+
+    for d in sd:
+        try:
+            dim = int(d)
+        except:
+            continue
+
+        output = []
+        for index in [0, 1]:
+            attn_weights = sd[dim][index]
+            keys = attn_weights.keys()
+
+            linears = filter(lambda a: a.endswith(".weight"), keys)
+            linears = list(map(lambda a: a[:-len(".weight")], linears))
+            layers = []
+
+            i = 0
+            while i < len(linears):
+                lin_name = linears[i]
+                last_layer = (i == (len(linears) - 1))
+                penultimate_layer = (i == (len(linears) - 2))
+
+                lin_weight = attn_weights['{}.weight'.format(lin_name)]
+                lin_bias = attn_weights['{}.bias'.format(lin_name)]
+                layer = torch.nn.Linear(lin_weight.shape[1], lin_weight.shape[0])
+                layer.load_state_dict({"weight": lin_weight, "bias": lin_bias})
+                layers.append(layer)
+                if activation_func != "linear":
+                    if (not last_layer) or (activate_output):
+                        layers.append(valid_activation[activation_func]())
+                if is_layer_norm:
+                    i += 1
+                    ln_name = linears[i]
+                    ln_weight = attn_weights['{}.weight'.format(ln_name)]
+                    ln_bias = attn_weights['{}.bias'.format(ln_name)]
+                    ln = torch.nn.LayerNorm(ln_weight.shape[0])
+                    ln.load_state_dict({"weight": ln_weight, "bias": ln_bias})
+                    layers.append(ln)
+                if use_dropout:
+                    if (not last_layer) and (not penultimate_layer or last_layer_dropout):
+                        layers.append(torch.nn.Dropout(p=0.3))
+                i += 1
+
+            output.append(torch.nn.Sequential(*layers))
+        out[dim] = torch.nn.ModuleList(output)
+
+    class hypernetwork_patch:
+        def __init__(self, hypernet, strength):
+            self.hypernet = hypernet
+            self.strength = strength
+        def __call__(self, q, k, v, extra_options):
+            dim = k.shape[-1]
+            if dim in self.hypernet:
+                hn = self.hypernet[dim]
+                k = k + hn[0](k) * self.strength
+                v = v + hn[1](v) * self.strength
+
+            return q, k, v
+
+        def to(self, device):
+            for d in self.hypernet.keys():
+                self.hypernet[d] = self.hypernet[d].to(device)
+            return self
+
+    return hypernetwork_patch(out, strength)
+
+class HypernetworkLoader:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model": ("MODEL",),
+                              "hypernetwork_name": (ldm_patched.utils.path_utils.get_filename_list("hypernetworks"), ),
+                              "strength": ("FLOAT", {"default": 1.0, "min": -10.0, "max": 10.0, "step": 0.01}),
+                              }}
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "load_hypernetwork"
+
+    CATEGORY = "loaders"
+
+    def load_hypernetwork(self, model, hypernetwork_name, strength):
+        hypernetwork_path = ldm_patched.utils.path_utils.get_full_path("hypernetworks", hypernetwork_name)
+        model_hypernetwork = model.clone()
+        patch = load_hypernetwork_patch(hypernetwork_path, strength)
+        if patch is not None:
+            model_hypernetwork.set_model_attn1_patch(patch)
+            model_hypernetwork.set_model_attn2_patch(patch)
+        return (model_hypernetwork,)
+
+NODE_CLASS_MAPPINGS = {
+    "HypernetworkLoader": HypernetworkLoader
+}

ldm_patched/contrib/external_hypertile.py

@@ -0,0 +1,85 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+#Taken from: https://github.com/tfernd/HyperTile/
+
+import math
+from einops import rearrange
+# Use torch rng for consistency across generations
+from torch import randint
+
+def random_divisor(value: int, min_value: int, /, max_options: int = 1) -> int:
+    min_value = min(min_value, value)
+
+    # All big divisors of value (inclusive)
+    divisors = [i for i in range(min_value, value + 1) if value % i == 0]
+
+    ns = [value // i for i in divisors[:max_options]]  # has at least 1 element
+
+    if len(ns) - 1 > 0:
+        idx = randint(low=0, high=len(ns) - 1, size=(1,)).item()
+    else:
+        idx = 0
+
+    return ns[idx]
+
+class HyperTile:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model": ("MODEL",),
+                             "tile_size": ("INT", {"default": 256, "min": 1, "max": 2048}),
+                             "swap_size": ("INT", {"default": 2, "min": 1, "max": 128}),
+                             "max_depth": ("INT", {"default": 0, "min": 0, "max": 10}),
+                             "scale_depth": ("BOOLEAN", {"default": False}),
+                              }}
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "patch"
+
+    CATEGORY = "_for_testing"
+
+    def patch(self, model, tile_size, swap_size, max_depth, scale_depth):
+        model_channels = model.model.model_config.unet_config["model_channels"]
+
+        apply_to = set()
+        temp = model_channels
+        for x in range(max_depth + 1):
+            apply_to.add(temp)
+            temp *= 2
+
+        latent_tile_size = max(32, tile_size) // 8
+        self.temp = None
+
+        def hypertile_in(q, k, v, extra_options):
+            if q.shape[-1] in apply_to:
+                shape = extra_options["original_shape"]
+                aspect_ratio = shape[-1] / shape[-2]
+
+                hw = q.size(1)
+                h, w = round(math.sqrt(hw * aspect_ratio)), round(math.sqrt(hw / aspect_ratio))
+
+                factor = 2**((q.shape[-1] // model_channels) - 1) if scale_depth else 1
+                nh = random_divisor(h, latent_tile_size * factor, swap_size)
+                nw = random_divisor(w, latent_tile_size * factor, swap_size)
+
+                if nh * nw > 1:
+                    q = rearrange(q, "b (nh h nw w) c -> (b nh nw) (h w) c", h=h // nh, w=w // nw, nh=nh, nw=nw)
+                    self.temp = (nh, nw, h, w)
+                return q, k, v
+
+            return q, k, v
+        def hypertile_out(out, extra_options):
+            if self.temp is not None:
+                nh, nw, h, w = self.temp
+                self.temp = None
+                out = rearrange(out, "(b nh nw) hw c -> b nh nw hw c", nh=nh, nw=nw)
+                out = rearrange(out, "b nh nw (h w) c -> b (nh h nw w) c", h=h // nh, w=w // nw)
+            return out
+
+
+        m = model.clone()
+        m.set_model_attn1_patch(hypertile_in)
+        m.set_model_attn1_output_patch(hypertile_out)
+        return (m, )
+
+NODE_CLASS_MAPPINGS = {
+    "HyperTile": HyperTile,
+}

ldm_patched/contrib/external_images.py

@@ -0,0 +1,177 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+import ldm_patched.contrib.external
+import ldm_patched.utils.path_utils
+from ldm_patched.modules.args_parser import args
+
+from PIL import Image
+from PIL.PngImagePlugin import PngInfo
+
+import numpy as np
+import json
+import os
+
+MAX_RESOLUTION = ldm_patched.contrib.external.MAX_RESOLUTION
+
+class ImageCrop:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "image": ("IMAGE",),
+                              "width": ("INT", {"default": 512, "min": 1, "max": MAX_RESOLUTION, "step": 1}),
+                              "height": ("INT", {"default": 512, "min": 1, "max": MAX_RESOLUTION, "step": 1}),
+                              "x": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION, "step": 1}),
+                              "y": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION, "step": 1}),
+                              }}
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "crop"
+
+    CATEGORY = "image/transform"
+
+    def crop(self, image, width, height, x, y):
+        x = min(x, image.shape[2] - 1)
+        y = min(y, image.shape[1] - 1)
+        to_x = width + x
+        to_y = height + y
+        img = image[:,y:to_y, x:to_x, :]
+        return (img,)
+
+class RepeatImageBatch:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "image": ("IMAGE",),
+                              "amount": ("INT", {"default": 1, "min": 1, "max": 64}),
+                              }}
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "repeat"
+
+    CATEGORY = "image/batch"
+
+    def repeat(self, image, amount):
+        s = image.repeat((amount, 1,1,1))
+        return (s,)
+
+class SaveAnimatedWEBP:
+    def __init__(self):
+        self.output_dir = ldm_patched.utils.path_utils.get_output_directory()
+        self.type = "output"
+        self.prefix_append = ""
+
+    methods = {"default": 4, "fastest": 0, "slowest": 6}
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required":
+                    {"images": ("IMAGE", ),
+                     "filename_prefix": ("STRING", {"default": "ldm_patched"}),
+                     "fps": ("FLOAT", {"default": 6.0, "min": 0.01, "max": 1000.0, "step": 0.01}),
+                     "lossless": ("BOOLEAN", {"default": True}),
+                     "quality": ("INT", {"default": 80, "min": 0, "max": 100}),
+                     "method": (list(s.methods.keys()),),
+                     # "num_frames": ("INT", {"default": 0, "min": 0, "max": 8192}),
+                     },
+                "hidden": {"prompt": "PROMPT", "extra_pnginfo": "EXTRA_PNGINFO"},
+                }
+
+    RETURN_TYPES = ()
+    FUNCTION = "save_images"
+
+    OUTPUT_NODE = True
+
+    CATEGORY = "image/animation"
+
+    def save_images(self, images, fps, filename_prefix, lossless, quality, method, num_frames=0, prompt=None, extra_pnginfo=None):
+        method = self.methods.get(method)
+        filename_prefix += self.prefix_append
+        full_output_folder, filename, counter, subfolder, filename_prefix = ldm_patched.utils.path_utils.get_save_image_path(filename_prefix, self.output_dir, images[0].shape[1], images[0].shape[0])
+        results = list()
+        pil_images = []
+        for image in images:
+            i = 255. * image.cpu().numpy()
+            img = Image.fromarray(np.clip(i, 0, 255).astype(np.uint8))
+            pil_images.append(img)
+
+        metadata = pil_images[0].getexif()
+        if not args.disable_server_info:
+            if prompt is not None:
+                metadata[0x0110] = "prompt:{}".format(json.dumps(prompt))
+            if extra_pnginfo is not None:
+                inital_exif = 0x010f
+                for x in extra_pnginfo:
+                    metadata[inital_exif] = "{}:{}".format(x, json.dumps(extra_pnginfo[x]))
+                    inital_exif -= 1
+
+        if num_frames == 0:
+            num_frames = len(pil_images)
+
+        c = len(pil_images)
+        for i in range(0, c, num_frames):
+            file = f"{filename}_{counter:05}_.webp"
+            pil_images[i].save(os.path.join(full_output_folder, file), save_all=True, duration=int(1000.0/fps), append_images=pil_images[i + 1:i + num_frames], exif=metadata, lossless=lossless, quality=quality, method=method)
+            results.append({
+                "filename": file,
+                "subfolder": subfolder,
+                "type": self.type
+            })
+            counter += 1
+
+        animated = num_frames != 1
+        return { "ui": { "images": results, "animated": (animated,) } }
+
+class SaveAnimatedPNG:
+    def __init__(self):
+        self.output_dir = ldm_patched.utils.path_utils.get_output_directory()
+        self.type = "output"
+        self.prefix_append = ""
+
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required":
+                    {"images": ("IMAGE", ),
+                     "filename_prefix": ("STRING", {"default": "ldm_patched"}),
+                     "fps": ("FLOAT", {"default": 6.0, "min": 0.01, "max": 1000.0, "step": 0.01}),
+                     "compress_level": ("INT", {"default": 4, "min": 0, "max": 9})
+                     },
+                "hidden": {"prompt": "PROMPT", "extra_pnginfo": "EXTRA_PNGINFO"},
+                }
+
+    RETURN_TYPES = ()
+    FUNCTION = "save_images"
+
+    OUTPUT_NODE = True
+
+    CATEGORY = "image/animation"
+
+    def save_images(self, images, fps, compress_level, filename_prefix="ldm_patched", prompt=None, extra_pnginfo=None):
+        filename_prefix += self.prefix_append
+        full_output_folder, filename, counter, subfolder, filename_prefix = ldm_patched.utils.path_utils.get_save_image_path(filename_prefix, self.output_dir, images[0].shape[1], images[0].shape[0])
+        results = list()
+        pil_images = []
+        for image in images:
+            i = 255. * image.cpu().numpy()
+            img = Image.fromarray(np.clip(i, 0, 255).astype(np.uint8))
+            pil_images.append(img)
+
+        metadata = None
+        if not args.disable_server_info:
+            metadata = PngInfo()
+            if prompt is not None:
+                metadata.add(b"ldm_patched", "prompt".encode("latin-1", "strict") + b"\0" + json.dumps(prompt).encode("latin-1", "strict"), after_idat=True)
+            if extra_pnginfo is not None:
+                for x in extra_pnginfo:
+                    metadata.add(b"ldm_patched", x.encode("latin-1", "strict") + b"\0" + json.dumps(extra_pnginfo[x]).encode("latin-1", "strict"), after_idat=True)
+
+        file = f"{filename}_{counter:05}_.png"
+        pil_images[0].save(os.path.join(full_output_folder, file), pnginfo=metadata, compress_level=compress_level, save_all=True, duration=int(1000.0/fps), append_images=pil_images[1:])
+        results.append({
+            "filename": file,
+            "subfolder": subfolder,
+            "type": self.type
+        })
+
+        return { "ui": { "images": results, "animated": (True,)} }
+
+NODE_CLASS_MAPPINGS = {
+    "ImageCrop": ImageCrop,
+    "RepeatImageBatch": RepeatImageBatch,
+    "SaveAnimatedWEBP": SaveAnimatedWEBP,
+    "SaveAnimatedPNG": SaveAnimatedPNG,
+}

ldm_patched/contrib/external_latent.py

@@ -0,0 +1,133 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+import ldm_patched.modules.utils
+import torch
+
+def reshape_latent_to(target_shape, latent):
+    if latent.shape[1:] != target_shape[1:]:
+        latent = ldm_patched.modules.utils.common_upscale(latent, target_shape[3], target_shape[2], "bilinear", "center")
+    return ldm_patched.modules.utils.repeat_to_batch_size(latent, target_shape[0])
+
+
+class LatentAdd:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "samples1": ("LATENT",), "samples2": ("LATENT",)}}
+
+    RETURN_TYPES = ("LATENT",)
+    FUNCTION = "op"
+
+    CATEGORY = "latent/advanced"
+
+    def op(self, samples1, samples2):
+        samples_out = samples1.copy()
+
+        s1 = samples1["samples"]
+        s2 = samples2["samples"]
+
+        s2 = reshape_latent_to(s1.shape, s2)
+        samples_out["samples"] = s1 + s2
+        return (samples_out,)
+
+class LatentSubtract:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "samples1": ("LATENT",), "samples2": ("LATENT",)}}
+
+    RETURN_TYPES = ("LATENT",)
+    FUNCTION = "op"
+
+    CATEGORY = "latent/advanced"
+
+    def op(self, samples1, samples2):
+        samples_out = samples1.copy()
+
+        s1 = samples1["samples"]
+        s2 = samples2["samples"]
+
+        s2 = reshape_latent_to(s1.shape, s2)
+        samples_out["samples"] = s1 - s2
+        return (samples_out,)
+
+class LatentMultiply:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "samples": ("LATENT",),
+                              "multiplier": ("FLOAT", {"default": 1.0, "min": -10.0, "max": 10.0, "step": 0.01}),
+                             }}
+
+    RETURN_TYPES = ("LATENT",)
+    FUNCTION = "op"
+
+    CATEGORY = "latent/advanced"
+
+    def op(self, samples, multiplier):
+        samples_out = samples.copy()
+
+        s1 = samples["samples"]
+        samples_out["samples"] = s1 * multiplier
+        return (samples_out,)
+
+class LatentInterpolate:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "samples1": ("LATENT",),
+                              "samples2": ("LATENT",),
+                              "ratio": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01}),
+                              }}
+
+    RETURN_TYPES = ("LATENT",)
+    FUNCTION = "op"
+
+    CATEGORY = "latent/advanced"
+
+    def op(self, samples1, samples2, ratio):
+        samples_out = samples1.copy()
+
+        s1 = samples1["samples"]
+        s2 = samples2["samples"]
+
+        s2 = reshape_latent_to(s1.shape, s2)
+
+        m1 = torch.linalg.vector_norm(s1, dim=(1))
+        m2 = torch.linalg.vector_norm(s2, dim=(1))
+
+        s1 = torch.nan_to_num(s1 / m1)
+        s2 = torch.nan_to_num(s2 / m2)
+
+        t = (s1 * ratio + s2 * (1.0 - ratio))
+        mt = torch.linalg.vector_norm(t, dim=(1))
+        st = torch.nan_to_num(t / mt)
+
+        samples_out["samples"] = st * (m1 * ratio + m2 * (1.0 - ratio))
+        return (samples_out,)
+
+class LatentBatch:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "samples1": ("LATENT",), "samples2": ("LATENT",)}}
+
+    RETURN_TYPES = ("LATENT",)
+    FUNCTION = "batch"
+
+    CATEGORY = "latent/batch"
+
+    def batch(self, samples1, samples2):
+        samples_out = samples1.copy()
+        s1 = samples1["samples"]
+        s2 = samples2["samples"]
+
+        if s1.shape[1:] != s2.shape[1:]:
+            s2 = ldm_patched.modules.utils.common_upscale(s2, s1.shape[3], s1.shape[2], "bilinear", "center")
+        s = torch.cat((s1, s2), dim=0)
+        samples_out["samples"] = s
+        samples_out["batch_index"] = samples1.get("batch_index", [x for x in range(0, s1.shape[0])]) + samples2.get("batch_index", [x for x in range(0, s2.shape[0])])
+        return (samples_out,)
+
+NODE_CLASS_MAPPINGS = {
+    "LatentAdd": LatentAdd,
+    "LatentSubtract": LatentSubtract,
+    "LatentMultiply": LatentMultiply,
+    "LatentInterpolate": LatentInterpolate,
+    "LatentBatch": LatentBatch,
+}

ldm_patched/contrib/external_mask.py

@@ -0,0 +1,364 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+import numpy as np
+import scipy.ndimage
+import torch
+import ldm_patched.modules.utils
+
+from ldm_patched.contrib.external import MAX_RESOLUTION
+
+def composite(destination, source, x, y, mask = None, multiplier = 8, resize_source = False):
+    if resize_source:
+        source = torch.nn.functional.interpolate(source, size=(destination.shape[2], destination.shape[3]), mode="bilinear")
+
+    source = ldm_patched.modules.utils.repeat_to_batch_size(source, destination.shape[0])
+
+    x = max(-source.shape[3] * multiplier, min(x, destination.shape[3] * multiplier))
+    y = max(-source.shape[2] * multiplier, min(y, destination.shape[2] * multiplier))
+
+    left, top = (x // multiplier, y // multiplier)
+    right, bottom = (left + source.shape[3], top + source.shape[2],)
+
+    if mask is None:
+        mask = torch.ones_like(source)
+    else:
+        mask = mask.clone()
+        mask = torch.nn.functional.interpolate(mask.reshape((-1, 1, mask.shape[-2], mask.shape[-1])), size=(source.shape[2], source.shape[3]), mode="bilinear")
+        mask = ldm_patched.modules.utils.repeat_to_batch_size(mask, source.shape[0])
+
+    # calculate the bounds of the source that will be overlapping the destination
+    # this prevents the source trying to overwrite latent pixels that are out of bounds
+    # of the destination
+    visible_width, visible_height = (destination.shape[3] - left + min(0, x), destination.shape[2] - top + min(0, y),)
+
+    mask = mask[:, :, :visible_height, :visible_width]
+    inverse_mask = torch.ones_like(mask) - mask
+
+    source_portion = mask * source[:, :, :visible_height, :visible_width]
+    destination_portion = inverse_mask  * destination[:, :, top:bottom, left:right]
+
+    destination[:, :, top:bottom, left:right] = source_portion + destination_portion
+    return destination
+
+class LatentCompositeMasked:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+                "destination": ("LATENT",),
+                "source": ("LATENT",),
+                "x": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION, "step": 8}),
+                "y": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION, "step": 8}),
+                "resize_source": ("BOOLEAN", {"default": False}),
+            },
+            "optional": {
+                "mask": ("MASK",),
+            }
+        }
+    RETURN_TYPES = ("LATENT",)
+    FUNCTION = "composite"
+
+    CATEGORY = "latent"
+
+    def composite(self, destination, source, x, y, resize_source, mask = None):
+        output = destination.copy()
+        destination = destination["samples"].clone()
+        source = source["samples"]
+        output["samples"] = composite(destination, source, x, y, mask, 8, resize_source)
+        return (output,)
+
+class ImageCompositeMasked:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+                "destination": ("IMAGE",),
+                "source": ("IMAGE",),
+                "x": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION, "step": 1}),
+                "y": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION, "step": 1}),
+                "resize_source": ("BOOLEAN", {"default": False}),
+            },
+            "optional": {
+                "mask": ("MASK",),
+            }
+        }
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "composite"
+
+    CATEGORY = "image"
+
+    def composite(self, destination, source, x, y, resize_source, mask = None):
+        destination = destination.clone().movedim(-1, 1)
+        output = composite(destination, source.movedim(-1, 1), x, y, mask, 1, resize_source).movedim(1, -1)
+        return (output,)
+
+class MaskToImage:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+                "required": {
+                    "mask": ("MASK",),
+                }
+        }
+
+    CATEGORY = "mask"
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "mask_to_image"
+
+    def mask_to_image(self, mask):
+        result = mask.reshape((-1, 1, mask.shape[-2], mask.shape[-1])).movedim(1, -1).expand(-1, -1, -1, 3)
+        return (result,)
+
+class ImageToMask:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+                "required": {
+                    "image": ("IMAGE",),
+                    "channel": (["red", "green", "blue", "alpha"],),
+                }
+        }
+
+    CATEGORY = "mask"
+
+    RETURN_TYPES = ("MASK",)
+    FUNCTION = "image_to_mask"
+
+    def image_to_mask(self, image, channel):
+        channels = ["red", "green", "blue", "alpha"]
+        mask = image[:, :, :, channels.index(channel)]
+        return (mask,)
+
+class ImageColorToMask:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+                "required": {
+                    "image": ("IMAGE",),
+                    "color": ("INT", {"default": 0, "min": 0, "max": 0xFFFFFF, "step": 1, "display": "color"}),
+                }
+        }
+
+    CATEGORY = "mask"
+
+    RETURN_TYPES = ("MASK",)
+    FUNCTION = "image_to_mask"
+
+    def image_to_mask(self, image, color):
+        temp = (torch.clamp(image, 0, 1.0) * 255.0).round().to(torch.int)
+        temp = torch.bitwise_left_shift(temp[:,:,:,0], 16) + torch.bitwise_left_shift(temp[:,:,:,1], 8) + temp[:,:,:,2]
+        mask = torch.where(temp == color, 255, 0).float()
+        return (mask,)
+
+class SolidMask:
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "value": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01}),
+                "width": ("INT", {"default": 512, "min": 1, "max": MAX_RESOLUTION, "step": 1}),
+                "height": ("INT", {"default": 512, "min": 1, "max": MAX_RESOLUTION, "step": 1}),
+            }
+        }
+
+    CATEGORY = "mask"
+
+    RETURN_TYPES = ("MASK",)
+
+    FUNCTION = "solid"
+
+    def solid(self, value, width, height):
+        out = torch.full((1, height, width), value, dtype=torch.float32, device="cpu")
+        return (out,)
+
+class InvertMask:
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "mask": ("MASK",),
+            }
+        }
+
+    CATEGORY = "mask"
+
+    RETURN_TYPES = ("MASK",)
+
+    FUNCTION = "invert"
+
+    def invert(self, mask):
+        out = 1.0 - mask
+        return (out,)
+
+class CropMask:
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "mask": ("MASK",),
+                "x": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION, "step": 1}),
+                "y": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION, "step": 1}),
+                "width": ("INT", {"default": 512, "min": 1, "max": MAX_RESOLUTION, "step": 1}),
+                "height": ("INT", {"default": 512, "min": 1, "max": MAX_RESOLUTION, "step": 1}),
+            }
+        }
+
+    CATEGORY = "mask"
+
+    RETURN_TYPES = ("MASK",)
+
+    FUNCTION = "crop"
+
+    def crop(self, mask, x, y, width, height):
+        mask = mask.reshape((-1, mask.shape[-2], mask.shape[-1]))
+        out = mask[:, y:y + height, x:x + width]
+        return (out,)
+
+class MaskComposite:
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "destination": ("MASK",),
+                "source": ("MASK",),
+                "x": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION, "step": 1}),
+                "y": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION, "step": 1}),
+                "operation": (["multiply", "add", "subtract", "and", "or", "xor"],),
+            }
+        }
+
+    CATEGORY = "mask"
+
+    RETURN_TYPES = ("MASK",)
+
+    FUNCTION = "combine"
+
+    def combine(self, destination, source, x, y, operation):
+        output = destination.reshape((-1, destination.shape[-2], destination.shape[-1])).clone()
+        source = source.reshape((-1, source.shape[-2], source.shape[-1]))
+
+        left, top = (x, y,)
+        right, bottom = (min(left + source.shape[-1], destination.shape[-1]), min(top + source.shape[-2], destination.shape[-2]))
+        visible_width, visible_height = (right - left, bottom - top,)
+
+        source_portion = source[:, :visible_height, :visible_width]
+        destination_portion = destination[:, top:bottom, left:right]
+
+        if operation == "multiply":
+            output[:, top:bottom, left:right] = destination_portion * source_portion
+        elif operation == "add":
+            output[:, top:bottom, left:right] = destination_portion + source_portion
+        elif operation == "subtract":
+            output[:, top:bottom, left:right] = destination_portion - source_portion
+        elif operation == "and":
+            output[:, top:bottom, left:right] = torch.bitwise_and(destination_portion.round().bool(), source_portion.round().bool()).float()
+        elif operation == "or":
+            output[:, top:bottom, left:right] = torch.bitwise_or(destination_portion.round().bool(), source_portion.round().bool()).float()
+        elif operation == "xor":
+            output[:, top:bottom, left:right] = torch.bitwise_xor(destination_portion.round().bool(), source_portion.round().bool()).float()
+
+        output = torch.clamp(output, 0.0, 1.0)
+
+        return (output,)
+
+class FeatherMask:
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "mask": ("MASK",),
+                "left": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION, "step": 1}),
+                "top": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION, "step": 1}),
+                "right": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION, "step": 1}),
+                "bottom": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION, "step": 1}),
+            }
+        }
+
+    CATEGORY = "mask"
+
+    RETURN_TYPES = ("MASK",)
+
+    FUNCTION = "feather"
+
+    def feather(self, mask, left, top, right, bottom):
+        output = mask.reshape((-1, mask.shape[-2], mask.shape[-1])).clone()
+
+        left = min(left, output.shape[-1])
+        right = min(right, output.shape[-1])
+        top = min(top, output.shape[-2])
+        bottom = min(bottom, output.shape[-2])
+
+        for x in range(left):
+            feather_rate = (x + 1.0) / left
+            output[:, :, x] *= feather_rate
+
+        for x in range(right):
+            feather_rate = (x + 1) / right
+            output[:, :, -x] *= feather_rate
+
+        for y in range(top):
+            feather_rate = (y + 1) / top
+            output[:, y, :] *= feather_rate
+
+        for y in range(bottom):
+            feather_rate = (y + 1) / bottom
+            output[:, -y, :] *= feather_rate
+
+        return (output,)
+    
+class GrowMask:
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "mask": ("MASK",),
+                "expand": ("INT", {"default": 0, "min": -MAX_RESOLUTION, "max": MAX_RESOLUTION, "step": 1}),
+                "tapered_corners": ("BOOLEAN", {"default": True}),
+            },
+        }
+    
+    CATEGORY = "mask"
+
+    RETURN_TYPES = ("MASK",)
+
+    FUNCTION = "expand_mask"
+
+    def expand_mask(self, mask, expand, tapered_corners):
+        c = 0 if tapered_corners else 1
+        kernel = np.array([[c, 1, c],
+                           [1, 1, 1],
+                           [c, 1, c]])
+        mask = mask.reshape((-1, mask.shape[-2], mask.shape[-1]))
+        out = []
+        for m in mask:
+            output = m.numpy()
+            for _ in range(abs(expand)):
+                if expand < 0:
+                    output = scipy.ndimage.grey_erosion(output, footprint=kernel)
+                else:
+                    output = scipy.ndimage.grey_dilation(output, footprint=kernel)
+            output = torch.from_numpy(output)
+            out.append(output)
+        return (torch.stack(out, dim=0),)
+
+
+
+NODE_CLASS_MAPPINGS = {
+    "LatentCompositeMasked": LatentCompositeMasked,
+    "ImageCompositeMasked": ImageCompositeMasked,
+    "MaskToImage": MaskToImage,
+    "ImageToMask": ImageToMask,
+    "ImageColorToMask": ImageColorToMask,
+    "SolidMask": SolidMask,
+    "InvertMask": InvertMask,
+    "CropMask": CropMask,
+    "MaskComposite": MaskComposite,
+    "FeatherMask": FeatherMask,
+    "GrowMask": GrowMask,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "ImageToMask": "Convert Image to Mask",
+    "MaskToImage": "Convert Mask to Image",
+}

ldm_patched/contrib/external_model_advanced.py

@@ -0,0 +1,177 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+import ldm_patched.utils.path_utils
+import ldm_patched.modules.sd
+import ldm_patched.modules.model_sampling
+import torch
+
+class LCM(ldm_patched.modules.model_sampling.EPS):
+    def calculate_denoised(self, sigma, model_output, model_input):
+        timestep = self.timestep(sigma).view(sigma.shape[:1] + (1,) * (model_output.ndim - 1))
+        sigma = sigma.view(sigma.shape[:1] + (1,) * (model_output.ndim - 1))
+        x0 = model_input - model_output * sigma
+
+        sigma_data = 0.5
+        scaled_timestep = timestep * 10.0 #timestep_scaling
+
+        c_skip = sigma_data**2 / (scaled_timestep**2 + sigma_data**2)
+        c_out = scaled_timestep / (scaled_timestep**2 + sigma_data**2) ** 0.5
+
+        return c_out * x0 + c_skip * model_input
+
+class ModelSamplingDiscreteDistilled(ldm_patched.modules.model_sampling.ModelSamplingDiscrete):
+    original_timesteps = 50
+
+    def __init__(self, model_config=None):
+        super().__init__(model_config)
+
+        self.skip_steps = self.num_timesteps // self.original_timesteps
+
+        sigmas_valid = torch.zeros((self.original_timesteps), dtype=torch.float32)
+        for x in range(self.original_timesteps):
+            sigmas_valid[self.original_timesteps - 1 - x] = self.sigmas[self.num_timesteps - 1 - x * self.skip_steps]
+
+        self.set_sigmas(sigmas_valid)
+
+    def timestep(self, sigma):
+        log_sigma = sigma.log()
+        dists = log_sigma.to(self.log_sigmas.device) - self.log_sigmas[:, None]
+        return (dists.abs().argmin(dim=0).view(sigma.shape) * self.skip_steps + (self.skip_steps - 1)).to(sigma.device)
+
+    def sigma(self, timestep):
+        t = torch.clamp(((timestep.float().to(self.log_sigmas.device) - (self.skip_steps - 1)) / self.skip_steps).float(), min=0, max=(len(self.sigmas) - 1))
+        low_idx = t.floor().long()
+        high_idx = t.ceil().long()
+        w = t.frac()
+        log_sigma = (1 - w) * self.log_sigmas[low_idx] + w * self.log_sigmas[high_idx]
+        return log_sigma.exp().to(timestep.device)
+
+
+def rescale_zero_terminal_snr_sigmas(sigmas):
+    alphas_cumprod = 1 / ((sigmas * sigmas) + 1)
+    alphas_bar_sqrt = alphas_cumprod.sqrt()
+
+    # Store old values.
+    alphas_bar_sqrt_0 = alphas_bar_sqrt[0].clone()
+    alphas_bar_sqrt_T = alphas_bar_sqrt[-1].clone()
+
+    # Shift so the last timestep is zero.
+    alphas_bar_sqrt -= (alphas_bar_sqrt_T)
+
+    # Scale so the first timestep is back to the old value.
+    alphas_bar_sqrt *= alphas_bar_sqrt_0 / (alphas_bar_sqrt_0 - alphas_bar_sqrt_T)
+
+    # Convert alphas_bar_sqrt to betas
+    alphas_bar = alphas_bar_sqrt**2  # Revert sqrt
+    alphas_bar[-1] = 4.8973451890853435e-08
+    return ((1 - alphas_bar) / alphas_bar) ** 0.5
+
+class ModelSamplingDiscrete:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model": ("MODEL",),
+                              "sampling": (["eps", "v_prediction", "lcm"],),
+                              "zsnr": ("BOOLEAN", {"default": False}),
+                              }}
+
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "patch"
+
+    CATEGORY = "advanced/model"
+
+    def patch(self, model, sampling, zsnr):
+        m = model.clone()
+
+        sampling_base = ldm_patched.modules.model_sampling.ModelSamplingDiscrete
+        if sampling == "eps":
+            sampling_type = ldm_patched.modules.model_sampling.EPS
+        elif sampling == "v_prediction":
+            sampling_type = ldm_patched.modules.model_sampling.V_PREDICTION
+        elif sampling == "lcm":
+            sampling_type = LCM
+            sampling_base = ModelSamplingDiscreteDistilled
+
+        class ModelSamplingAdvanced(sampling_base, sampling_type):
+            pass
+
+        model_sampling = ModelSamplingAdvanced(model.model.model_config)
+        if zsnr:
+            model_sampling.set_sigmas(rescale_zero_terminal_snr_sigmas(model_sampling.sigmas))
+
+        m.add_object_patch("model_sampling", model_sampling)
+        return (m, )
+
+class ModelSamplingContinuousEDM:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model": ("MODEL",),
+                              "sampling": (["v_prediction", "eps"],),
+                              "sigma_max": ("FLOAT", {"default": 120.0, "min": 0.0, "max": 1000.0, "step":0.001, "round": False}),
+                              "sigma_min": ("FLOAT", {"default": 0.002, "min": 0.0, "max": 1000.0, "step":0.001, "round": False}),
+                              }}
+
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "patch"
+
+    CATEGORY = "advanced/model"
+
+    def patch(self, model, sampling, sigma_max, sigma_min):
+        m = model.clone()
+
+        if sampling == "eps":
+            sampling_type = ldm_patched.modules.model_sampling.EPS
+        elif sampling == "v_prediction":
+            sampling_type = ldm_patched.modules.model_sampling.V_PREDICTION
+
+        class ModelSamplingAdvanced(ldm_patched.modules.model_sampling.ModelSamplingContinuousEDM, sampling_type):
+            pass
+
+        model_sampling = ModelSamplingAdvanced(model.model.model_config)
+        model_sampling.set_sigma_range(sigma_min, sigma_max)
+        m.add_object_patch("model_sampling", model_sampling)
+        return (m, )
+
+class RescaleCFG:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model": ("MODEL",),
+                              "multiplier": ("FLOAT", {"default": 0.7, "min": 0.0, "max": 1.0, "step": 0.01}),
+                              }}
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "patch"
+
+    CATEGORY = "advanced/model"
+
+    def patch(self, model, multiplier):
+        def rescale_cfg(args):
+            cond = args["cond"]
+            uncond = args["uncond"]
+            cond_scale = args["cond_scale"]
+            sigma = args["sigma"]
+            sigma = sigma.view(sigma.shape[:1] + (1,) * (cond.ndim - 1))
+            x_orig = args["input"]
+
+            #rescale cfg has to be done on v-pred model output
+            x = x_orig / (sigma * sigma + 1.0)
+            cond = ((x - (x_orig - cond)) * (sigma ** 2 + 1.0) ** 0.5) / (sigma)
+            uncond = ((x - (x_orig - uncond)) * (sigma ** 2 + 1.0) ** 0.5) / (sigma)
+
+            #rescalecfg
+            x_cfg = uncond + cond_scale * (cond - uncond)
+            ro_pos = torch.std(cond, dim=(1,2,3), keepdim=True)
+            ro_cfg = torch.std(x_cfg, dim=(1,2,3), keepdim=True)
+
+            x_rescaled = x_cfg * (ro_pos / ro_cfg)
+            x_final = multiplier * x_rescaled + (1.0 - multiplier) * x_cfg
+
+            return x_orig - (x - x_final * sigma / (sigma * sigma + 1.0) ** 0.5)
+
+        m = model.clone()
+        m.set_model_sampler_cfg_function(rescale_cfg)
+        return (m, )
+
+NODE_CLASS_MAPPINGS = {
+    "ModelSamplingDiscrete": ModelSamplingDiscrete,
+    "ModelSamplingContinuousEDM": ModelSamplingContinuousEDM,
+    "RescaleCFG": RescaleCFG,
+}

ldm_patched/contrib/external_model_downscale.py

@@ -0,0 +1,55 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+import torch
+import ldm_patched.modules.utils
+
+class PatchModelAddDownscale:
+    upscale_methods = ["bicubic", "nearest-exact", "bilinear", "area", "bislerp"]
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model": ("MODEL",),
+                              "block_number": ("INT", {"default": 3, "min": 1, "max": 32, "step": 1}),
+                              "downscale_factor": ("FLOAT", {"default": 2.0, "min": 0.1, "max": 9.0, "step": 0.001}),
+                              "start_percent": ("FLOAT", {"default": 0.0, "min": 0.0, "max": 1.0, "step": 0.001}),
+                              "end_percent": ("FLOAT", {"default": 0.35, "min": 0.0, "max": 1.0, "step": 0.001}),
+                              "downscale_after_skip": ("BOOLEAN", {"default": True}),
+                              "downscale_method": (s.upscale_methods,),
+                              "upscale_method": (s.upscale_methods,),
+                              }}
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "patch"
+
+    CATEGORY = "_for_testing"
+
+    def patch(self, model, block_number, downscale_factor, start_percent, end_percent, downscale_after_skip, downscale_method, upscale_method):
+        sigma_start = model.model.model_sampling.percent_to_sigma(start_percent)
+        sigma_end = model.model.model_sampling.percent_to_sigma(end_percent)
+
+        def input_block_patch(h, transformer_options):
+            if transformer_options["block"][1] == block_number:
+                sigma = transformer_options["sigmas"][0].item()
+                if sigma <= sigma_start and sigma >= sigma_end:
+                    h = ldm_patched.modules.utils.common_upscale(h, round(h.shape[-1] * (1.0 / downscale_factor)), round(h.shape[-2] * (1.0 / downscale_factor)), downscale_method, "disabled")
+            return h
+
+        def output_block_patch(h, hsp, transformer_options):
+            if h.shape[2] != hsp.shape[2]:
+                h = ldm_patched.modules.utils.common_upscale(h, hsp.shape[-1], hsp.shape[-2], upscale_method, "disabled")
+            return h, hsp
+
+        m = model.clone()
+        if downscale_after_skip:
+            m.set_model_input_block_patch_after_skip(input_block_patch)
+        else:
+            m.set_model_input_block_patch(input_block_patch)
+        m.set_model_output_block_patch(output_block_patch)
+        return (m, )
+
+NODE_CLASS_MAPPINGS = {
+    "PatchModelAddDownscale": PatchModelAddDownscale,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    # Sampling
+    "PatchModelAddDownscale": "PatchModelAddDownscale (Kohya Deep Shrink)",
+}

ldm_patched/contrib/external_model_merging.py

@@ -0,0 +1,283 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+import ldm_patched.modules.sd
+import ldm_patched.modules.utils
+import ldm_patched.modules.model_base
+import ldm_patched.modules.model_management
+
+import ldm_patched.utils.path_utils
+import json
+import os
+
+from ldm_patched.modules.args_parser import args
+
+class ModelMergeSimple:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model1": ("MODEL",),
+                              "model2": ("MODEL",),
+                              "ratio": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01}),
+                              }}
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "merge"
+
+    CATEGORY = "advanced/model_merging"
+
+    def merge(self, model1, model2, ratio):
+        m = model1.clone()
+        kp = model2.get_key_patches("diffusion_model.")
+        for k in kp:
+            m.add_patches({k: kp[k]}, 1.0 - ratio, ratio)
+        return (m, )
+
+class ModelSubtract:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model1": ("MODEL",),
+                              "model2": ("MODEL",),
+                              "multiplier": ("FLOAT", {"default": 1.0, "min": -10.0, "max": 10.0, "step": 0.01}),
+                              }}
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "merge"
+
+    CATEGORY = "advanced/model_merging"
+
+    def merge(self, model1, model2, multiplier):
+        m = model1.clone()
+        kp = model2.get_key_patches("diffusion_model.")
+        for k in kp:
+            m.add_patches({k: kp[k]}, - multiplier, multiplier)
+        return (m, )
+
+class ModelAdd:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model1": ("MODEL",),
+                              "model2": ("MODEL",),
+                              }}
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "merge"
+
+    CATEGORY = "advanced/model_merging"
+
+    def merge(self, model1, model2):
+        m = model1.clone()
+        kp = model2.get_key_patches("diffusion_model.")
+        for k in kp:
+            m.add_patches({k: kp[k]}, 1.0, 1.0)
+        return (m, )
+
+
+class CLIPMergeSimple:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "clip1": ("CLIP",),
+                              "clip2": ("CLIP",),
+                              "ratio": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01}),
+                              }}
+    RETURN_TYPES = ("CLIP",)
+    FUNCTION = "merge"
+
+    CATEGORY = "advanced/model_merging"
+
+    def merge(self, clip1, clip2, ratio):
+        m = clip1.clone()
+        kp = clip2.get_key_patches()
+        for k in kp:
+            if k.endswith(".position_ids") or k.endswith(".logit_scale"):
+                continue
+            m.add_patches({k: kp[k]}, 1.0 - ratio, ratio)
+        return (m, )
+
+class ModelMergeBlocks:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model1": ("MODEL",),
+                              "model2": ("MODEL",),
+                              "input": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01}),
+                              "middle": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01}),
+                              "out": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01})
+                              }}
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "merge"
+
+    CATEGORY = "advanced/model_merging"
+
+    def merge(self, model1, model2, **kwargs):
+        m = model1.clone()
+        kp = model2.get_key_patches("diffusion_model.")
+        default_ratio = next(iter(kwargs.values()))
+
+        for k in kp:
+            ratio = default_ratio
+            k_unet = k[len("diffusion_model."):]
+
+            last_arg_size = 0
+            for arg in kwargs:
+                if k_unet.startswith(arg) and last_arg_size < len(arg):
+                    ratio = kwargs[arg]
+                    last_arg_size = len(arg)
+
+            m.add_patches({k: kp[k]}, 1.0 - ratio, ratio)
+        return (m, )
+
+class CheckpointSave:
+    def __init__(self):
+        self.output_dir = ldm_patched.utils.path_utils.get_output_directory()
+
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model": ("MODEL",),
+                              "clip": ("CLIP",),
+                              "vae": ("VAE",),
+                              "filename_prefix": ("STRING", {"default": "checkpoints/ldm_patched"}),},
+                "hidden": {"prompt": "PROMPT", "extra_pnginfo": "EXTRA_PNGINFO"},}
+    RETURN_TYPES = ()
+    FUNCTION = "save"
+    OUTPUT_NODE = True
+
+    CATEGORY = "advanced/model_merging"
+
+    def save(self, model, clip, vae, filename_prefix, prompt=None, extra_pnginfo=None):
+        full_output_folder, filename, counter, subfolder, filename_prefix = ldm_patched.utils.path_utils.get_save_image_path(filename_prefix, self.output_dir)
+        prompt_info = ""
+        if prompt is not None:
+            prompt_info = json.dumps(prompt)
+
+        metadata = {}
+
+        enable_modelspec = True
+        if isinstance(model.model, ldm_patched.modules.model_base.SDXL):
+            metadata["modelspec.architecture"] = "stable-diffusion-xl-v1-base"
+        elif isinstance(model.model, ldm_patched.modules.model_base.SDXLRefiner):
+            metadata["modelspec.architecture"] = "stable-diffusion-xl-v1-refiner"
+        else:
+            enable_modelspec = False
+
+        if enable_modelspec:
+            metadata["modelspec.sai_model_spec"] = "1.0.0"
+            metadata["modelspec.implementation"] = "sgm"
+            metadata["modelspec.title"] = "{} {}".format(filename, counter)
+
+        #TODO:
+        # "stable-diffusion-v1", "stable-diffusion-v1-inpainting", "stable-diffusion-v2-512",
+        # "stable-diffusion-v2-768-v", "stable-diffusion-v2-unclip-l", "stable-diffusion-v2-unclip-h",
+        # "v2-inpainting"
+
+        if model.model.model_type == ldm_patched.modules.model_base.ModelType.EPS:
+            metadata["modelspec.predict_key"] = "epsilon"
+        elif model.model.model_type == ldm_patched.modules.model_base.ModelType.V_PREDICTION:
+            metadata["modelspec.predict_key"] = "v"
+
+        if not args.disable_server_info:
+            metadata["prompt"] = prompt_info
+            if extra_pnginfo is not None:
+                for x in extra_pnginfo:
+                    metadata[x] = json.dumps(extra_pnginfo[x])
+
+        output_checkpoint = f"{filename}_{counter:05}_.safetensors"
+        output_checkpoint = os.path.join(full_output_folder, output_checkpoint)
+
+        ldm_patched.modules.sd.save_checkpoint(output_checkpoint, model, clip, vae, metadata=metadata)
+        return {}
+
+class CLIPSave:
+    def __init__(self):
+        self.output_dir = ldm_patched.utils.path_utils.get_output_directory()
+
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "clip": ("CLIP",),
+                              "filename_prefix": ("STRING", {"default": "clip/ldm_patched"}),},
+                "hidden": {"prompt": "PROMPT", "extra_pnginfo": "EXTRA_PNGINFO"},}
+    RETURN_TYPES = ()
+    FUNCTION = "save"
+    OUTPUT_NODE = True
+
+    CATEGORY = "advanced/model_merging"
+
+    def save(self, clip, filename_prefix, prompt=None, extra_pnginfo=None):
+        prompt_info = ""
+        if prompt is not None:
+            prompt_info = json.dumps(prompt)
+
+        metadata = {}
+        if not args.disable_server_info:
+            metadata["prompt"] = prompt_info
+            if extra_pnginfo is not None:
+                for x in extra_pnginfo:
+                    metadata[x] = json.dumps(extra_pnginfo[x])
+
+        ldm_patched.modules.model_management.load_models_gpu([clip.load_model()])
+        clip_sd = clip.get_sd()
+
+        for prefix in ["clip_l.", "clip_g.", ""]:
+            k = list(filter(lambda a: a.startswith(prefix), clip_sd.keys()))
+            current_clip_sd = {}
+            for x in k:
+                current_clip_sd[x] = clip_sd.pop(x)
+            if len(current_clip_sd) == 0:
+                continue
+
+            p = prefix[:-1]
+            replace_prefix = {}
+            filename_prefix_ = filename_prefix
+            if len(p) > 0:
+                filename_prefix_ = "{}_{}".format(filename_prefix_, p)
+                replace_prefix[prefix] = ""
+            replace_prefix["transformer."] = ""
+
+            full_output_folder, filename, counter, subfolder, filename_prefix_ = ldm_patched.utils.path_utils.get_save_image_path(filename_prefix_, self.output_dir)
+
+            output_checkpoint = f"{filename}_{counter:05}_.safetensors"
+            output_checkpoint = os.path.join(full_output_folder, output_checkpoint)
+
+            current_clip_sd = ldm_patched.modules.utils.state_dict_prefix_replace(current_clip_sd, replace_prefix)
+
+            ldm_patched.modules.utils.save_torch_file(current_clip_sd, output_checkpoint, metadata=metadata)
+        return {}
+
+class VAESave:
+    def __init__(self):
+        self.output_dir = ldm_patched.utils.path_utils.get_output_directory()
+
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "vae": ("VAE",),
+                              "filename_prefix": ("STRING", {"default": "vae/ldm_patched_vae"}),},
+                "hidden": {"prompt": "PROMPT", "extra_pnginfo": "EXTRA_PNGINFO"},}
+    RETURN_TYPES = ()
+    FUNCTION = "save"
+    OUTPUT_NODE = True
+
+    CATEGORY = "advanced/model_merging"
+
+    def save(self, vae, filename_prefix, prompt=None, extra_pnginfo=None):
+        full_output_folder, filename, counter, subfolder, filename_prefix = ldm_patched.utils.path_utils.get_save_image_path(filename_prefix, self.output_dir)
+        prompt_info = ""
+        if prompt is not None:
+            prompt_info = json.dumps(prompt)
+
+        metadata = {}
+        if not args.disable_server_info:
+            metadata["prompt"] = prompt_info
+            if extra_pnginfo is not None:
+                for x in extra_pnginfo:
+                    metadata[x] = json.dumps(extra_pnginfo[x])
+
+        output_checkpoint = f"{filename}_{counter:05}_.safetensors"
+        output_checkpoint = os.path.join(full_output_folder, output_checkpoint)
+
+        ldm_patched.modules.utils.save_torch_file(vae.get_sd(), output_checkpoint, metadata=metadata)
+        return {}
+
+NODE_CLASS_MAPPINGS = {
+    "ModelMergeSimple": ModelMergeSimple,
+    "ModelMergeBlocks": ModelMergeBlocks,
+    "ModelMergeSubtract": ModelSubtract,
+    "ModelMergeAdd": ModelAdd,
+    "CheckpointSave": CheckpointSave,
+    "CLIPMergeSimple": CLIPMergeSimple,
+    "CLIPSave": CLIPSave,
+    "VAESave": VAESave,
+}

ldm_patched/contrib/external_perpneg.py

@@ -0,0 +1,57 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+import torch
+import ldm_patched.modules.model_management
+import ldm_patched.modules.sample
+import ldm_patched.modules.samplers
+import ldm_patched.modules.utils
+
+
+class PerpNeg:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": {"model": ("MODEL", ),
+                             "empty_conditioning": ("CONDITIONING", ),
+                             "neg_scale": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0}),
+                            }}
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "patch"
+
+    CATEGORY = "_for_testing"
+
+    def patch(self, model, empty_conditioning, neg_scale):
+        m = model.clone()
+        nocond = ldm_patched.modules.sample.convert_cond(empty_conditioning)
+
+        def cfg_function(args):
+            model = args["model"]
+            noise_pred_pos = args["cond_denoised"]
+            noise_pred_neg = args["uncond_denoised"]
+            cond_scale = args["cond_scale"]
+            x = args["input"]
+            sigma = args["sigma"]
+            model_options = args["model_options"]
+            nocond_processed = ldm_patched.modules.samplers.encode_model_conds(model.extra_conds, nocond, x, x.device, "negative")
+
+            (noise_pred_nocond, _) = ldm_patched.modules.samplers.calc_cond_uncond_batch(model, nocond_processed, None, x, sigma, model_options)
+
+            pos = noise_pred_pos - noise_pred_nocond
+            neg = noise_pred_neg - noise_pred_nocond
+            perp = ((torch.mul(pos, neg).sum())/(torch.norm(neg)**2)) * neg
+            perp_neg = perp * neg_scale
+            cfg_result = noise_pred_nocond + cond_scale*(pos - perp_neg)
+            cfg_result = x - cfg_result
+            return cfg_result
+
+        m.set_model_sampler_cfg_function(cfg_function)
+
+        return (m, )
+
+
+NODE_CLASS_MAPPINGS = {
+    "PerpNeg": PerpNeg,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "PerpNeg": "Perp-Neg",
+}

ldm_patched/contrib/external_post_processing.py

@@ -0,0 +1,277 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+import numpy as np
+import torch
+import torch.nn.functional as F
+from PIL import Image
+import math
+
+import ldm_patched.modules.utils
+
+
+class Blend:
+    def __init__(self):
+        pass
+
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+                "image1": ("IMAGE",),
+                "image2": ("IMAGE",),
+                "blend_factor": ("FLOAT", {
+                    "default": 0.5,
+                    "min": 0.0,
+                    "max": 1.0,
+                    "step": 0.01
+                }),
+                "blend_mode": (["normal", "multiply", "screen", "overlay", "soft_light", "difference"],),
+            },
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "blend_images"
+
+    CATEGORY = "image/postprocessing"
+
+    def blend_images(self, image1: torch.Tensor, image2: torch.Tensor, blend_factor: float, blend_mode: str):
+        if image1.shape != image2.shape:
+            image2 = image2.permute(0, 3, 1, 2)
+            image2 = ldm_patched.modules.utils.common_upscale(image2, image1.shape[2], image1.shape[1], upscale_method='bicubic', crop='center')
+            image2 = image2.permute(0, 2, 3, 1)
+
+        blended_image = self.blend_mode(image1, image2, blend_mode)
+        blended_image = image1 * (1 - blend_factor) + blended_image * blend_factor
+        blended_image = torch.clamp(blended_image, 0, 1)
+        return (blended_image,)
+
+    def blend_mode(self, img1, img2, mode):
+        if mode == "normal":
+            return img2
+        elif mode == "multiply":
+            return img1 * img2
+        elif mode == "screen":
+            return 1 - (1 - img1) * (1 - img2)
+        elif mode == "overlay":
+            return torch.where(img1 <= 0.5, 2 * img1 * img2, 1 - 2 * (1 - img1) * (1 - img2))
+        elif mode == "soft_light":
+            return torch.where(img2 <= 0.5, img1 - (1 - 2 * img2) * img1 * (1 - img1), img1 + (2 * img2 - 1) * (self.g(img1) - img1))
+        elif mode == "difference":
+            return img1 - img2
+        else:
+            raise ValueError(f"Unsupported blend mode: {mode}")
+
+    def g(self, x):
+        return torch.where(x <= 0.25, ((16 * x - 12) * x + 4) * x, torch.sqrt(x))
+
+def gaussian_kernel(kernel_size: int, sigma: float, device=None):
+    x, y = torch.meshgrid(torch.linspace(-1, 1, kernel_size, device=device), torch.linspace(-1, 1, kernel_size, device=device), indexing="ij")
+    d = torch.sqrt(x * x + y * y)
+    g = torch.exp(-(d * d) / (2.0 * sigma * sigma))
+    return g / g.sum()
+
+class Blur:
+    def __init__(self):
+        pass
+
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+                "image": ("IMAGE",),
+                "blur_radius": ("INT", {
+                    "default": 1,
+                    "min": 1,
+                    "max": 31,
+                    "step": 1
+                }),
+                "sigma": ("FLOAT", {
+                    "default": 1.0,
+                    "min": 0.1,
+                    "max": 10.0,
+                    "step": 0.1
+                }),
+            },
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "blur"
+
+    CATEGORY = "image/postprocessing"
+
+    def blur(self, image: torch.Tensor, blur_radius: int, sigma: float):
+        if blur_radius == 0:
+            return (image,)
+
+        batch_size, height, width, channels = image.shape
+
+        kernel_size = blur_radius * 2 + 1
+        kernel = gaussian_kernel(kernel_size, sigma, device=image.device).repeat(channels, 1, 1).unsqueeze(1)
+
+        image = image.permute(0, 3, 1, 2) # Torch wants (B, C, H, W) we use (B, H, W, C)
+        padded_image = F.pad(image, (blur_radius,blur_radius,blur_radius,blur_radius), 'reflect')
+        blurred = F.conv2d(padded_image, kernel, padding=kernel_size // 2, groups=channels)[:,:,blur_radius:-blur_radius, blur_radius:-blur_radius]
+        blurred = blurred.permute(0, 2, 3, 1)
+
+        return (blurred,)
+
+class Quantize:
+    def __init__(self):
+        pass
+
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+                "image": ("IMAGE",),
+                "colors": ("INT", {
+                    "default": 256,
+                    "min": 1,
+                    "max": 256,
+                    "step": 1
+                }),
+                "dither": (["none", "floyd-steinberg", "bayer-2", "bayer-4", "bayer-8", "bayer-16"],),
+            },
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "quantize"
+
+    CATEGORY = "image/postprocessing"
+
+    def bayer(im, pal_im, order):
+        def normalized_bayer_matrix(n):
+            if n == 0:
+                return np.zeros((1,1), "float32")
+            else:
+                q = 4 ** n
+                m = q * normalized_bayer_matrix(n - 1)
+                return np.bmat(((m-1.5, m+0.5), (m+1.5, m-0.5))) / q
+
+        num_colors = len(pal_im.getpalette()) // 3
+        spread = 2 * 256 / num_colors
+        bayer_n = int(math.log2(order))
+        bayer_matrix = torch.from_numpy(spread * normalized_bayer_matrix(bayer_n) + 0.5)
+
+        result = torch.from_numpy(np.array(im).astype(np.float32))
+        tw = math.ceil(result.shape[0] / bayer_matrix.shape[0])
+        th = math.ceil(result.shape[1] / bayer_matrix.shape[1])
+        tiled_matrix = bayer_matrix.tile(tw, th).unsqueeze(-1)
+        result.add_(tiled_matrix[:result.shape[0],:result.shape[1]]).clamp_(0, 255)
+        result = result.to(dtype=torch.uint8)
+
+        im = Image.fromarray(result.cpu().numpy())
+        im = im.quantize(palette=pal_im, dither=Image.Dither.NONE)
+        return im
+
+    def quantize(self, image: torch.Tensor, colors: int, dither: str):
+        batch_size, height, width, _ = image.shape
+        result = torch.zeros_like(image)
+
+        for b in range(batch_size):
+            im = Image.fromarray((image[b] * 255).to(torch.uint8).numpy(), mode='RGB')
+
+            pal_im = im.quantize(colors=colors) # Required as described in https://github.com/python-pillow/Pillow/issues/5836
+
+            if dither == "none":
+                quantized_image = im.quantize(palette=pal_im, dither=Image.Dither.NONE)
+            elif dither == "floyd-steinberg":
+                quantized_image = im.quantize(palette=pal_im, dither=Image.Dither.FLOYDSTEINBERG)
+            elif dither.startswith("bayer"):
+                order = int(dither.split('-')[-1])
+                quantized_image = Quantize.bayer(im, pal_im, order)
+
+            quantized_array = torch.tensor(np.array(quantized_image.convert("RGB"))).float() / 255
+            result[b] = quantized_array
+
+        return (result,)
+
+class Sharpen:
+    def __init__(self):
+        pass
+
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+                "image": ("IMAGE",),
+                "sharpen_radius": ("INT", {
+                    "default": 1,
+                    "min": 1,
+                    "max": 31,
+                    "step": 1
+                }),
+                "sigma": ("FLOAT", {
+                    "default": 1.0,
+                    "min": 0.1,
+                    "max": 10.0,
+                    "step": 0.1
+                }),
+                "alpha": ("FLOAT", {
+                    "default": 1.0,
+                    "min": 0.0,
+                    "max": 5.0,
+                    "step": 0.1
+                }),
+            },
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "sharpen"
+
+    CATEGORY = "image/postprocessing"
+
+    def sharpen(self, image: torch.Tensor, sharpen_radius: int, sigma:float, alpha: float):
+        if sharpen_radius == 0:
+            return (image,)
+
+        batch_size, height, width, channels = image.shape
+
+        kernel_size = sharpen_radius * 2 + 1
+        kernel = gaussian_kernel(kernel_size, sigma, device=image.device) * -(alpha*10)
+        center = kernel_size // 2
+        kernel[center, center] = kernel[center, center] - kernel.sum() + 1.0
+        kernel = kernel.repeat(channels, 1, 1).unsqueeze(1)
+
+        tensor_image = image.permute(0, 3, 1, 2) # Torch wants (B, C, H, W) we use (B, H, W, C)
+        tensor_image = F.pad(tensor_image, (sharpen_radius,sharpen_radius,sharpen_radius,sharpen_radius), 'reflect')
+        sharpened = F.conv2d(tensor_image, kernel, padding=center, groups=channels)[:,:,sharpen_radius:-sharpen_radius, sharpen_radius:-sharpen_radius]
+        sharpened = sharpened.permute(0, 2, 3, 1)
+
+        result = torch.clamp(sharpened, 0, 1)
+
+        return (result,)
+
+class ImageScaleToTotalPixels:
+    upscale_methods = ["nearest-exact", "bilinear", "area", "bicubic", "lanczos"]
+    crop_methods = ["disabled", "center"]
+
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "image": ("IMAGE",), "upscale_method": (s.upscale_methods,),
+                              "megapixels": ("FLOAT", {"default": 1.0, "min": 0.01, "max": 16.0, "step": 0.01}),
+                            }}
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "upscale"
+
+    CATEGORY = "image/upscaling"
+
+    def upscale(self, image, upscale_method, megapixels):
+        samples = image.movedim(-1,1)
+        total = int(megapixels * 1024 * 1024)
+
+        scale_by = math.sqrt(total / (samples.shape[3] * samples.shape[2]))
+        width = round(samples.shape[3] * scale_by)
+        height = round(samples.shape[2] * scale_by)
+
+        s = ldm_patched.modules.utils.common_upscale(samples, width, height, upscale_method, "disabled")
+        s = s.movedim(1,-1)
+        return (s,)
+
+NODE_CLASS_MAPPINGS = {
+    "ImageBlend": Blend,
+    "ImageBlur": Blur,
+    "ImageQuantize": Quantize,
+    "ImageSharpen": Sharpen,
+    "ImageScaleToTotalPixels": ImageScaleToTotalPixels,
+}

ldm_patched/contrib/external_rebatch.py

@@ -0,0 +1,110 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+import torch
+
+class LatentRebatch:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "latents": ("LATENT",),
+                              "batch_size": ("INT", {"default": 1, "min": 1, "max": 4096}),
+                              }}
+    RETURN_TYPES = ("LATENT",)
+    INPUT_IS_LIST = True
+    OUTPUT_IS_LIST = (True, )
+
+    FUNCTION = "rebatch"
+
+    CATEGORY = "latent/batch"
+
+    @staticmethod
+    def get_batch(latents, list_ind, offset):
+        '''prepare a batch out of the list of latents'''
+        samples = latents[list_ind]['samples']
+        shape = samples.shape
+        mask = latents[list_ind]['noise_mask'] if 'noise_mask' in latents[list_ind] else torch.ones((shape[0], 1, shape[2]*8, shape[3]*8), device='cpu')
+        if mask.shape[-1] != shape[-1] * 8 or mask.shape[-2] != shape[-2]:
+            torch.nn.functional.interpolate(mask.reshape((-1, 1, mask.shape[-2], mask.shape[-1])), size=(shape[-2]*8, shape[-1]*8), mode="bilinear")
+        if mask.shape[0] < samples.shape[0]:
+            mask = mask.repeat((shape[0] - 1) // mask.shape[0] + 1, 1, 1, 1)[:shape[0]]
+        if 'batch_index' in latents[list_ind]:
+            batch_inds = latents[list_ind]['batch_index']
+        else:
+            batch_inds = [x+offset for x in range(shape[0])]
+        return samples, mask, batch_inds
+
+    @staticmethod
+    def get_slices(indexable, num, batch_size):
+        '''divides an indexable object into num slices of length batch_size, and a remainder'''
+        slices = []
+        for i in range(num):
+            slices.append(indexable[i*batch_size:(i+1)*batch_size])
+        if num * batch_size < len(indexable):
+            return slices, indexable[num * batch_size:]
+        else:
+            return slices, None
+    
+    @staticmethod
+    def slice_batch(batch, num, batch_size):
+        result = [LatentRebatch.get_slices(x, num, batch_size) for x in batch]
+        return list(zip(*result))
+
+    @staticmethod
+    def cat_batch(batch1, batch2):
+        if batch1[0] is None:
+            return batch2
+        result = [torch.cat((b1, b2)) if torch.is_tensor(b1) else b1 + b2 for b1, b2 in zip(batch1, batch2)]
+        return result
+
+    def rebatch(self, latents, batch_size):
+        batch_size = batch_size[0]
+
+        output_list = []
+        current_batch = (None, None, None)
+        processed = 0
+
+        for i in range(len(latents)):
+            # fetch new entry of list
+            #samples, masks, indices = self.get_batch(latents, i)
+            next_batch = self.get_batch(latents, i, processed)
+            processed += len(next_batch[2])
+            # set to current if current is None
+            if current_batch[0] is None:
+                current_batch = next_batch
+            # add previous to list if dimensions do not match
+            elif next_batch[0].shape[-1] != current_batch[0].shape[-1] or next_batch[0].shape[-2] != current_batch[0].shape[-2]:
+                sliced, _ = self.slice_batch(current_batch, 1, batch_size)
+                output_list.append({'samples': sliced[0][0], 'noise_mask': sliced[1][0], 'batch_index': sliced[2][0]})
+                current_batch = next_batch
+            # cat if everything checks out
+            else:
+                current_batch = self.cat_batch(current_batch, next_batch)
+
+            # add to list if dimensions gone above target batch size
+            if current_batch[0].shape[0] > batch_size:
+                num = current_batch[0].shape[0] // batch_size
+                sliced, remainder = self.slice_batch(current_batch, num, batch_size)
+                
+                for i in range(num):
+                    output_list.append({'samples': sliced[0][i], 'noise_mask': sliced[1][i], 'batch_index': sliced[2][i]})
+
+                current_batch = remainder
+
+        #add remainder
+        if current_batch[0] is not None:
+            sliced, _ = self.slice_batch(current_batch, 1, batch_size)
+            output_list.append({'samples': sliced[0][0], 'noise_mask': sliced[1][0], 'batch_index': sliced[2][0]})
+
+        #get rid of empty masks
+        for s in output_list:
+            if s['noise_mask'].mean() == 1.0:
+                del s['noise_mask']
+
+        return (output_list,)
+
+NODE_CLASS_MAPPINGS = {
+    "RebatchLatents": LatentRebatch,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "RebatchLatents": "Rebatch Latents",
+}

ldm_patched/contrib/external_sag.py

@@ -0,0 +1,170 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+import torch
+from torch import einsum
+import torch.nn.functional as F
+import math
+
+from einops import rearrange, repeat
+import os
+from ldm_patched.ldm.modules.attention import optimized_attention, _ATTN_PRECISION
+import ldm_patched.modules.samplers
+
+# from ldm_patched.modules/ldm/modules/attention.py
+# but modified to return attention scores as well as output
+def attention_basic_with_sim(q, k, v, heads, mask=None):
+    b, _, dim_head = q.shape
+    dim_head //= heads
+    scale = dim_head ** -0.5
+
+    h = heads
+    q, k, v = map(
+        lambda t: t.unsqueeze(3)
+        .reshape(b, -1, heads, dim_head)
+        .permute(0, 2, 1, 3)
+        .reshape(b * heads, -1, dim_head)
+        .contiguous(),
+        (q, k, v),
+    )
+
+    # force cast to fp32 to avoid overflowing
+    if _ATTN_PRECISION =="fp32":
+        sim = einsum('b i d, b j d -> b i j', q.float(), k.float()) * scale
+    else:
+        sim = einsum('b i d, b j d -> b i j', q, k) * scale
+
+    del q, k
+
+    if mask is not None:
+        mask = rearrange(mask, 'b ... -> b (...)')
+        max_neg_value = -torch.finfo(sim.dtype).max
+        mask = repeat(mask, 'b j -> (b h) () j', h=h)
+        sim.masked_fill_(~mask, max_neg_value)
+
+    # attention, what we cannot get enough of
+    sim = sim.softmax(dim=-1)
+
+    out = einsum('b i j, b j d -> b i d', sim.to(v.dtype), v)
+    out = (
+        out.unsqueeze(0)
+        .reshape(b, heads, -1, dim_head)
+        .permute(0, 2, 1, 3)
+        .reshape(b, -1, heads * dim_head)
+    )
+    return (out, sim)
+
+def create_blur_map(x0, attn, sigma=3.0, threshold=1.0):
+    # reshape and GAP the attention map
+    _, hw1, hw2 = attn.shape
+    b, _, lh, lw = x0.shape
+    attn = attn.reshape(b, -1, hw1, hw2)
+    # Global Average Pool
+    mask = attn.mean(1, keepdim=False).sum(1, keepdim=False) > threshold
+    ratio = math.ceil(math.sqrt(lh * lw / hw1))
+    mid_shape = [math.ceil(lh / ratio), math.ceil(lw / ratio)]
+
+    # Reshape
+    mask = (
+        mask.reshape(b, *mid_shape)
+        .unsqueeze(1)
+        .type(attn.dtype)
+    )
+    # Upsample
+    mask = F.interpolate(mask, (lh, lw))
+
+    blurred = gaussian_blur_2d(x0, kernel_size=9, sigma=sigma)
+    blurred = blurred * mask + x0 * (1 - mask)
+    return blurred
+
+def gaussian_blur_2d(img, kernel_size, sigma):
+    ksize_half = (kernel_size - 1) * 0.5
+
+    x = torch.linspace(-ksize_half, ksize_half, steps=kernel_size)
+
+    pdf = torch.exp(-0.5 * (x / sigma).pow(2))
+
+    x_kernel = pdf / pdf.sum()
+    x_kernel = x_kernel.to(device=img.device, dtype=img.dtype)
+
+    kernel2d = torch.mm(x_kernel[:, None], x_kernel[None, :])
+    kernel2d = kernel2d.expand(img.shape[-3], 1, kernel2d.shape[0], kernel2d.shape[1])
+
+    padding = [kernel_size // 2, kernel_size // 2, kernel_size // 2, kernel_size // 2]
+
+    img = F.pad(img, padding, mode="reflect")
+    img = F.conv2d(img, kernel2d, groups=img.shape[-3])
+    return img
+
+class SelfAttentionGuidance:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model": ("MODEL",),
+                             "scale": ("FLOAT", {"default": 0.5, "min": -2.0, "max": 5.0, "step": 0.1}),
+                             "blur_sigma": ("FLOAT", {"default": 2.0, "min": 0.0, "max": 10.0, "step": 0.1}),
+                              }}
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "patch"
+
+    CATEGORY = "_for_testing"
+
+    def patch(self, model, scale, blur_sigma):
+        m = model.clone()
+
+        attn_scores = None
+
+        # TODO: make this work properly with chunked batches
+        #       currently, we can only save the attn from one UNet call
+        def attn_and_record(q, k, v, extra_options):
+            nonlocal attn_scores
+            # if uncond, save the attention scores
+            heads = extra_options["n_heads"]
+            cond_or_uncond = extra_options["cond_or_uncond"]
+            b = q.shape[0] // len(cond_or_uncond)
+            if 1 in cond_or_uncond:
+                uncond_index = cond_or_uncond.index(1)
+                # do the entire attention operation, but save the attention scores to attn_scores
+                (out, sim) = attention_basic_with_sim(q, k, v, heads=heads)
+                # when using a higher batch size, I BELIEVE the result batch dimension is [uc1, ... ucn, c1, ... cn]
+                n_slices = heads * b
+                attn_scores = sim[n_slices * uncond_index:n_slices * (uncond_index+1)]
+                return out
+            else:
+                return optimized_attention(q, k, v, heads=heads)
+
+        def post_cfg_function(args):
+            nonlocal attn_scores
+            uncond_attn = attn_scores
+
+            sag_scale = scale
+            sag_sigma = blur_sigma
+            sag_threshold = 1.0
+            model = args["model"]
+            uncond_pred = args["uncond_denoised"]
+            uncond = args["uncond"]
+            cfg_result = args["denoised"]
+            sigma = args["sigma"]
+            model_options = args["model_options"]
+            x = args["input"]
+
+            # create the adversarially blurred image
+            degraded = create_blur_map(uncond_pred, uncond_attn, sag_sigma, sag_threshold)
+            degraded_noised = degraded + x - uncond_pred
+            # call into the UNet
+            (sag, _) = ldm_patched.modules.samplers.calc_cond_uncond_batch(model, uncond, None, degraded_noised, sigma, model_options)
+            return cfg_result + (degraded - sag) * sag_scale
+
+        m.set_model_sampler_post_cfg_function(post_cfg_function)
+
+        # from diffusers:
+        # unet.mid_block.attentions[0].transformer_blocks[0].attn1.patch
+        m.set_model_attn1_replace(attn_and_record, "middle", 0, 0)
+
+        return (m, )
+
+NODE_CLASS_MAPPINGS = {
+    "SelfAttentionGuidance": SelfAttentionGuidance,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "SelfAttentionGuidance": "Self-Attention Guidance",
+}

ldm_patched/contrib/external_tomesd.py

@@ -0,0 +1,179 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+#Taken from: https://github.com/dbolya/tomesd
+
+import torch
+from typing import Tuple, Callable
+import math
+
+def do_nothing(x: torch.Tensor, mode:str=None):
+    return x
+
+
+def mps_gather_workaround(input, dim, index):
+    if input.shape[-1] == 1:
+        return torch.gather(
+            input.unsqueeze(-1),
+            dim - 1 if dim < 0 else dim,
+            index.unsqueeze(-1)
+        ).squeeze(-1)
+    else:
+        return torch.gather(input, dim, index)
+
+
+def bipartite_soft_matching_random2d(metric: torch.Tensor,
+                                     w: int, h: int, sx: int, sy: int, r: int,
+                                     no_rand: bool = False) -> Tuple[Callable, Callable]:
+    """
+    Partitions the tokens into src and dst and merges r tokens from src to dst.
+    Dst tokens are partitioned by choosing one randomy in each (sx, sy) region.
+    Args:
+     - metric [B, N, C]: metric to use for similarity
+     - w: image width in tokens
+     - h: image height in tokens
+     - sx: stride in the x dimension for dst, must divide w
+     - sy: stride in the y dimension for dst, must divide h
+     - r: number of tokens to remove (by merging)
+     - no_rand: if true, disable randomness (use top left corner only)
+    """
+    B, N, _ = metric.shape
+
+    if r <= 0 or w == 1 or h == 1:
+        return do_nothing, do_nothing
+
+    gather = mps_gather_workaround if metric.device.type == "mps" else torch.gather
+    
+    with torch.no_grad():
+        
+        hsy, wsx = h // sy, w // sx
+
+        # For each sy by sx kernel, randomly assign one token to be dst and the rest src
+        if no_rand:
+            rand_idx = torch.zeros(hsy, wsx, 1, device=metric.device, dtype=torch.int64)
+        else:
+            rand_idx = torch.randint(sy*sx, size=(hsy, wsx, 1), device=metric.device)
+        
+        # The image might not divide sx and sy, so we need to work on a view of the top left if the idx buffer instead
+        idx_buffer_view = torch.zeros(hsy, wsx, sy*sx, device=metric.device, dtype=torch.int64)
+        idx_buffer_view.scatter_(dim=2, index=rand_idx, src=-torch.ones_like(rand_idx, dtype=rand_idx.dtype))
+        idx_buffer_view = idx_buffer_view.view(hsy, wsx, sy, sx).transpose(1, 2).reshape(hsy * sy, wsx * sx)
+
+        # Image is not divisible by sx or sy so we need to move it into a new buffer
+        if (hsy * sy) < h or (wsx * sx) < w:
+            idx_buffer = torch.zeros(h, w, device=metric.device, dtype=torch.int64)
+            idx_buffer[:(hsy * sy), :(wsx * sx)] = idx_buffer_view
+        else:
+            idx_buffer = idx_buffer_view
+
+        # We set dst tokens to be -1 and src to be 0, so an argsort gives us dst|src indices
+        rand_idx = idx_buffer.reshape(1, -1, 1).argsort(dim=1)
+
+        # We're finished with these
+        del idx_buffer, idx_buffer_view
+
+        # rand_idx is currently dst|src, so split them
+        num_dst = hsy * wsx
+        a_idx = rand_idx[:, num_dst:, :] # src
+        b_idx = rand_idx[:, :num_dst, :] # dst
+
+        def split(x):
+            C = x.shape[-1]
+            src = gather(x, dim=1, index=a_idx.expand(B, N - num_dst, C))
+            dst = gather(x, dim=1, index=b_idx.expand(B, num_dst, C))
+            return src, dst
+
+        # Cosine similarity between A and B
+        metric = metric / metric.norm(dim=-1, keepdim=True)
+        a, b = split(metric)
+        scores = a @ b.transpose(-1, -2)
+
+        # Can't reduce more than the # tokens in src
+        r = min(a.shape[1], r)
+
+        # Find the most similar greedily
+        node_max, node_idx = scores.max(dim=-1)
+        edge_idx = node_max.argsort(dim=-1, descending=True)[..., None]
+
+        unm_idx = edge_idx[..., r:, :]  # Unmerged Tokens
+        src_idx = edge_idx[..., :r, :]  # Merged Tokens
+        dst_idx = gather(node_idx[..., None], dim=-2, index=src_idx)
+
+    def merge(x: torch.Tensor, mode="mean") -> torch.Tensor:
+        src, dst = split(x)
+        n, t1, c = src.shape
+        
+        unm = gather(src, dim=-2, index=unm_idx.expand(n, t1 - r, c))
+        src = gather(src, dim=-2, index=src_idx.expand(n, r, c))
+        dst = dst.scatter_reduce(-2, dst_idx.expand(n, r, c), src, reduce=mode)
+
+        return torch.cat([unm, dst], dim=1)
+
+    def unmerge(x: torch.Tensor) -> torch.Tensor:
+        unm_len = unm_idx.shape[1]
+        unm, dst = x[..., :unm_len, :], x[..., unm_len:, :]
+        _, _, c = unm.shape
+
+        src = gather(dst, dim=-2, index=dst_idx.expand(B, r, c))
+
+        # Combine back to the original shape
+        out = torch.zeros(B, N, c, device=x.device, dtype=x.dtype)
+        out.scatter_(dim=-2, index=b_idx.expand(B, num_dst, c), src=dst)
+        out.scatter_(dim=-2, index=gather(a_idx.expand(B, a_idx.shape[1], 1), dim=1, index=unm_idx).expand(B, unm_len, c), src=unm)
+        out.scatter_(dim=-2, index=gather(a_idx.expand(B, a_idx.shape[1], 1), dim=1, index=src_idx).expand(B, r, c), src=src)
+
+        return out
+
+    return merge, unmerge
+
+
+def get_functions(x, ratio, original_shape):
+    b, c, original_h, original_w = original_shape
+    original_tokens = original_h * original_w
+    downsample = int(math.ceil(math.sqrt(original_tokens // x.shape[1])))
+    stride_x = 2
+    stride_y = 2
+    max_downsample = 1
+
+    if downsample <= max_downsample:
+        w = int(math.ceil(original_w / downsample))
+        h = int(math.ceil(original_h / downsample))
+        r = int(x.shape[1] * ratio)
+        no_rand = False
+        m, u = bipartite_soft_matching_random2d(x, w, h, stride_x, stride_y, r, no_rand)
+        return m, u
+
+    nothing = lambda y: y
+    return nothing, nothing
+
+
+
+class TomePatchModel:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model": ("MODEL",),
+                              "ratio": ("FLOAT", {"default": 0.3, "min": 0.0, "max": 1.0, "step": 0.01}),
+                              }}
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "patch"
+
+    CATEGORY = "_for_testing"
+
+    def patch(self, model, ratio):
+        self.u = None
+        def tomesd_m(q, k, v, extra_options):
+            #NOTE: In the reference code get_functions takes x (input of the transformer block) as the argument instead of q
+            #however from my basic testing it seems that using q instead gives better results
+            m, self.u = get_functions(q, ratio, extra_options["original_shape"])
+            return m(q), k, v
+        def tomesd_u(n, extra_options):
+            return self.u(n)
+
+        m = model.clone()
+        m.set_model_attn1_patch(tomesd_m)
+        m.set_model_attn1_output_patch(tomesd_u)
+        return (m, )
+
+
+NODE_CLASS_MAPPINGS = {
+    "TomePatchModel": TomePatchModel,
+}

ldm_patched/contrib/external_upscale_model.py

@@ -0,0 +1,68 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+import os
+from ldm_patched.pfn import model_loading
+from ldm_patched.modules import model_management
+import torch
+import ldm_patched.modules.utils
+import ldm_patched.utils.path_utils
+
+class UpscaleModelLoader:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model_name": (ldm_patched.utils.path_utils.get_filename_list("upscale_models"), ),
+                             }}
+    RETURN_TYPES = ("UPSCALE_MODEL",)
+    FUNCTION = "load_model"
+
+    CATEGORY = "loaders"
+
+    def load_model(self, model_name):
+        model_path = ldm_patched.utils.path_utils.get_full_path("upscale_models", model_name)
+        sd = ldm_patched.modules.utils.load_torch_file(model_path, safe_load=True)
+        if "module.layers.0.residual_group.blocks.0.norm1.weight" in sd:
+            sd = ldm_patched.modules.utils.state_dict_prefix_replace(sd, {"module.":""})
+        out = model_loading.load_state_dict(sd).eval()
+        return (out, )
+
+
+class ImageUpscaleWithModel:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "upscale_model": ("UPSCALE_MODEL",),
+                              "image": ("IMAGE",),
+                              }}
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "upscale"
+
+    CATEGORY = "image/upscaling"
+
+    def upscale(self, upscale_model, image):
+        device = model_management.get_torch_device()
+        upscale_model.to(device)
+        in_img = image.movedim(-1,-3).to(device)
+        free_memory = model_management.get_free_memory(device)
+
+        tile = 512
+        overlap = 32
+
+        oom = True
+        while oom:
+            try:
+                steps = in_img.shape[0] * ldm_patched.modules.utils.get_tiled_scale_steps(in_img.shape[3], in_img.shape[2], tile_x=tile, tile_y=tile, overlap=overlap)
+                pbar = ldm_patched.modules.utils.ProgressBar(steps)
+                s = ldm_patched.modules.utils.tiled_scale(in_img, lambda a: upscale_model(a), tile_x=tile, tile_y=tile, overlap=overlap, upscale_amount=upscale_model.scale, pbar=pbar)
+                oom = False
+            except model_management.OOM_EXCEPTION as e:
+                tile //= 2
+                if tile < 128:
+                    raise e
+
+        upscale_model.cpu()
+        s = torch.clamp(s.movedim(-3,-1), min=0, max=1.0)
+        return (s,)
+
+NODE_CLASS_MAPPINGS = {
+    "UpscaleModelLoader": UpscaleModelLoader,
+    "ImageUpscaleWithModel": ImageUpscaleWithModel
+}

ldm_patched/contrib/external_video_model.py

@@ -0,0 +1,91 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+import ldm_patched.contrib.external
+import torch
+import ldm_patched.modules.utils
+import ldm_patched.modules.sd
+import ldm_patched.utils.path_utils
+
+
+class ImageOnlyCheckpointLoader:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "ckpt_name": (ldm_patched.utils.path_utils.get_filename_list("checkpoints"), ),
+                             }}
+    RETURN_TYPES = ("MODEL", "CLIP_VISION", "VAE")
+    FUNCTION = "load_checkpoint"
+
+    CATEGORY = "loaders/video_models"
+
+    def load_checkpoint(self, ckpt_name, output_vae=True, output_clip=True):
+        ckpt_path = ldm_patched.utils.path_utils.get_full_path("checkpoints", ckpt_name)
+        out = ldm_patched.modules.sd.load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=False, output_clipvision=True, embedding_directory=ldm_patched.utils.path_utils.get_folder_paths("embeddings"))
+        return (out[0], out[3], out[2])
+
+
+class SVD_img2vid_Conditioning:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "clip_vision": ("CLIP_VISION",),
+                              "init_image": ("IMAGE",),
+                              "vae": ("VAE",),
+                              "width": ("INT", {"default": 1024, "min": 16, "max": ldm_patched.contrib.external.MAX_RESOLUTION, "step": 8}),
+                              "height": ("INT", {"default": 576, "min": 16, "max": ldm_patched.contrib.external.MAX_RESOLUTION, "step": 8}),
+                              "video_frames": ("INT", {"default": 14, "min": 1, "max": 4096}),
+                              "motion_bucket_id": ("INT", {"default": 127, "min": 1, "max": 1023}),
+                              "fps": ("INT", {"default": 6, "min": 1, "max": 1024}),
+                              "augmentation_level": ("FLOAT", {"default": 0.0, "min": 0.0, "max": 10.0, "step": 0.01})
+                             }}
+    RETURN_TYPES = ("CONDITIONING", "CONDITIONING", "LATENT")
+    RETURN_NAMES = ("positive", "negative", "latent")
+
+    FUNCTION = "encode"
+
+    CATEGORY = "conditioning/video_models"
+
+    def encode(self, clip_vision, init_image, vae, width, height, video_frames, motion_bucket_id, fps, augmentation_level):
+        output = clip_vision.encode_image(init_image)
+        pooled = output.image_embeds.unsqueeze(0)
+        pixels = ldm_patched.modules.utils.common_upscale(init_image.movedim(-1,1), width, height, "bilinear", "center").movedim(1,-1)
+        encode_pixels = pixels[:,:,:,:3]
+        if augmentation_level > 0:
+            encode_pixels += torch.randn_like(pixels) * augmentation_level
+        t = vae.encode(encode_pixels)
+        positive = [[pooled, {"motion_bucket_id": motion_bucket_id, "fps": fps, "augmentation_level": augmentation_level, "concat_latent_image": t}]]
+        negative = [[torch.zeros_like(pooled), {"motion_bucket_id": motion_bucket_id, "fps": fps, "augmentation_level": augmentation_level, "concat_latent_image": torch.zeros_like(t)}]]
+        latent = torch.zeros([video_frames, 4, height // 8, width // 8])
+        return (positive, negative, {"samples":latent})
+
+class VideoLinearCFGGuidance:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model": ("MODEL",),
+                              "min_cfg": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0, "step":0.5, "round": 0.01}),
+                              }}
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "patch"
+
+    CATEGORY = "sampling/video_models"
+
+    def patch(self, model, min_cfg):
+        def linear_cfg(args):
+            cond = args["cond"]
+            uncond = args["uncond"]
+            cond_scale = args["cond_scale"]
+
+            scale = torch.linspace(min_cfg, cond_scale, cond.shape[0], device=cond.device).reshape((cond.shape[0], 1, 1, 1))
+            return uncond + scale * (cond - uncond)
+
+        m = model.clone()
+        m.set_model_sampler_cfg_function(linear_cfg)
+        return (m, )
+
+NODE_CLASS_MAPPINGS = {
+    "ImageOnlyCheckpointLoader": ImageOnlyCheckpointLoader,
+    "SVD_img2vid_Conditioning": SVD_img2vid_Conditioning,
+    "VideoLinearCFGGuidance": VideoLinearCFGGuidance,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "ImageOnlyCheckpointLoader": "Image Only Checkpoint Loader (img2vid model)",
+}

ldm_patched/controlnet/cldm.py

@@ -0,0 +1,312 @@
+#taken from: https://github.com/lllyasviel/ControlNet
+#and modified
+
+import torch
+import torch as th
+import torch.nn as nn
+
+from ldm_patched.ldm.modules.diffusionmodules.util import (
+    zero_module,
+    timestep_embedding,
+)
+
+from ldm_patched.ldm.modules.attention import SpatialTransformer
+from ldm_patched.ldm.modules.diffusionmodules.openaimodel import UNetModel, TimestepEmbedSequential, ResBlock, Downsample
+from ldm_patched.ldm.util import exists
+import ldm_patched.modules.ops
+
+class ControlledUnetModel(UNetModel):
+    #implemented in the ldm unet
+    pass
+
+class ControlNet(nn.Module):
+    def __init__(
+        self,
+        image_size,
+        in_channels,
+        model_channels,
+        hint_channels,
+        num_res_blocks,
+        dropout=0,
+        channel_mult=(1, 2, 4, 8),
+        conv_resample=True,
+        dims=2,
+        num_classes=None,
+        use_checkpoint=False,
+        dtype=torch.float32,
+        num_heads=-1,
+        num_head_channels=-1,
+        num_heads_upsample=-1,
+        use_scale_shift_norm=False,
+        resblock_updown=False,
+        use_new_attention_order=False,
+        use_spatial_transformer=False,    # custom transformer support
+        transformer_depth=1,              # custom transformer support
+        context_dim=None,                 # custom transformer support
+        n_embed=None,                     # custom support for prediction of discrete ids into codebook of first stage vq model
+        legacy=True,
+        disable_self_attentions=None,
+        num_attention_blocks=None,
+        disable_middle_self_attn=False,
+        use_linear_in_transformer=False,
+        adm_in_channels=None,
+        transformer_depth_middle=None,
+        transformer_depth_output=None,
+        device=None,
+        operations=ldm_patched.modules.ops.disable_weight_init,
+        **kwargs,
+    ):
+        super().__init__()
+        assert use_spatial_transformer == True, "use_spatial_transformer has to be true"
+        if use_spatial_transformer:
+            assert context_dim is not None, 'Fool!! You forgot to include the dimension of your cross-attention conditioning...'
+
+        if context_dim is not None:
+            assert use_spatial_transformer, 'Fool!! You forgot to use the spatial transformer for your cross-attention conditioning...'
+            # from omegaconf.listconfig import ListConfig
+            # if type(context_dim) == ListConfig:
+            #     context_dim = list(context_dim)
+
+        if num_heads_upsample == -1:
+            num_heads_upsample = num_heads
+
+        if num_heads == -1:
+            assert num_head_channels != -1, 'Either num_heads or num_head_channels has to be set'
+
+        if num_head_channels == -1:
+            assert num_heads != -1, 'Either num_heads or num_head_channels has to be set'
+
+        self.dims = dims
+        self.image_size = image_size
+        self.in_channels = in_channels
+        self.model_channels = model_channels
+
+        if isinstance(num_res_blocks, int):
+            self.num_res_blocks = len(channel_mult) * [num_res_blocks]
+        else:
+            if len(num_res_blocks) != len(channel_mult):
+                raise ValueError("provide num_res_blocks either as an int (globally constant) or "
+                                 "as a list/tuple (per-level) with the same length as channel_mult")
+            self.num_res_blocks = num_res_blocks
+
+        if disable_self_attentions is not None:
+            # should be a list of booleans, indicating whether to disable self-attention in TransformerBlocks or not
+            assert len(disable_self_attentions) == len(channel_mult)
+        if num_attention_blocks is not None:
+            assert len(num_attention_blocks) == len(self.num_res_blocks)
+            assert all(map(lambda i: self.num_res_blocks[i] >= num_attention_blocks[i], range(len(num_attention_blocks))))
+
+        transformer_depth = transformer_depth[:]
+
+        self.dropout = dropout
+        self.channel_mult = channel_mult
+        self.conv_resample = conv_resample
+        self.num_classes = num_classes
+        self.use_checkpoint = use_checkpoint
+        self.dtype = dtype
+        self.num_heads = num_heads
+        self.num_head_channels = num_head_channels
+        self.num_heads_upsample = num_heads_upsample
+        self.predict_codebook_ids = n_embed is not None
+
+        time_embed_dim = model_channels * 4
+        self.time_embed = nn.Sequential(
+            operations.Linear(model_channels, time_embed_dim, dtype=self.dtype, device=device),
+            nn.SiLU(),
+            operations.Linear(time_embed_dim, time_embed_dim, dtype=self.dtype, device=device),
+        )
+
+        if self.num_classes is not None:
+            if isinstance(self.num_classes, int):
+                self.label_emb = nn.Embedding(num_classes, time_embed_dim)
+            elif self.num_classes == "continuous":
+                print("setting up linear c_adm embedding layer")
+                self.label_emb = nn.Linear(1, time_embed_dim)
+            elif self.num_classes == "sequential":
+                assert adm_in_channels is not None
+                self.label_emb = nn.Sequential(
+                    nn.Sequential(
+                        operations.Linear(adm_in_channels, time_embed_dim, dtype=self.dtype, device=device),
+                        nn.SiLU(),
+                        operations.Linear(time_embed_dim, time_embed_dim, dtype=self.dtype, device=device),
+                    )
+                )
+            else:
+                raise ValueError()
+
+        self.input_blocks = nn.ModuleList(
+            [
+                TimestepEmbedSequential(
+                    operations.conv_nd(dims, in_channels, model_channels, 3, padding=1, dtype=self.dtype, device=device)
+                )
+            ]
+        )
+        self.zero_convs = nn.ModuleList([self.make_zero_conv(model_channels, operations=operations, dtype=self.dtype, device=device)])
+
+        self.input_hint_block = TimestepEmbedSequential(
+                    operations.conv_nd(dims, hint_channels, 16, 3, padding=1, dtype=self.dtype, device=device),
+                    nn.SiLU(),
+                    operations.conv_nd(dims, 16, 16, 3, padding=1, dtype=self.dtype, device=device),
+                    nn.SiLU(),
+                    operations.conv_nd(dims, 16, 32, 3, padding=1, stride=2, dtype=self.dtype, device=device),
+                    nn.SiLU(),
+                    operations.conv_nd(dims, 32, 32, 3, padding=1, dtype=self.dtype, device=device),
+                    nn.SiLU(),
+                    operations.conv_nd(dims, 32, 96, 3, padding=1, stride=2, dtype=self.dtype, device=device),
+                    nn.SiLU(),
+                    operations.conv_nd(dims, 96, 96, 3, padding=1, dtype=self.dtype, device=device),
+                    nn.SiLU(),
+                    operations.conv_nd(dims, 96, 256, 3, padding=1, stride=2, dtype=self.dtype, device=device),
+                    nn.SiLU(),
+                    operations.conv_nd(dims, 256, model_channels, 3, padding=1, dtype=self.dtype, device=device)
+        )
+
+        self._feature_size = model_channels
+        input_block_chans = [model_channels]
+        ch = model_channels
+        ds = 1
+        for level, mult in enumerate(channel_mult):
+            for nr in range(self.num_res_blocks[level]):
+                layers = [
+                    ResBlock(
+                        ch,
+                        time_embed_dim,
+                        dropout,
+                        out_channels=mult * model_channels,
+                        dims=dims,
+                        use_checkpoint=use_checkpoint,
+                        use_scale_shift_norm=use_scale_shift_norm,
+                        dtype=self.dtype,
+                        device=device,
+                        operations=operations,
+                    )
+                ]
+                ch = mult * model_channels
+                num_transformers = transformer_depth.pop(0)
+                if num_transformers > 0:
+                    if num_head_channels == -1:
+                        dim_head = ch // num_heads
+                    else:
+                        num_heads = ch // num_head_channels
+                        dim_head = num_head_channels
+                    if legacy:
+                        #num_heads = 1
+                        dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
+                    if exists(disable_self_attentions):
+                        disabled_sa = disable_self_attentions[level]
+                    else:
+                        disabled_sa = False
+
+                    if not exists(num_attention_blocks) or nr < num_attention_blocks[level]:
+                        layers.append(
+                            SpatialTransformer(
+                                ch, num_heads, dim_head, depth=num_transformers, context_dim=context_dim,
+                                disable_self_attn=disabled_sa, use_linear=use_linear_in_transformer,
+                                use_checkpoint=use_checkpoint, dtype=self.dtype, device=device, operations=operations
+                            )
+                        )
+                self.input_blocks.append(TimestepEmbedSequential(*layers))
+                self.zero_convs.append(self.make_zero_conv(ch, operations=operations, dtype=self.dtype, device=device))
+                self._feature_size += ch
+                input_block_chans.append(ch)
+            if level != len(channel_mult) - 1:
+                out_ch = ch
+                self.input_blocks.append(
+                    TimestepEmbedSequential(
+                        ResBlock(
+                            ch,
+                            time_embed_dim,
+                            dropout,
+                            out_channels=out_ch,
+                            dims=dims,
+                            use_checkpoint=use_checkpoint,
+                            use_scale_shift_norm=use_scale_shift_norm,
+                            down=True,
+                            dtype=self.dtype,
+                            device=device,
+                            operations=operations
+                        )
+                        if resblock_updown
+                        else Downsample(
+                            ch, conv_resample, dims=dims, out_channels=out_ch, dtype=self.dtype, device=device, operations=operations
+                        )
+                    )
+                )
+                ch = out_ch
+                input_block_chans.append(ch)
+                self.zero_convs.append(self.make_zero_conv(ch, operations=operations, dtype=self.dtype, device=device))
+                ds *= 2
+                self._feature_size += ch
+
+        if num_head_channels == -1:
+            dim_head = ch // num_heads
+        else:
+            num_heads = ch // num_head_channels
+            dim_head = num_head_channels
+        if legacy:
+            #num_heads = 1
+            dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
+        mid_block = [
+            ResBlock(
+                ch,
+                time_embed_dim,
+                dropout,
+                dims=dims,
+                use_checkpoint=use_checkpoint,
+                use_scale_shift_norm=use_scale_shift_norm,
+                dtype=self.dtype,
+                device=device,
+                operations=operations
+            )]
+        if transformer_depth_middle >= 0:
+            mid_block += [SpatialTransformer(  # always uses a self-attn
+                            ch, num_heads, dim_head, depth=transformer_depth_middle, context_dim=context_dim,
+                            disable_self_attn=disable_middle_self_attn, use_linear=use_linear_in_transformer,
+                            use_checkpoint=use_checkpoint, dtype=self.dtype, device=device, operations=operations
+                        ),
+            ResBlock(
+                ch,
+                time_embed_dim,
+                dropout,
+                dims=dims,
+                use_checkpoint=use_checkpoint,
+                use_scale_shift_norm=use_scale_shift_norm,
+                dtype=self.dtype,
+                device=device,
+                operations=operations
+            )]
+        self.middle_block = TimestepEmbedSequential(*mid_block)
+        self.middle_block_out = self.make_zero_conv(ch, operations=operations, dtype=self.dtype, device=device)
+        self._feature_size += ch
+
+    def make_zero_conv(self, channels, operations=None, dtype=None, device=None):
+        return TimestepEmbedSequential(operations.conv_nd(self.dims, channels, channels, 1, padding=0, dtype=dtype, device=device))
+
+    def forward(self, x, hint, timesteps, context, y=None, **kwargs):
+        t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False).to(x.dtype)
+        emb = self.time_embed(t_emb)
+
+        guided_hint = self.input_hint_block(hint, emb, context)
+
+        outs = []
+
+        hs = []
+        if self.num_classes is not None:
+            assert y.shape[0] == x.shape[0]
+            emb = emb + self.label_emb(y)
+
+        h = x
+        for module, zero_conv in zip(self.input_blocks, self.zero_convs):
+            if guided_hint is not None:
+                h = module(h, emb, context)
+                h += guided_hint
+                guided_hint = None
+            else:
+                h = module(h, emb, context)
+            outs.append(zero_conv(h, emb, context))
+
+        h = self.middle_block(h, emb, context)
+        outs.append(self.middle_block_out(h, emb, context))
+
+        return outs
+

ldm_patched/k_diffusion/sampling.py

@@ -0,0 +1,810 @@
+import math
+
+from scipy import integrate
+import torch
+from torch import nn
+import torchsde
+from tqdm.auto import trange, tqdm
+
+from . import utils
+
+
+def append_zero(x):
+    return torch.cat([x, x.new_zeros([1])])
+
+
+def get_sigmas_karras(n, sigma_min, sigma_max, rho=7., device='cpu'):
+    """Constructs the noise schedule of Karras et al. (2022)."""
+    ramp = torch.linspace(0, 1, n, device=device)
+    min_inv_rho = sigma_min ** (1 / rho)
+    max_inv_rho = sigma_max ** (1 / rho)
+    sigmas = (max_inv_rho + ramp * (min_inv_rho - max_inv_rho)) ** rho
+    return append_zero(sigmas).to(device)
+
+
+def get_sigmas_exponential(n, sigma_min, sigma_max, device='cpu'):
+    """Constructs an exponential noise schedule."""
+    sigmas = torch.linspace(math.log(sigma_max), math.log(sigma_min), n, device=device).exp()
+    return append_zero(sigmas)
+
+
+def get_sigmas_polyexponential(n, sigma_min, sigma_max, rho=1., device='cpu'):
+    """Constructs an polynomial in log sigma noise schedule."""
+    ramp = torch.linspace(1, 0, n, device=device) ** rho
+    sigmas = torch.exp(ramp * (math.log(sigma_max) - math.log(sigma_min)) + math.log(sigma_min))
+    return append_zero(sigmas)
+
+
+def get_sigmas_vp(n, beta_d=19.9, beta_min=0.1, eps_s=1e-3, device='cpu'):
+    """Constructs a continuous VP noise schedule."""
+    t = torch.linspace(1, eps_s, n, device=device)
+    sigmas = torch.sqrt(torch.exp(beta_d * t ** 2 / 2 + beta_min * t) - 1)
+    return append_zero(sigmas)
+
+
+def to_d(x, sigma, denoised):
+    """Converts a denoiser output to a Karras ODE derivative."""
+    return (x - denoised) / utils.append_dims(sigma, x.ndim)
+
+
+def get_ancestral_step(sigma_from, sigma_to, eta=1.):
+    """Calculates the noise level (sigma_down) to step down to and the amount
+    of noise to add (sigma_up) when doing an ancestral sampling step."""
+    if not eta:
+        return sigma_to, 0.
+    sigma_up = min(sigma_to, eta * (sigma_to ** 2 * (sigma_from ** 2 - sigma_to ** 2) / sigma_from ** 2) ** 0.5)
+    sigma_down = (sigma_to ** 2 - sigma_up ** 2) ** 0.5
+    return sigma_down, sigma_up
+
+
+def default_noise_sampler(x):
+    return lambda sigma, sigma_next: torch.randn_like(x)
+
+
+class BatchedBrownianTree:
+    """A wrapper around torchsde.BrownianTree that enables batches of entropy."""
+
+    def __init__(self, x, t0, t1, seed=None, **kwargs):
+        self.cpu_tree = True
+        if "cpu" in kwargs:
+            self.cpu_tree = kwargs.pop("cpu")
+        t0, t1, self.sign = self.sort(t0, t1)
+        w0 = kwargs.get('w0', torch.zeros_like(x))
+        if seed is None:
+            seed = torch.randint(0, 2 ** 63 - 1, []).item()
+        self.batched = True
+        try:
+            assert len(seed) == x.shape[0]
+            w0 = w0[0]
+        except TypeError:
+            seed = [seed]
+            self.batched = False
+        if self.cpu_tree:
+            self.trees = [torchsde.BrownianTree(t0.cpu(), w0.cpu(), t1.cpu(), entropy=s, **kwargs) for s in seed]
+        else:
+            self.trees = [torchsde.BrownianTree(t0, w0, t1, entropy=s, **kwargs) for s in seed]
+
+    @staticmethod
+    def sort(a, b):
+        return (a, b, 1) if a < b else (b, a, -1)
+
+    def __call__(self, t0, t1):
+        t0, t1, sign = self.sort(t0, t1)
+        if self.cpu_tree:
+            w = torch.stack([tree(t0.cpu().float(), t1.cpu().float()).to(t0.dtype).to(t0.device) for tree in self.trees]) * (self.sign * sign)
+        else:
+            w = torch.stack([tree(t0, t1) for tree in self.trees]) * (self.sign * sign)
+
+        return w if self.batched else w[0]
+
+
+class BrownianTreeNoiseSampler:
+    """A noise sampler backed by a torchsde.BrownianTree.
+
+    Args:
+        x (Tensor): The tensor whose shape, device and dtype to use to generate
+            random samples.
+        sigma_min (float): The low end of the valid interval.
+        sigma_max (float): The high end of the valid interval.
+        seed (int or List[int]): The random seed. If a list of seeds is
+            supplied instead of a single integer, then the noise sampler will
+            use one BrownianTree per batch item, each with its own seed.
+        transform (callable): A function that maps sigma to the sampler's
+            internal timestep.
+    """
+
+    def __init__(self, x, sigma_min, sigma_max, seed=None, transform=lambda x: x, cpu=False):
+        self.transform = transform
+        t0, t1 = self.transform(torch.as_tensor(sigma_min)), self.transform(torch.as_tensor(sigma_max))
+        self.tree = BatchedBrownianTree(x, t0, t1, seed, cpu=cpu)
+
+    def __call__(self, sigma, sigma_next):
+        t0, t1 = self.transform(torch.as_tensor(sigma)), self.transform(torch.as_tensor(sigma_next))
+        return self.tree(t0, t1) / (t1 - t0).abs().sqrt()
+
+
+@torch.no_grad()
+def sample_euler(model, x, sigmas, extra_args=None, callback=None, disable=None, s_churn=0., s_tmin=0., s_tmax=float('inf'), s_noise=1.):
+    """Implements Algorithm 2 (Euler steps) from Karras et al. (2022)."""
+    extra_args = {} if extra_args is None else extra_args
+    s_in = x.new_ones([x.shape[0]])
+    for i in trange(len(sigmas) - 1, disable=disable):
+        gamma = min(s_churn / (len(sigmas) - 1), 2 ** 0.5 - 1) if s_tmin <= sigmas[i] <= s_tmax else 0.
+        sigma_hat = sigmas[i] * (gamma + 1)
+        if gamma > 0:
+            eps = torch.randn_like(x) * s_noise
+            x = x + eps * (sigma_hat ** 2 - sigmas[i] ** 2) ** 0.5
+        denoised = model(x, sigma_hat * s_in, **extra_args)
+        d = to_d(x, sigma_hat, denoised)
+        if callback is not None:
+            callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigma_hat, 'denoised': denoised})
+        dt = sigmas[i + 1] - sigma_hat
+        # Euler method
+        x = x + d * dt
+    return x
+
+
+@torch.no_grad()
+def sample_euler_ancestral(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None):
+    """Ancestral sampling with Euler method steps."""
+    extra_args = {} if extra_args is None else extra_args
+    noise_sampler = default_noise_sampler(x) if noise_sampler is None else noise_sampler
+    s_in = x.new_ones([x.shape[0]])
+    for i in trange(len(sigmas) - 1, disable=disable):
+        denoised = model(x, sigmas[i] * s_in, **extra_args)
+        sigma_down, sigma_up = get_ancestral_step(sigmas[i], sigmas[i + 1], eta=eta)
+        if callback is not None:
+            callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised})
+        d = to_d(x, sigmas[i], denoised)
+        # Euler method
+        dt = sigma_down - sigmas[i]
+        x = x + d * dt
+        if sigmas[i + 1] > 0:
+            x = x + noise_sampler(sigmas[i], sigmas[i + 1]) * s_noise * sigma_up
+    return x
+
+
+@torch.no_grad()
+def sample_heun(model, x, sigmas, extra_args=None, callback=None, disable=None, s_churn=0., s_tmin=0., s_tmax=float('inf'), s_noise=1.):
+    """Implements Algorithm 2 (Heun steps) from Karras et al. (2022)."""
+    extra_args = {} if extra_args is None else extra_args
+    s_in = x.new_ones([x.shape[0]])
+    for i in trange(len(sigmas) - 1, disable=disable):
+        gamma = min(s_churn / (len(sigmas) - 1), 2 ** 0.5 - 1) if s_tmin <= sigmas[i] <= s_tmax else 0.
+        sigma_hat = sigmas[i] * (gamma + 1)
+        if gamma > 0:
+            eps = torch.randn_like(x) * s_noise
+            x = x + eps * (sigma_hat ** 2 - sigmas[i] ** 2) ** 0.5
+        denoised = model(x, sigma_hat * s_in, **extra_args)
+        d = to_d(x, sigma_hat, denoised)
+        if callback is not None:
+            callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigma_hat, 'denoised': denoised})
+        dt = sigmas[i + 1] - sigma_hat
+        if sigmas[i + 1] == 0:
+            # Euler method
+            x = x + d * dt
+        else:
+            # Heun's method
+            x_2 = x + d * dt
+            denoised_2 = model(x_2, sigmas[i + 1] * s_in, **extra_args)
+            d_2 = to_d(x_2, sigmas[i + 1], denoised_2)
+            d_prime = (d + d_2) / 2
+            x = x + d_prime * dt
+    return x
+
+
+@torch.no_grad()
+def sample_dpm_2(model, x, sigmas, extra_args=None, callback=None, disable=None, s_churn=0., s_tmin=0., s_tmax=float('inf'), s_noise=1.):
+    """A sampler inspired by DPM-Solver-2 and Algorithm 2 from Karras et al. (2022)."""
+    extra_args = {} if extra_args is None else extra_args
+    s_in = x.new_ones([x.shape[0]])
+    for i in trange(len(sigmas) - 1, disable=disable):
+        gamma = min(s_churn / (len(sigmas) - 1), 2 ** 0.5 - 1) if s_tmin <= sigmas[i] <= s_tmax else 0.
+        sigma_hat = sigmas[i] * (gamma + 1)
+        if gamma > 0:
+            eps = torch.randn_like(x) * s_noise
+            x = x + eps * (sigma_hat ** 2 - sigmas[i] ** 2) ** 0.5
+        denoised = model(x, sigma_hat * s_in, **extra_args)
+        d = to_d(x, sigma_hat, denoised)
+        if callback is not None:
+            callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigma_hat, 'denoised': denoised})
+        if sigmas[i + 1] == 0:
+            # Euler method
+            dt = sigmas[i + 1] - sigma_hat
+            x = x + d * dt
+        else:
+            # DPM-Solver-2
+            sigma_mid = sigma_hat.log().lerp(sigmas[i + 1].log(), 0.5).exp()
+            dt_1 = sigma_mid - sigma_hat
+            dt_2 = sigmas[i + 1] - sigma_hat
+            x_2 = x + d * dt_1
+            denoised_2 = model(x_2, sigma_mid * s_in, **extra_args)
+            d_2 = to_d(x_2, sigma_mid, denoised_2)
+            x = x + d_2 * dt_2
+    return x
+
+
+@torch.no_grad()
+def sample_dpm_2_ancestral(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None):
+    """Ancestral sampling with DPM-Solver second-order steps."""
+    extra_args = {} if extra_args is None else extra_args
+    noise_sampler = default_noise_sampler(x) if noise_sampler is None else noise_sampler
+    s_in = x.new_ones([x.shape[0]])
+    for i in trange(len(sigmas) - 1, disable=disable):
+        denoised = model(x, sigmas[i] * s_in, **extra_args)
+        sigma_down, sigma_up = get_ancestral_step(sigmas[i], sigmas[i + 1], eta=eta)
+        if callback is not None:
+            callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised})
+        d = to_d(x, sigmas[i], denoised)
+        if sigma_down == 0:
+            # Euler method
+            dt = sigma_down - sigmas[i]
+            x = x + d * dt
+        else:
+            # DPM-Solver-2
+            sigma_mid = sigmas[i].log().lerp(sigma_down.log(), 0.5).exp()
+            dt_1 = sigma_mid - sigmas[i]
+            dt_2 = sigma_down - sigmas[i]
+            x_2 = x + d * dt_1
+            denoised_2 = model(x_2, sigma_mid * s_in, **extra_args)
+            d_2 = to_d(x_2, sigma_mid, denoised_2)
+            x = x + d_2 * dt_2
+            x = x + noise_sampler(sigmas[i], sigmas[i + 1]) * s_noise * sigma_up
+    return x
+
+
+def linear_multistep_coeff(order, t, i, j):
+    if order - 1 > i:
+        raise ValueError(f'Order {order} too high for step {i}')
+    def fn(tau):
+        prod = 1.
+        for k in range(order):
+            if j == k:
+                continue
+            prod *= (tau - t[i - k]) / (t[i - j] - t[i - k])
+        return prod
+    return integrate.quad(fn, t[i], t[i + 1], epsrel=1e-4)[0]
+
+
+@torch.no_grad()
+def sample_lms(model, x, sigmas, extra_args=None, callback=None, disable=None, order=4):
+    extra_args = {} if extra_args is None else extra_args
+    s_in = x.new_ones([x.shape[0]])
+    sigmas_cpu = sigmas.detach().cpu().numpy()
+    ds = []
+    for i in trange(len(sigmas) - 1, disable=disable):
+        denoised = model(x, sigmas[i] * s_in, **extra_args)
+        d = to_d(x, sigmas[i], denoised)
+        ds.append(d)
+        if len(ds) > order:
+            ds.pop(0)
+        if callback is not None:
+            callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised})
+        cur_order = min(i + 1, order)
+        coeffs = [linear_multistep_coeff(cur_order, sigmas_cpu, i, j) for j in range(cur_order)]
+        x = x + sum(coeff * d for coeff, d in zip(coeffs, reversed(ds)))
+    return x
+
+
+class PIDStepSizeController:
+    """A PID controller for ODE adaptive step size control."""
+    def __init__(self, h, pcoeff, icoeff, dcoeff, order=1, accept_safety=0.81, eps=1e-8):
+        self.h = h
+        self.b1 = (pcoeff + icoeff + dcoeff) / order
+        self.b2 = -(pcoeff + 2 * dcoeff) / order
+        self.b3 = dcoeff / order
+        self.accept_safety = accept_safety
+        self.eps = eps
+        self.errs = []
+
+    def limiter(self, x):
+        return 1 + math.atan(x - 1)
+
+    def propose_step(self, error):
+        inv_error = 1 / (float(error) + self.eps)
+        if not self.errs:
+            self.errs = [inv_error, inv_error, inv_error]
+        self.errs[0] = inv_error
+        factor = self.errs[0] ** self.b1 * self.errs[1] ** self.b2 * self.errs[2] ** self.b3
+        factor = self.limiter(factor)
+        accept = factor >= self.accept_safety
+        if accept:
+            self.errs[2] = self.errs[1]
+            self.errs[1] = self.errs[0]
+        self.h *= factor
+        return accept
+
+
+class DPMSolver(nn.Module):
+    """DPM-Solver. See https://arxiv.org/abs/2206.00927."""
+
+    def __init__(self, model, extra_args=None, eps_callback=None, info_callback=None):
+        super().__init__()
+        self.model = model
+        self.extra_args = {} if extra_args is None else extra_args
+        self.eps_callback = eps_callback
+        self.info_callback = info_callback
+
+    def t(self, sigma):
+        return -sigma.log()
+
+    def sigma(self, t):
+        return t.neg().exp()
+
+    def eps(self, eps_cache, key, x, t, *args, **kwargs):
+        if key in eps_cache:
+            return eps_cache[key], eps_cache
+        sigma = self.sigma(t) * x.new_ones([x.shape[0]])
+        eps = (x - self.model(x, sigma, *args, **self.extra_args, **kwargs)) / self.sigma(t)
+        if self.eps_callback is not None:
+            self.eps_callback()
+        return eps, {key: eps, **eps_cache}
+
+    def dpm_solver_1_step(self, x, t, t_next, eps_cache=None):
+        eps_cache = {} if eps_cache is None else eps_cache
+        h = t_next - t
+        eps, eps_cache = self.eps(eps_cache, 'eps', x, t)
+        x_1 = x - self.sigma(t_next) * h.expm1() * eps
+        return x_1, eps_cache
+
+    def dpm_solver_2_step(self, x, t, t_next, r1=1 / 2, eps_cache=None):
+        eps_cache = {} if eps_cache is None else eps_cache
+        h = t_next - t
+        eps, eps_cache = self.eps(eps_cache, 'eps', x, t)
+        s1 = t + r1 * h
+        u1 = x - self.sigma(s1) * (r1 * h).expm1() * eps
+        eps_r1, eps_cache = self.eps(eps_cache, 'eps_r1', u1, s1)
+        x_2 = x - self.sigma(t_next) * h.expm1() * eps - self.sigma(t_next) / (2 * r1) * h.expm1() * (eps_r1 - eps)
+        return x_2, eps_cache
+
+    def dpm_solver_3_step(self, x, t, t_next, r1=1 / 3, r2=2 / 3, eps_cache=None):
+        eps_cache = {} if eps_cache is None else eps_cache
+        h = t_next - t
+        eps, eps_cache = self.eps(eps_cache, 'eps', x, t)
+        s1 = t + r1 * h
+        s2 = t + r2 * h
+        u1 = x - self.sigma(s1) * (r1 * h).expm1() * eps
+        eps_r1, eps_cache = self.eps(eps_cache, 'eps_r1', u1, s1)
+        u2 = x - self.sigma(s2) * (r2 * h).expm1() * eps - self.sigma(s2) * (r2 / r1) * ((r2 * h).expm1() / (r2 * h) - 1) * (eps_r1 - eps)
+        eps_r2, eps_cache = self.eps(eps_cache, 'eps_r2', u2, s2)
+        x_3 = x - self.sigma(t_next) * h.expm1() * eps - self.sigma(t_next) / r2 * (h.expm1() / h - 1) * (eps_r2 - eps)
+        return x_3, eps_cache
+
+    def dpm_solver_fast(self, x, t_start, t_end, nfe, eta=0., s_noise=1., noise_sampler=None):
+        noise_sampler = default_noise_sampler(x) if noise_sampler is None else noise_sampler
+        if not t_end > t_start and eta:
+            raise ValueError('eta must be 0 for reverse sampling')
+
+        m = math.floor(nfe / 3) + 1
+        ts = torch.linspace(t_start, t_end, m + 1, device=x.device)
+
+        if nfe % 3 == 0:
+            orders = [3] * (m - 2) + [2, 1]
+        else:
+            orders = [3] * (m - 1) + [nfe % 3]
+
+        for i in range(len(orders)):
+            eps_cache = {}
+            t, t_next = ts[i], ts[i + 1]
+            if eta:
+                sd, su = get_ancestral_step(self.sigma(t), self.sigma(t_next), eta)
+                t_next_ = torch.minimum(t_end, self.t(sd))
+                su = (self.sigma(t_next) ** 2 - self.sigma(t_next_) ** 2) ** 0.5
+            else:
+                t_next_, su = t_next, 0.
+
+            eps, eps_cache = self.eps(eps_cache, 'eps', x, t)
+            denoised = x - self.sigma(t) * eps
+            if self.info_callback is not None:
+                self.info_callback({'x': x, 'i': i, 't': ts[i], 't_up': t, 'denoised': denoised})
+
+            if orders[i] == 1:
+                x, eps_cache = self.dpm_solver_1_step(x, t, t_next_, eps_cache=eps_cache)
+            elif orders[i] == 2:
+                x, eps_cache = self.dpm_solver_2_step(x, t, t_next_, eps_cache=eps_cache)
+            else:
+                x, eps_cache = self.dpm_solver_3_step(x, t, t_next_, eps_cache=eps_cache)
+
+            x = x + su * s_noise * noise_sampler(self.sigma(t), self.sigma(t_next))
+
+        return x
+
+    def dpm_solver_adaptive(self, x, t_start, t_end, order=3, rtol=0.05, atol=0.0078, h_init=0.05, pcoeff=0., icoeff=1., dcoeff=0., accept_safety=0.81, eta=0., s_noise=1., noise_sampler=None):
+        noise_sampler = default_noise_sampler(x) if noise_sampler is None else noise_sampler
+        if order not in {2, 3}:
+            raise ValueError('order should be 2 or 3')
+        forward = t_end > t_start
+        if not forward and eta:
+            raise ValueError('eta must be 0 for reverse sampling')
+        h_init = abs(h_init) * (1 if forward else -1)
+        atol = torch.tensor(atol)
+        rtol = torch.tensor(rtol)
+        s = t_start
+        x_prev = x
+        accept = True
+        pid = PIDStepSizeController(h_init, pcoeff, icoeff, dcoeff, 1.5 if eta else order, accept_safety)
+        info = {'steps': 0, 'nfe': 0, 'n_accept': 0, 'n_reject': 0}
+
+        while s < t_end - 1e-5 if forward else s > t_end + 1e-5:
+            eps_cache = {}
+            t = torch.minimum(t_end, s + pid.h) if forward else torch.maximum(t_end, s + pid.h)
+            if eta:
+                sd, su = get_ancestral_step(self.sigma(s), self.sigma(t), eta)
+                t_ = torch.minimum(t_end, self.t(sd))
+                su = (self.sigma(t) ** 2 - self.sigma(t_) ** 2) ** 0.5
+            else:
+                t_, su = t, 0.
+
+            eps, eps_cache = self.eps(eps_cache, 'eps', x, s)
+            denoised = x - self.sigma(s) * eps
+
+            if order == 2:
+                x_low, eps_cache = self.dpm_solver_1_step(x, s, t_, eps_cache=eps_cache)
+                x_high, eps_cache = self.dpm_solver_2_step(x, s, t_, eps_cache=eps_cache)
+            else:
+                x_low, eps_cache = self.dpm_solver_2_step(x, s, t_, r1=1 / 3, eps_cache=eps_cache)
+                x_high, eps_cache = self.dpm_solver_3_step(x, s, t_, eps_cache=eps_cache)
+            delta = torch.maximum(atol, rtol * torch.maximum(x_low.abs(), x_prev.abs()))
+            error = torch.linalg.norm((x_low - x_high) / delta) / x.numel() ** 0.5
+            accept = pid.propose_step(error)
+            if accept:
+                x_prev = x_low
+                x = x_high + su * s_noise * noise_sampler(self.sigma(s), self.sigma(t))
+                s = t
+                info['n_accept'] += 1
+            else:
+                info['n_reject'] += 1
+            info['nfe'] += order
+            info['steps'] += 1
+
+            if self.info_callback is not None:
+                self.info_callback({'x': x, 'i': info['steps'] - 1, 't': s, 't_up': s, 'denoised': denoised, 'error': error, 'h': pid.h, **info})
+
+        return x, info
+
+
+@torch.no_grad()
+def sample_dpm_fast(model, x, sigma_min, sigma_max, n, extra_args=None, callback=None, disable=None, eta=0., s_noise=1., noise_sampler=None):
+    """DPM-Solver-Fast (fixed step size). See https://arxiv.org/abs/2206.00927."""
+    if sigma_min <= 0 or sigma_max <= 0:
+        raise ValueError('sigma_min and sigma_max must not be 0')
+    with tqdm(total=n, disable=disable) as pbar:
+        dpm_solver = DPMSolver(model, extra_args, eps_callback=pbar.update)
+        if callback is not None:
+            dpm_solver.info_callback = lambda info: callback({'sigma': dpm_solver.sigma(info['t']), 'sigma_hat': dpm_solver.sigma(info['t_up']), **info})
+        return dpm_solver.dpm_solver_fast(x, dpm_solver.t(torch.tensor(sigma_max)), dpm_solver.t(torch.tensor(sigma_min)), n, eta, s_noise, noise_sampler)
+
+
+@torch.no_grad()
+def sample_dpm_adaptive(model, x, sigma_min, sigma_max, extra_args=None, callback=None, disable=None, order=3, rtol=0.05, atol=0.0078, h_init=0.05, pcoeff=0., icoeff=1., dcoeff=0., accept_safety=0.81, eta=0., s_noise=1., noise_sampler=None, return_info=False):
+    """DPM-Solver-12 and 23 (adaptive step size). See https://arxiv.org/abs/2206.00927."""
+    if sigma_min <= 0 or sigma_max <= 0:
+        raise ValueError('sigma_min and sigma_max must not be 0')
+    with tqdm(disable=disable) as pbar:
+        dpm_solver = DPMSolver(model, extra_args, eps_callback=pbar.update)
+        if callback is not None:
+            dpm_solver.info_callback = lambda info: callback({'sigma': dpm_solver.sigma(info['t']), 'sigma_hat': dpm_solver.sigma(info['t_up']), **info})
+        x, info = dpm_solver.dpm_solver_adaptive(x, dpm_solver.t(torch.tensor(sigma_max)), dpm_solver.t(torch.tensor(sigma_min)), order, rtol, atol, h_init, pcoeff, icoeff, dcoeff, accept_safety, eta, s_noise, noise_sampler)
+    if return_info:
+        return x, info
+    return x
+
+
+@torch.no_grad()
+def sample_dpmpp_2s_ancestral(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None):
+    """Ancestral sampling with DPM-Solver++(2S) second-order steps."""
+    extra_args = {} if extra_args is None else extra_args
+    noise_sampler = default_noise_sampler(x) if noise_sampler is None else noise_sampler
+    s_in = x.new_ones([x.shape[0]])
+    sigma_fn = lambda t: t.neg().exp()
+    t_fn = lambda sigma: sigma.log().neg()
+
+    for i in trange(len(sigmas) - 1, disable=disable):
+        denoised = model(x, sigmas[i] * s_in, **extra_args)
+        sigma_down, sigma_up = get_ancestral_step(sigmas[i], sigmas[i + 1], eta=eta)
+        if callback is not None:
+            callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised})
+        if sigma_down == 0:
+            # Euler method
+            d = to_d(x, sigmas[i], denoised)
+            dt = sigma_down - sigmas[i]
+            x = x + d * dt
+        else:
+            # DPM-Solver++(2S)
+            t, t_next = t_fn(sigmas[i]), t_fn(sigma_down)
+            r = 1 / 2
+            h = t_next - t
+            s = t + r * h
+            x_2 = (sigma_fn(s) / sigma_fn(t)) * x - (-h * r).expm1() * denoised
+            denoised_2 = model(x_2, sigma_fn(s) * s_in, **extra_args)
+            x = (sigma_fn(t_next) / sigma_fn(t)) * x - (-h).expm1() * denoised_2
+        # Noise addition
+        if sigmas[i + 1] > 0:
+            x = x + noise_sampler(sigmas[i], sigmas[i + 1]) * s_noise * sigma_up
+    return x
+
+
+@torch.no_grad()
+def sample_dpmpp_sde(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None, r=1 / 2):
+    """DPM-Solver++ (stochastic)."""
+    sigma_min, sigma_max = sigmas[sigmas > 0].min(), sigmas.max()
+    seed = extra_args.get("seed", None)
+    noise_sampler = BrownianTreeNoiseSampler(x, sigma_min, sigma_max, seed=seed, cpu=True) if noise_sampler is None else noise_sampler
+    extra_args = {} if extra_args is None else extra_args
+    s_in = x.new_ones([x.shape[0]])
+    sigma_fn = lambda t: t.neg().exp()
+    t_fn = lambda sigma: sigma.log().neg()
+
+    for i in trange(len(sigmas) - 1, disable=disable):
+        denoised = model(x, sigmas[i] * s_in, **extra_args)
+        if callback is not None:
+            callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised})
+        if sigmas[i + 1] == 0:
+            # Euler method
+            d = to_d(x, sigmas[i], denoised)
+            dt = sigmas[i + 1] - sigmas[i]
+            x = x + d * dt
+        else:
+            # DPM-Solver++
+            t, t_next = t_fn(sigmas[i]), t_fn(sigmas[i + 1])
+            h = t_next - t
+            s = t + h * r
+            fac = 1 / (2 * r)
+
+            # Step 1
+            sd, su = get_ancestral_step(sigma_fn(t), sigma_fn(s), eta)
+            s_ = t_fn(sd)
+            x_2 = (sigma_fn(s_) / sigma_fn(t)) * x - (t - s_).expm1() * denoised
+            x_2 = x_2 + noise_sampler(sigma_fn(t), sigma_fn(s)) * s_noise * su
+            denoised_2 = model(x_2, sigma_fn(s) * s_in, **extra_args)
+
+            # Step 2
+            sd, su = get_ancestral_step(sigma_fn(t), sigma_fn(t_next), eta)
+            t_next_ = t_fn(sd)
+            denoised_d = (1 - fac) * denoised + fac * denoised_2
+            x = (sigma_fn(t_next_) / sigma_fn(t)) * x - (t - t_next_).expm1() * denoised_d
+            x = x + noise_sampler(sigma_fn(t), sigma_fn(t_next)) * s_noise * su
+    return x
+
+
+@torch.no_grad()
+def sample_dpmpp_2m(model, x, sigmas, extra_args=None, callback=None, disable=None):
+    """DPM-Solver++(2M)."""
+    extra_args = {} if extra_args is None else extra_args
+    s_in = x.new_ones([x.shape[0]])
+    sigma_fn = lambda t: t.neg().exp()
+    t_fn = lambda sigma: sigma.log().neg()
+    old_denoised = None
+
+    for i in trange(len(sigmas) - 1, disable=disable):
+        denoised = model(x, sigmas[i] * s_in, **extra_args)
+        if callback is not None:
+            callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised})
+        t, t_next = t_fn(sigmas[i]), t_fn(sigmas[i + 1])
+        h = t_next - t
+        if old_denoised is None or sigmas[i + 1] == 0:
+            x = (sigma_fn(t_next) / sigma_fn(t)) * x - (-h).expm1() * denoised
+        else:
+            h_last = t - t_fn(sigmas[i - 1])
+            r = h_last / h
+            denoised_d = (1 + 1 / (2 * r)) * denoised - (1 / (2 * r)) * old_denoised
+            x = (sigma_fn(t_next) / sigma_fn(t)) * x - (-h).expm1() * denoised_d
+        old_denoised = denoised
+    return x
+
+@torch.no_grad()
+def sample_dpmpp_2m_sde(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None, solver_type='midpoint'):
+    """DPM-Solver++(2M) SDE."""
+
+    if solver_type not in {'heun', 'midpoint'}:
+        raise ValueError('solver_type must be \'heun\' or \'midpoint\'')
+
+    seed = extra_args.get("seed", None)
+    sigma_min, sigma_max = sigmas[sigmas > 0].min(), sigmas.max()
+    noise_sampler = BrownianTreeNoiseSampler(x, sigma_min, sigma_max, seed=seed, cpu=True) if noise_sampler is None else noise_sampler
+    extra_args = {} if extra_args is None else extra_args
+    s_in = x.new_ones([x.shape[0]])
+
+    old_denoised = None
+    h_last = None
+    h = None
+
+    for i in trange(len(sigmas) - 1, disable=disable):
+        denoised = model(x, sigmas[i] * s_in, **extra_args)
+        if callback is not None:
+            callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised})
+        if sigmas[i + 1] == 0:
+            # Denoising step
+            x = denoised
+        else:
+            # DPM-Solver++(2M) SDE
+            t, s = -sigmas[i].log(), -sigmas[i + 1].log()
+            h = s - t
+            eta_h = eta * h
+
+            x = sigmas[i + 1] / sigmas[i] * (-eta_h).exp() * x + (-h - eta_h).expm1().neg() * denoised
+
+            if old_denoised is not None:
+                r = h_last / h
+                if solver_type == 'heun':
+                    x = x + ((-h - eta_h).expm1().neg() / (-h - eta_h) + 1) * (1 / r) * (denoised - old_denoised)
+                elif solver_type == 'midpoint':
+                    x = x + 0.5 * (-h - eta_h).expm1().neg() * (1 / r) * (denoised - old_denoised)
+
+            if eta:
+                x = x + noise_sampler(sigmas[i], sigmas[i + 1]) * sigmas[i + 1] * (-2 * eta_h).expm1().neg().sqrt() * s_noise
+
+        old_denoised = denoised
+        h_last = h
+    return x
+
+@torch.no_grad()
+def sample_dpmpp_3m_sde(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None):
+    """DPM-Solver++(3M) SDE."""
+
+    seed = extra_args.get("seed", None)
+    sigma_min, sigma_max = sigmas[sigmas > 0].min(), sigmas.max()
+    noise_sampler = BrownianTreeNoiseSampler(x, sigma_min, sigma_max, seed=seed, cpu=True) if noise_sampler is None else noise_sampler
+    extra_args = {} if extra_args is None else extra_args
+    s_in = x.new_ones([x.shape[0]])
+
+    denoised_1, denoised_2 = None, None
+    h, h_1, h_2 = None, None, None
+
+    for i in trange(len(sigmas) - 1, disable=disable):
+        denoised = model(x, sigmas[i] * s_in, **extra_args)
+        if callback is not None:
+            callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised})
+        if sigmas[i + 1] == 0:
+            # Denoising step
+            x = denoised
+        else:
+            t, s = -sigmas[i].log(), -sigmas[i + 1].log()
+            h = s - t
+            h_eta = h * (eta + 1)
+
+            x = torch.exp(-h_eta) * x + (-h_eta).expm1().neg() * denoised
+
+            if h_2 is not None:
+                r0 = h_1 / h
+                r1 = h_2 / h
+                d1_0 = (denoised - denoised_1) / r0
+                d1_1 = (denoised_1 - denoised_2) / r1
+                d1 = d1_0 + (d1_0 - d1_1) * r0 / (r0 + r1)
+                d2 = (d1_0 - d1_1) / (r0 + r1)
+                phi_2 = h_eta.neg().expm1() / h_eta + 1
+                phi_3 = phi_2 / h_eta - 0.5
+                x = x + phi_2 * d1 - phi_3 * d2
+            elif h_1 is not None:
+                r = h_1 / h
+                d = (denoised - denoised_1) / r
+                phi_2 = h_eta.neg().expm1() / h_eta + 1
+                x = x + phi_2 * d
+
+            if eta:
+                x = x + noise_sampler(sigmas[i], sigmas[i + 1]) * sigmas[i + 1] * (-2 * h * eta).expm1().neg().sqrt() * s_noise
+
+        denoised_1, denoised_2 = denoised, denoised_1
+        h_1, h_2 = h, h_1
+    return x
+
+@torch.no_grad()
+def sample_dpmpp_3m_sde_gpu(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None):
+    sigma_min, sigma_max = sigmas[sigmas > 0].min(), sigmas.max()
+    noise_sampler = BrownianTreeNoiseSampler(x, sigma_min, sigma_max, seed=extra_args.get("seed", None), cpu=False) if noise_sampler is None else noise_sampler
+    return sample_dpmpp_3m_sde(model, x, sigmas, extra_args=extra_args, callback=callback, disable=disable, eta=eta, s_noise=s_noise, noise_sampler=noise_sampler)
+
+@torch.no_grad()
+def sample_dpmpp_2m_sde_gpu(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None, solver_type='midpoint'):
+    sigma_min, sigma_max = sigmas[sigmas > 0].min(), sigmas.max()
+    noise_sampler = BrownianTreeNoiseSampler(x, sigma_min, sigma_max, seed=extra_args.get("seed", None), cpu=False) if noise_sampler is None else noise_sampler
+    return sample_dpmpp_2m_sde(model, x, sigmas, extra_args=extra_args, callback=callback, disable=disable, eta=eta, s_noise=s_noise, noise_sampler=noise_sampler, solver_type=solver_type)
+
+@torch.no_grad()
+def sample_dpmpp_sde_gpu(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None, r=1 / 2):
+    sigma_min, sigma_max = sigmas[sigmas > 0].min(), sigmas.max()
+    noise_sampler = BrownianTreeNoiseSampler(x, sigma_min, sigma_max, seed=extra_args.get("seed", None), cpu=False) if noise_sampler is None else noise_sampler
+    return sample_dpmpp_sde(model, x, sigmas, extra_args=extra_args, callback=callback, disable=disable, eta=eta, s_noise=s_noise, noise_sampler=noise_sampler, r=r)
+
+
+def DDPMSampler_step(x, sigma, sigma_prev, noise, noise_sampler):
+    alpha_cumprod = 1 / ((sigma * sigma) + 1)
+    alpha_cumprod_prev = 1 / ((sigma_prev * sigma_prev) + 1)
+    alpha = (alpha_cumprod / alpha_cumprod_prev)
+
+    mu = (1.0 / alpha).sqrt() * (x - (1 - alpha) * noise / (1 - alpha_cumprod).sqrt())
+    if sigma_prev > 0:
+        mu += ((1 - alpha) * (1. - alpha_cumprod_prev) / (1. - alpha_cumprod)).sqrt() * noise_sampler(sigma, sigma_prev)
+    return mu
+
+def generic_step_sampler(model, x, sigmas, extra_args=None, callback=None, disable=None, noise_sampler=None, step_function=None):
+    extra_args = {} if extra_args is None else extra_args
+    noise_sampler = default_noise_sampler(x) if noise_sampler is None else noise_sampler
+    s_in = x.new_ones([x.shape[0]])
+
+    for i in trange(len(sigmas) - 1, disable=disable):
+        denoised = model(x, sigmas[i] * s_in, **extra_args)
+        if callback is not None:
+            callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised})
+        x = step_function(x / torch.sqrt(1.0 + sigmas[i] ** 2.0), sigmas[i], sigmas[i + 1], (x - denoised) / sigmas[i], noise_sampler)
+        if sigmas[i + 1] != 0:
+            x *= torch.sqrt(1.0 + sigmas[i + 1] ** 2.0)
+    return x
+
+
+@torch.no_grad()
+def sample_ddpm(model, x, sigmas, extra_args=None, callback=None, disable=None, noise_sampler=None):
+    return generic_step_sampler(model, x, sigmas, extra_args, callback, disable, noise_sampler, DDPMSampler_step)
+
+@torch.no_grad()
+def sample_lcm(model, x, sigmas, extra_args=None, callback=None, disable=None, noise_sampler=None):
+    extra_args = {} if extra_args is None else extra_args
+    noise_sampler = default_noise_sampler(x) if noise_sampler is None else noise_sampler
+    s_in = x.new_ones([x.shape[0]])
+    for i in trange(len(sigmas) - 1, disable=disable):
+        denoised = model(x, sigmas[i] * s_in, **extra_args)
+        if callback is not None:
+            callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised})
+
+        x = denoised
+        if sigmas[i + 1] > 0:
+            x += sigmas[i + 1] * noise_sampler(sigmas[i], sigmas[i + 1])
+    return x
+
+
+
+@torch.no_grad()
+def sample_heunpp2(model, x, sigmas, extra_args=None, callback=None, disable=None, s_churn=0., s_tmin=0., s_tmax=float('inf'), s_noise=1.):
+    # From MIT licensed: https://github.com/Carzit/sd-webui-samplers-scheduler/
+    extra_args = {} if extra_args is None else extra_args
+    s_in = x.new_ones([x.shape[0]])
+    s_end = sigmas[-1]
+    for i in trange(len(sigmas) - 1, disable=disable):
+        gamma = min(s_churn / (len(sigmas) - 1), 2 ** 0.5 - 1) if s_tmin <= sigmas[i] <= s_tmax else 0.
+        eps = torch.randn_like(x) * s_noise
+        sigma_hat = sigmas[i] * (gamma + 1)
+        if gamma > 0:
+            x = x + eps * (sigma_hat ** 2 - sigmas[i] ** 2) ** 0.5
+        denoised = model(x, sigma_hat * s_in, **extra_args)
+        d = to_d(x, sigma_hat, denoised)
+        if callback is not None:
+            callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigma_hat, 'denoised': denoised})
+        dt = sigmas[i + 1] - sigma_hat
+        if sigmas[i + 1] == s_end:
+            # Euler method
+            x = x + d * dt
+        elif sigmas[i + 2] == s_end:
+
+            # Heun's method
+            x_2 = x + d * dt
+            denoised_2 = model(x_2, sigmas[i + 1] * s_in, **extra_args)
+            d_2 = to_d(x_2, sigmas[i + 1], denoised_2)
+
+            w = 2 * sigmas[0]
+            w2 = sigmas[i+1]/w
+            w1 = 1 - w2
+
+            d_prime = d * w1 + d_2 * w2
+
+
+            x = x + d_prime * dt
+
+        else:
+            # Heun++
+            x_2 = x + d * dt
+            denoised_2 = model(x_2, sigmas[i + 1] * s_in, **extra_args)
+            d_2 = to_d(x_2, sigmas[i + 1], denoised_2)
+            dt_2 = sigmas[i + 2] - sigmas[i + 1]
+
+            x_3 = x_2 + d_2 * dt_2
+            denoised_3 = model(x_3, sigmas[i + 2] * s_in, **extra_args)
+            d_3 = to_d(x_3, sigmas[i + 2], denoised_3)
+
+            w = 3 * sigmas[0]
+            w2 = sigmas[i + 1] / w
+            w3 = sigmas[i + 2] / w
+            w1 = 1 - w2 - w3
+
+            d_prime = w1 * d + w2 * d_2 + w3 * d_3
+            x = x + d_prime * dt
+    return x

ldm_patched/k_diffusion/utils.py

@@ -0,0 +1,313 @@
+from contextlib import contextmanager
+import hashlib
+import math
+from pathlib import Path
+import shutil
+import urllib
+import warnings
+
+from PIL import Image
+import torch
+from torch import nn, optim
+from torch.utils import data
+
+
+def hf_datasets_augs_helper(examples, transform, image_key, mode='RGB'):
+    """Apply passed in transforms for HuggingFace Datasets."""
+    images = [transform(image.convert(mode)) for image in examples[image_key]]
+    return {image_key: images}
+
+
+def append_dims(x, target_dims):
+    """Appends dimensions to the end of a tensor until it has target_dims dimensions."""
+    dims_to_append = target_dims - x.ndim
+    if dims_to_append < 0:
+        raise ValueError(f'input has {x.ndim} dims but target_dims is {target_dims}, which is less')
+    expanded = x[(...,) + (None,) * dims_to_append]
+    # MPS will get inf values if it tries to index into the new axes, but detaching fixes this.
+    # https://github.com/pytorch/pytorch/issues/84364
+    return expanded.detach().clone() if expanded.device.type == 'mps' else expanded
+
+
+def n_params(module):
+    """Returns the number of trainable parameters in a module."""
+    return sum(p.numel() for p in module.parameters())
+
+
+def download_file(path, url, digest=None):
+    """Downloads a file if it does not exist, optionally checking its SHA-256 hash."""
+    path = Path(path)
+    path.parent.mkdir(parents=True, exist_ok=True)
+    if not path.exists():
+        with urllib.request.urlopen(url) as response, open(path, 'wb') as f:
+            shutil.copyfileobj(response, f)
+    if digest is not None:
+        file_digest = hashlib.sha256(open(path, 'rb').read()).hexdigest()
+        if digest != file_digest:
+            raise OSError(f'hash of {path} (url: {url}) failed to validate')
+    return path
+
+
+@contextmanager
+def train_mode(model, mode=True):
+    """A context manager that places a model into training mode and restores
+    the previous mode on exit."""
+    modes = [module.training for module in model.modules()]
+    try:
+        yield model.train(mode)
+    finally:
+        for i, module in enumerate(model.modules()):
+            module.training = modes[i]
+
+
+def eval_mode(model):
+    """A context manager that places a model into evaluation mode and restores
+    the previous mode on exit."""
+    return train_mode(model, False)
+
+
+@torch.no_grad()
+def ema_update(model, averaged_model, decay):
+    """Incorporates updated model parameters into an exponential moving averaged
+    version of a model. It should be called after each optimizer step."""
+    model_params = dict(model.named_parameters())
+    averaged_params = dict(averaged_model.named_parameters())
+    assert model_params.keys() == averaged_params.keys()
+
+    for name, param in model_params.items():
+        averaged_params[name].mul_(decay).add_(param, alpha=1 - decay)
+
+    model_buffers = dict(model.named_buffers())
+    averaged_buffers = dict(averaged_model.named_buffers())
+    assert model_buffers.keys() == averaged_buffers.keys()
+
+    for name, buf in model_buffers.items():
+        averaged_buffers[name].copy_(buf)
+
+
+class EMAWarmup:
+    """Implements an EMA warmup using an inverse decay schedule.
+    If inv_gamma=1 and power=1, implements a simple average. inv_gamma=1, power=2/3 are
+    good values for models you plan to train for a million or more steps (reaches decay
+    factor 0.999 at 31.6K steps, 0.9999 at 1M steps), inv_gamma=1, power=3/4 for models
+    you plan to train for less (reaches decay factor 0.999 at 10K steps, 0.9999 at
+    215.4k steps).
+    Args:
+        inv_gamma (float): Inverse multiplicative factor of EMA warmup. Default: 1.
+        power (float): Exponential factor of EMA warmup. Default: 1.
+        min_value (float): The minimum EMA decay rate. Default: 0.
+        max_value (float): The maximum EMA decay rate. Default: 1.
+        start_at (int): The epoch to start averaging at. Default: 0.
+        last_epoch (int): The index of last epoch. Default: 0.
+    """
+
+    def __init__(self, inv_gamma=1., power=1., min_value=0., max_value=1., start_at=0,
+                 last_epoch=0):
+        self.inv_gamma = inv_gamma
+        self.power = power
+        self.min_value = min_value
+        self.max_value = max_value
+        self.start_at = start_at
+        self.last_epoch = last_epoch
+
+    def state_dict(self):
+        """Returns the state of the class as a :class:`dict`."""
+        return dict(self.__dict__.items())
+
+    def load_state_dict(self, state_dict):
+        """Loads the class's state.
+        Args:
+            state_dict (dict): scaler state. Should be an object returned
+                from a call to :meth:`state_dict`.
+        """
+        self.__dict__.update(state_dict)
+
+    def get_value(self):
+        """Gets the current EMA decay rate."""
+        epoch = max(0, self.last_epoch - self.start_at)
+        value = 1 - (1 + epoch / self.inv_gamma) ** -self.power
+        return 0. if epoch < 0 else min(self.max_value, max(self.min_value, value))
+
+    def step(self):
+        """Updates the step count."""
+        self.last_epoch += 1
+
+
+class InverseLR(optim.lr_scheduler._LRScheduler):
+    """Implements an inverse decay learning rate schedule with an optional exponential
+    warmup. When last_epoch=-1, sets initial lr as lr.
+    inv_gamma is the number of steps/epochs required for the learning rate to decay to
+    (1 / 2)**power of its original value.
+    Args:
+        optimizer (Optimizer): Wrapped optimizer.
+        inv_gamma (float): Inverse multiplicative factor of learning rate decay. Default: 1.
+        power (float): Exponential factor of learning rate decay. Default: 1.
+        warmup (float): Exponential warmup factor (0 <= warmup < 1, 0 to disable)
+            Default: 0.
+        min_lr (float): The minimum learning rate. Default: 0.
+        last_epoch (int): The index of last epoch. Default: -1.
+        verbose (bool): If ``True``, prints a message to stdout for
+            each update. Default: ``False``.
+    """
+
+    def __init__(self, optimizer, inv_gamma=1., power=1., warmup=0., min_lr=0.,
+                 last_epoch=-1, verbose=False):
+        self.inv_gamma = inv_gamma
+        self.power = power
+        if not 0. <= warmup < 1:
+            raise ValueError('Invalid value for warmup')
+        self.warmup = warmup
+        self.min_lr = min_lr
+        super().__init__(optimizer, last_epoch, verbose)
+
+    def get_lr(self):
+        if not self._get_lr_called_within_step:
+            warnings.warn("To get the last learning rate computed by the scheduler, "
+                          "please use `get_last_lr()`.")
+
+        return self._get_closed_form_lr()
+
+    def _get_closed_form_lr(self):
+        warmup = 1 - self.warmup ** (self.last_epoch + 1)
+        lr_mult = (1 + self.last_epoch / self.inv_gamma) ** -self.power
+        return [warmup * max(self.min_lr, base_lr * lr_mult)
+                for base_lr in self.base_lrs]
+
+
+class ExponentialLR(optim.lr_scheduler._LRScheduler):
+    """Implements an exponential learning rate schedule with an optional exponential
+    warmup. When last_epoch=-1, sets initial lr as lr. Decays the learning rate
+    continuously by decay (default 0.5) every num_steps steps.
+    Args:
+        optimizer (Optimizer): Wrapped optimizer.
+        num_steps (float): The number of steps to decay the learning rate by decay in.
+        decay (float): The factor by which to decay the learning rate every num_steps
+            steps. Default: 0.5.
+        warmup (float): Exponential warmup factor (0 <= warmup < 1, 0 to disable)
+            Default: 0.
+        min_lr (float): The minimum learning rate. Default: 0.
+        last_epoch (int): The index of last epoch. Default: -1.
+        verbose (bool): If ``True``, prints a message to stdout for
+            each update. Default: ``False``.
+    """
+
+    def __init__(self, optimizer, num_steps, decay=0.5, warmup=0., min_lr=0.,
+                 last_epoch=-1, verbose=False):
+        self.num_steps = num_steps
+        self.decay = decay
+        if not 0. <= warmup < 1:
+            raise ValueError('Invalid value for warmup')
+        self.warmup = warmup
+        self.min_lr = min_lr
+        super().__init__(optimizer, last_epoch, verbose)
+
+    def get_lr(self):
+        if not self._get_lr_called_within_step:
+            warnings.warn("To get the last learning rate computed by the scheduler, "
+                          "please use `get_last_lr()`.")
+
+        return self._get_closed_form_lr()
+
+    def _get_closed_form_lr(self):
+        warmup = 1 - self.warmup ** (self.last_epoch + 1)
+        lr_mult = (self.decay ** (1 / self.num_steps)) ** self.last_epoch
+        return [warmup * max(self.min_lr, base_lr * lr_mult)
+                for base_lr in self.base_lrs]
+
+
+def rand_log_normal(shape, loc=0., scale=1., device='cpu', dtype=torch.float32):
+    """Draws samples from an lognormal distribution."""
+    return (torch.randn(shape, device=device, dtype=dtype) * scale + loc).exp()
+
+
+def rand_log_logistic(shape, loc=0., scale=1., min_value=0., max_value=float('inf'), device='cpu', dtype=torch.float32):
+    """Draws samples from an optionally truncated log-logistic distribution."""
+    min_value = torch.as_tensor(min_value, device=device, dtype=torch.float64)
+    max_value = torch.as_tensor(max_value, device=device, dtype=torch.float64)
+    min_cdf = min_value.log().sub(loc).div(scale).sigmoid()
+    max_cdf = max_value.log().sub(loc).div(scale).sigmoid()
+    u = torch.rand(shape, device=device, dtype=torch.float64) * (max_cdf - min_cdf) + min_cdf
+    return u.logit().mul(scale).add(loc).exp().to(dtype)
+
+
+def rand_log_uniform(shape, min_value, max_value, device='cpu', dtype=torch.float32):
+    """Draws samples from an log-uniform distribution."""
+    min_value = math.log(min_value)
+    max_value = math.log(max_value)
+    return (torch.rand(shape, device=device, dtype=dtype) * (max_value - min_value) + min_value).exp()
+
+
+def rand_v_diffusion(shape, sigma_data=1., min_value=0., max_value=float('inf'), device='cpu', dtype=torch.float32):
+    """Draws samples from a truncated v-diffusion training timestep distribution."""
+    min_cdf = math.atan(min_value / sigma_data) * 2 / math.pi
+    max_cdf = math.atan(max_value / sigma_data) * 2 / math.pi
+    u = torch.rand(shape, device=device, dtype=dtype) * (max_cdf - min_cdf) + min_cdf
+    return torch.tan(u * math.pi / 2) * sigma_data
+
+
+def rand_split_log_normal(shape, loc, scale_1, scale_2, device='cpu', dtype=torch.float32):
+    """Draws samples from a split lognormal distribution."""
+    n = torch.randn(shape, device=device, dtype=dtype).abs()
+    u = torch.rand(shape, device=device, dtype=dtype)
+    n_left = n * -scale_1 + loc
+    n_right = n * scale_2 + loc
+    ratio = scale_1 / (scale_1 + scale_2)
+    return torch.where(u < ratio, n_left, n_right).exp()
+
+
+class FolderOfImages(data.Dataset):
+    """Recursively finds all images in a directory. It does not support
+    classes/targets."""
+
+    IMG_EXTENSIONS = {'.jpg', '.jpeg', '.png', '.ppm', '.bmp', '.pgm', '.tif', '.tiff', '.webp'}
+
+    def __init__(self, root, transform=None):
+        super().__init__()
+        self.root = Path(root)
+        self.transform = nn.Identity() if transform is None else transform
+        self.paths = sorted(path for path in self.root.rglob('*') if path.suffix.lower() in self.IMG_EXTENSIONS)
+
+    def __repr__(self):
+        return f'FolderOfImages(root="{self.root}", len: {len(self)})'
+
+    def __len__(self):
+        return len(self.paths)
+
+    def __getitem__(self, key):
+        path = self.paths[key]
+        with open(path, 'rb') as f:
+            image = Image.open(f).convert('RGB')
+        image = self.transform(image)
+        return image,
+
+
+class CSVLogger:
+    def __init__(self, filename, columns):
+        self.filename = Path(filename)
+        self.columns = columns
+        if self.filename.exists():
+            self.file = open(self.filename, 'a')
+        else:
+            self.file = open(self.filename, 'w')
+            self.write(*self.columns)
+
+    def write(self, *args):
+        print(*args, sep=',', file=self.file, flush=True)
+
+
+@contextmanager
+def tf32_mode(cudnn=None, matmul=None):
+    """A context manager that sets whether TF32 is allowed on cuDNN or matmul."""
+    cudnn_old = torch.backends.cudnn.allow_tf32
+    matmul_old = torch.backends.cuda.matmul.allow_tf32
+    try:
+        if cudnn is not None:
+            torch.backends.cudnn.allow_tf32 = cudnn
+        if matmul is not None:
+            torch.backends.cuda.matmul.allow_tf32 = matmul
+        yield
+    finally:
+        if cudnn is not None:
+            torch.backends.cudnn.allow_tf32 = cudnn_old
+        if matmul is not None:
+            torch.backends.cuda.matmul.allow_tf32 = matmul_old

ldm_patched/ldm/models/autoencoder.py

@@ -0,0 +1,227 @@
+import torch
+# import pytorch_lightning as pl
+import torch.nn.functional as F
+from contextlib import contextmanager
+from typing import Any, Dict, List, Optional, Tuple, Union
+
+from ldm_patched.ldm.modules.distributions.distributions import DiagonalGaussianDistribution
+
+from ldm_patched.ldm.util import instantiate_from_config
+from ldm_patched.ldm.modules.ema import LitEma
+
+class DiagonalGaussianRegularizer(torch.nn.Module):
+    def __init__(self, sample: bool = True):
+        super().__init__()
+        self.sample = sample
+
+    def get_trainable_parameters(self) -> Any:
+        yield from ()
+
+    def forward(self, z: torch.Tensor) -> Tuple[torch.Tensor, dict]:
+        log = dict()
+        posterior = DiagonalGaussianDistribution(z)
+        if self.sample:
+            z = posterior.sample()
+        else:
+            z = posterior.mode()
+        kl_loss = posterior.kl()
+        kl_loss = torch.sum(kl_loss) / kl_loss.shape[0]
+        log["kl_loss"] = kl_loss
+        return z, log
+
+
+class AbstractAutoencoder(torch.nn.Module):
+    """
+    This is the base class for all autoencoders, including image autoencoders, image autoencoders with discriminators,
+    unCLIP models, etc. Hence, it is fairly general, and specific features
+    (e.g. discriminator training, encoding, decoding) must be implemented in subclasses.
+    """
+
+    def __init__(
+        self,
+        ema_decay: Union[None, float] = None,
+        monitor: Union[None, str] = None,
+        input_key: str = "jpg",
+        **kwargs,
+    ):
+        super().__init__()
+
+        self.input_key = input_key
+        self.use_ema = ema_decay is not None
+        if monitor is not None:
+            self.monitor = monitor
+
+        if self.use_ema:
+            self.model_ema = LitEma(self, decay=ema_decay)
+            logpy.info(f"Keeping EMAs of {len(list(self.model_ema.buffers()))}.")
+
+    def get_input(self, batch) -> Any:
+        raise NotImplementedError()
+
+    def on_train_batch_end(self, *args, **kwargs):
+        # for EMA computation
+        if self.use_ema:
+            self.model_ema(self)
+
+    @contextmanager
+    def ema_scope(self, context=None):
+        if self.use_ema:
+            self.model_ema.store(self.parameters())
+            self.model_ema.copy_to(self)
+            if context is not None:
+                logpy.info(f"{context}: Switched to EMA weights")
+        try:
+            yield None
+        finally:
+            if self.use_ema:
+                self.model_ema.restore(self.parameters())
+                if context is not None:
+                    logpy.info(f"{context}: Restored training weights")
+
+    def encode(self, *args, **kwargs) -> torch.Tensor:
+        raise NotImplementedError("encode()-method of abstract base class called")
+
+    def decode(self, *args, **kwargs) -> torch.Tensor:
+        raise NotImplementedError("decode()-method of abstract base class called")
+
+    def instantiate_optimizer_from_config(self, params, lr, cfg):
+        logpy.info(f"loading >>> {cfg['target']} <<< optimizer from config")
+        return get_obj_from_str(cfg["target"])(
+            params, lr=lr, **cfg.get("params", dict())
+        )
+
+    def configure_optimizers(self) -> Any:
+        raise NotImplementedError()
+
+
+class AutoencodingEngine(AbstractAutoencoder):
+    """
+    Base class for all image autoencoders that we train, like VQGAN or AutoencoderKL
+    (we also restore them explicitly as special cases for legacy reasons).
+    Regularizations such as KL or VQ are moved to the regularizer class.
+    """
+
+    def __init__(
+        self,
+        *args,
+        encoder_config: Dict,
+        decoder_config: Dict,
+        regularizer_config: Dict,
+        **kwargs,
+    ):
+        super().__init__(*args, **kwargs)
+
+        self.encoder: torch.nn.Module = instantiate_from_config(encoder_config)
+        self.decoder: torch.nn.Module = instantiate_from_config(decoder_config)
+        self.regularization: AbstractRegularizer = instantiate_from_config(
+            regularizer_config
+        )
+
+    def get_last_layer(self):
+        return self.decoder.get_last_layer()
+
+    def encode(
+        self,
+        x: torch.Tensor,
+        return_reg_log: bool = False,
+        unregularized: bool = False,
+    ) -> Union[torch.Tensor, Tuple[torch.Tensor, dict]]:
+        z = self.encoder(x)
+        if unregularized:
+            return z, dict()
+        z, reg_log = self.regularization(z)
+        if return_reg_log:
+            return z, reg_log
+        return z
+
+    def decode(self, z: torch.Tensor, **kwargs) -> torch.Tensor:
+        x = self.decoder(z, **kwargs)
+        return x
+
+    def forward(
+        self, x: torch.Tensor, **additional_decode_kwargs
+    ) -> Tuple[torch.Tensor, torch.Tensor, dict]:
+        z, reg_log = self.encode(x, return_reg_log=True)
+        dec = self.decode(z, **additional_decode_kwargs)
+        return z, dec, reg_log
+
+
+class AutoencodingEngineLegacy(AutoencodingEngine):
+    def __init__(self, embed_dim: int, **kwargs):
+        self.max_batch_size = kwargs.pop("max_batch_size", None)
+        ddconfig = kwargs.pop("ddconfig")
+        super().__init__(
+            encoder_config={
+                "target": "ldm_patched.ldm.modules.diffusionmodules.model.Encoder",
+                "params": ddconfig,
+            },
+            decoder_config={
+                "target": "ldm_patched.ldm.modules.diffusionmodules.model.Decoder",
+                "params": ddconfig,
+            },
+            **kwargs,
+        )
+        self.quant_conv = torch.nn.Conv2d(
+            (1 + ddconfig["double_z"]) * ddconfig["z_channels"],
+            (1 + ddconfig["double_z"]) * embed_dim,
+            1,
+        )
+        self.post_quant_conv = torch.nn.Conv2d(embed_dim, ddconfig["z_channels"], 1)
+        self.embed_dim = embed_dim
+
+    def get_autoencoder_params(self) -> list:
+        params = super().get_autoencoder_params()
+        return params
+
+    def encode(
+        self, x: torch.Tensor, return_reg_log: bool = False
+    ) -> Union[torch.Tensor, Tuple[torch.Tensor, dict]]:
+        if self.max_batch_size is None:
+            z = self.encoder(x)
+            z = self.quant_conv(z)
+        else:
+            N = x.shape[0]
+            bs = self.max_batch_size
+            n_batches = int(math.ceil(N / bs))
+            z = list()
+            for i_batch in range(n_batches):
+                z_batch = self.encoder(x[i_batch * bs : (i_batch + 1) * bs])
+                z_batch = self.quant_conv(z_batch)
+                z.append(z_batch)
+            z = torch.cat(z, 0)
+
+        z, reg_log = self.regularization(z)
+        if return_reg_log:
+            return z, reg_log
+        return z
+
+    def decode(self, z: torch.Tensor, **decoder_kwargs) -> torch.Tensor:
+        if self.max_batch_size is None:
+            dec = self.post_quant_conv(z)
+            dec = self.decoder(dec, **decoder_kwargs)
+        else:
+            N = z.shape[0]
+            bs = self.max_batch_size
+            n_batches = int(math.ceil(N / bs))
+            dec = list()
+            for i_batch in range(n_batches):
+                dec_batch = self.post_quant_conv(z[i_batch * bs : (i_batch + 1) * bs])
+                dec_batch = self.decoder(dec_batch, **decoder_kwargs)
+                dec.append(dec_batch)
+            dec = torch.cat(dec, 0)
+
+        return dec
+
+
+class AutoencoderKL(AutoencodingEngineLegacy):
+    def __init__(self, **kwargs):
+        if "lossconfig" in kwargs:
+            kwargs["loss_config"] = kwargs.pop("lossconfig")
+        super().__init__(
+            regularizer_config={
+                "target": (
+                    "ldm_patched.ldm.models.autoencoder.DiagonalGaussianRegularizer"
+                )
+            },
+            **kwargs,
+        )

ldm_patched/ldm/modules/attention.py

@@ -0,0 +1,772 @@
+from inspect import isfunction
+import math
+import torch
+import torch.nn.functional as F
+from torch import nn, einsum
+from einops import rearrange, repeat
+from typing import Optional, Any
+from functools import partial
+
+
+from .diffusionmodules.util import checkpoint, AlphaBlender, timestep_embedding
+from .sub_quadratic_attention import efficient_dot_product_attention
+
+from ldm_patched.modules import model_management
+
+if model_management.xformers_enabled():
+    import xformers
+    import xformers.ops
+
+from ldm_patched.modules.args_parser import args
+import ldm_patched.modules.ops
+ops = ldm_patched.modules.ops.disable_weight_init
+
+# CrossAttn precision handling
+if args.disable_attention_upcast:
+    print("disabling upcasting of attention")
+    _ATTN_PRECISION = "fp16"
+else:
+    _ATTN_PRECISION = "fp32"
+
+
+def exists(val):
+    return val is not None
+
+
+def uniq(arr):
+    return{el: True for el in arr}.keys()
+
+
+def default(val, d):
+    if exists(val):
+        return val
+    return d
+
+
+def max_neg_value(t):
+    return -torch.finfo(t.dtype).max
+
+
+def init_(tensor):
+    dim = tensor.shape[-1]
+    std = 1 / math.sqrt(dim)
+    tensor.uniform_(-std, std)
+    return tensor
+
+
+# feedforward
+class GEGLU(nn.Module):
+    def __init__(self, dim_in, dim_out, dtype=None, device=None, operations=ops):
+        super().__init__()
+        self.proj = operations.Linear(dim_in, dim_out * 2, dtype=dtype, device=device)
+
+    def forward(self, x):
+        x, gate = self.proj(x).chunk(2, dim=-1)
+        return x * F.gelu(gate)
+
+
+class FeedForward(nn.Module):
+    def __init__(self, dim, dim_out=None, mult=4, glu=False, dropout=0., dtype=None, device=None, operations=ops):
+        super().__init__()
+        inner_dim = int(dim * mult)
+        dim_out = default(dim_out, dim)
+        project_in = nn.Sequential(
+            operations.Linear(dim, inner_dim, dtype=dtype, device=device),
+            nn.GELU()
+        ) if not glu else GEGLU(dim, inner_dim, dtype=dtype, device=device, operations=operations)
+
+        self.net = nn.Sequential(
+            project_in,
+            nn.Dropout(dropout),
+            operations.Linear(inner_dim, dim_out, dtype=dtype, device=device)
+        )
+
+    def forward(self, x):
+        return self.net(x)
+
+def Normalize(in_channels, dtype=None, device=None):
+    return torch.nn.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True, dtype=dtype, device=device)
+
+def attention_basic(q, k, v, heads, mask=None):
+    b, _, dim_head = q.shape
+    dim_head //= heads
+    scale = dim_head ** -0.5
+
+    h = heads
+    q, k, v = map(
+        lambda t: t.unsqueeze(3)
+        .reshape(b, -1, heads, dim_head)
+        .permute(0, 2, 1, 3)
+        .reshape(b * heads, -1, dim_head)
+        .contiguous(),
+        (q, k, v),
+    )
+
+    # force cast to fp32 to avoid overflowing
+    if _ATTN_PRECISION =="fp32":
+        sim = einsum('b i d, b j d -> b i j', q.float(), k.float()) * scale
+    else:
+        sim = einsum('b i d, b j d -> b i j', q, k) * scale
+
+    del q, k
+
+    if exists(mask):
+        if mask.dtype == torch.bool:
+            mask = rearrange(mask, 'b ... -> b (...)') #TODO: check if this bool part matches pytorch attention
+            max_neg_value = -torch.finfo(sim.dtype).max
+            mask = repeat(mask, 'b j -> (b h) () j', h=h)
+            sim.masked_fill_(~mask, max_neg_value)
+        else:
+            sim += mask
+
+    # attention, what we cannot get enough of
+    sim = sim.softmax(dim=-1)
+
+    out = einsum('b i j, b j d -> b i d', sim.to(v.dtype), v)
+    out = (
+        out.unsqueeze(0)
+        .reshape(b, heads, -1, dim_head)
+        .permute(0, 2, 1, 3)
+        .reshape(b, -1, heads * dim_head)
+    )
+    return out
+
+
+def attention_sub_quad(query, key, value, heads, mask=None):
+    b, _, dim_head = query.shape
+    dim_head //= heads
+
+    scale = dim_head ** -0.5
+    query = query.unsqueeze(3).reshape(b, -1, heads, dim_head).permute(0, 2, 1, 3).reshape(b * heads, -1, dim_head)
+    value = value.unsqueeze(3).reshape(b, -1, heads, dim_head).permute(0, 2, 1, 3).reshape(b * heads, -1, dim_head)
+
+    key = key.unsqueeze(3).reshape(b, -1, heads, dim_head).permute(0, 2, 3, 1).reshape(b * heads, dim_head, -1)
+
+    dtype = query.dtype
+    upcast_attention = _ATTN_PRECISION =="fp32" and query.dtype != torch.float32
+    if upcast_attention:
+        bytes_per_token = torch.finfo(torch.float32).bits//8
+    else:
+        bytes_per_token = torch.finfo(query.dtype).bits//8
+    batch_x_heads, q_tokens, _ = query.shape
+    _, _, k_tokens = key.shape
+    qk_matmul_size_bytes = batch_x_heads * bytes_per_token * q_tokens * k_tokens
+
+    mem_free_total, mem_free_torch = model_management.get_free_memory(query.device, True)
+
+    kv_chunk_size_min = None
+    kv_chunk_size = None
+    query_chunk_size = None
+
+    for x in [4096, 2048, 1024, 512, 256]:
+        count = mem_free_total / (batch_x_heads * bytes_per_token * x * 4.0)
+        if count >= k_tokens:
+            kv_chunk_size = k_tokens
+            query_chunk_size = x
+            break
+
+    if query_chunk_size is None:
+        query_chunk_size = 512
+
+    hidden_states = efficient_dot_product_attention(
+        query,
+        key,
+        value,
+        query_chunk_size=query_chunk_size,
+        kv_chunk_size=kv_chunk_size,
+        kv_chunk_size_min=kv_chunk_size_min,
+        use_checkpoint=False,
+        upcast_attention=upcast_attention,
+    )
+
+    hidden_states = hidden_states.to(dtype)
+
+    hidden_states = hidden_states.unflatten(0, (-1, heads)).transpose(1,2).flatten(start_dim=2)
+    return hidden_states
+
+def attention_split(q, k, v, heads, mask=None):
+    b, _, dim_head = q.shape
+    dim_head //= heads
+    scale = dim_head ** -0.5
+
+    h = heads
+    q, k, v = map(
+        lambda t: t.unsqueeze(3)
+        .reshape(b, -1, heads, dim_head)
+        .permute(0, 2, 1, 3)
+        .reshape(b * heads, -1, dim_head)
+        .contiguous(),
+        (q, k, v),
+    )
+
+    r1 = torch.zeros(q.shape[0], q.shape[1], v.shape[2], device=q.device, dtype=q.dtype)
+
+    mem_free_total = model_management.get_free_memory(q.device)
+
+    if _ATTN_PRECISION =="fp32":
+        element_size = 4
+    else:
+        element_size = q.element_size()
+
+    gb = 1024 ** 3
+    tensor_size = q.shape[0] * q.shape[1] * k.shape[1] * element_size
+    modifier = 3
+    mem_required = tensor_size * modifier
+    steps = 1
+
+
+    if mem_required > mem_free_total:
+        steps = 2**(math.ceil(math.log(mem_required / mem_free_total, 2)))
+        # print(f"Expected tensor size:{tensor_size/gb:0.1f}GB, cuda free:{mem_free_cuda/gb:0.1f}GB "
+        #      f"torch free:{mem_free_torch/gb:0.1f} total:{mem_free_total/gb:0.1f} steps:{steps}")
+
+    if steps > 64:
+        max_res = math.floor(math.sqrt(math.sqrt(mem_free_total / 2.5)) / 8) * 64
+        raise RuntimeError(f'Not enough memory, use lower resolution (max approx. {max_res}x{max_res}). '
+                            f'Need: {mem_required/64/gb:0.1f}GB free, Have:{mem_free_total/gb:0.1f}GB free')
+
+    # print("steps", steps, mem_required, mem_free_total, modifier, q.element_size(), tensor_size)
+    first_op_done = False
+    cleared_cache = False
+    while True:
+        try:
+            slice_size = q.shape[1] // steps if (q.shape[1] % steps) == 0 else q.shape[1]
+            for i in range(0, q.shape[1], slice_size):
+                end = i + slice_size
+                if _ATTN_PRECISION =="fp32":
+                    with torch.autocast(enabled=False, device_type = 'cuda'):
+                        s1 = einsum('b i d, b j d -> b i j', q[:, i:end].float(), k.float()) * scale
+                else:
+                    s1 = einsum('b i d, b j d -> b i j', q[:, i:end], k) * scale
+
+                s2 = s1.softmax(dim=-1).to(v.dtype)
+                del s1
+                first_op_done = True
+
+                r1[:, i:end] = einsum('b i j, b j d -> b i d', s2, v)
+                del s2
+            break
+        except model_management.OOM_EXCEPTION as e:
+            if first_op_done == False:
+                model_management.soft_empty_cache(True)
+                if cleared_cache == False:
+                    cleared_cache = True
+                    print("out of memory error, emptying cache and trying again")
+                    continue
+                steps *= 2
+                if steps > 64:
+                    raise e
+                print("out of memory error, increasing steps and trying again", steps)
+            else:
+                raise e
+
+    del q, k, v
+
+    r1 = (
+        r1.unsqueeze(0)
+        .reshape(b, heads, -1, dim_head)
+        .permute(0, 2, 1, 3)
+        .reshape(b, -1, heads * dim_head)
+    )
+    return r1
+
+BROKEN_XFORMERS = False
+try:
+    x_vers = xformers.__version__
+    #I think 0.0.23 is also broken (q with bs bigger than 65535 gives CUDA error)
+    BROKEN_XFORMERS = x_vers.startswith("0.0.21") or x_vers.startswith("0.0.22") or x_vers.startswith("0.0.23")
+except:
+    pass
+
+def attention_xformers(q, k, v, heads, mask=None):
+    b, _, dim_head = q.shape
+    dim_head //= heads
+    if BROKEN_XFORMERS:
+        if b * heads > 65535:
+            return attention_pytorch(q, k, v, heads, mask)
+
+    q, k, v = map(
+        lambda t: t.unsqueeze(3)
+        .reshape(b, -1, heads, dim_head)
+        .permute(0, 2, 1, 3)
+        .reshape(b * heads, -1, dim_head)
+        .contiguous(),
+        (q, k, v),
+    )
+
+    # actually compute the attention, what we cannot get enough of
+    out = xformers.ops.memory_efficient_attention(q, k, v, attn_bias=None)
+
+    if exists(mask):
+        raise NotImplementedError
+    out = (
+        out.unsqueeze(0)
+        .reshape(b, heads, -1, dim_head)
+        .permute(0, 2, 1, 3)
+        .reshape(b, -1, heads * dim_head)
+    )
+    return out
+
+def attention_pytorch(q, k, v, heads, mask=None):
+    b, _, dim_head = q.shape
+    dim_head //= heads
+    q, k, v = map(
+        lambda t: t.view(b, -1, heads, dim_head).transpose(1, 2),
+        (q, k, v),
+    )
+
+    out = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False)
+    out = (
+        out.transpose(1, 2).reshape(b, -1, heads * dim_head)
+    )
+    return out
+
+
+optimized_attention = attention_basic
+optimized_attention_masked = attention_basic
+
+if model_management.xformers_enabled():
+    print("Using xformers cross attention")
+    optimized_attention = attention_xformers
+elif model_management.pytorch_attention_enabled():
+    print("Using pytorch cross attention")
+    optimized_attention = attention_pytorch
+else:
+    if args.attention_split:
+        print("Using split optimization for cross attention")
+        optimized_attention = attention_split
+    else:
+        print("Using sub quadratic optimization for cross attention, if you have memory or speed issues try using: --attention-split")
+        optimized_attention = attention_sub_quad
+
+if model_management.pytorch_attention_enabled():
+    optimized_attention_masked = attention_pytorch
+
+def optimized_attention_for_device(device, mask=False):
+    if device == torch.device("cpu"): #TODO
+        if model_management.pytorch_attention_enabled():
+            return attention_pytorch
+        else:
+            return attention_basic
+    if mask:
+        return optimized_attention_masked
+
+    return optimized_attention
+
+
+class CrossAttention(nn.Module):
+    def __init__(self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0., dtype=None, device=None, operations=ops):
+        super().__init__()
+        inner_dim = dim_head * heads
+        context_dim = default(context_dim, query_dim)
+
+        self.heads = heads
+        self.dim_head = dim_head
+
+        self.to_q = operations.Linear(query_dim, inner_dim, bias=False, dtype=dtype, device=device)
+        self.to_k = operations.Linear(context_dim, inner_dim, bias=False, dtype=dtype, device=device)
+        self.to_v = operations.Linear(context_dim, inner_dim, bias=False, dtype=dtype, device=device)
+
+        self.to_out = nn.Sequential(operations.Linear(inner_dim, query_dim, dtype=dtype, device=device), nn.Dropout(dropout))
+
+    def forward(self, x, context=None, value=None, mask=None):
+        q = self.to_q(x)
+        context = default(context, x)
+        k = self.to_k(context)
+        if value is not None:
+            v = self.to_v(value)
+            del value
+        else:
+            v = self.to_v(context)
+
+        if mask is None:
+            out = optimized_attention(q, k, v, self.heads)
+        else:
+            out = optimized_attention_masked(q, k, v, self.heads, mask)
+        return self.to_out(out)
+
+
+class BasicTransformerBlock(nn.Module):
+    def __init__(self, dim, n_heads, d_head, dropout=0., context_dim=None, gated_ff=True, checkpoint=True, ff_in=False, inner_dim=None,
+                 disable_self_attn=False, disable_temporal_crossattention=False, switch_temporal_ca_to_sa=False, dtype=None, device=None, operations=ops):
+        super().__init__()
+
+        self.ff_in = ff_in or inner_dim is not None
+        if inner_dim is None:
+            inner_dim = dim
+
+        self.is_res = inner_dim == dim
+
+        if self.ff_in:
+            self.norm_in = operations.LayerNorm(dim, dtype=dtype, device=device)
+            self.ff_in = FeedForward(dim, dim_out=inner_dim, dropout=dropout, glu=gated_ff, dtype=dtype, device=device, operations=operations)
+
+        self.disable_self_attn = disable_self_attn
+        self.attn1 = CrossAttention(query_dim=inner_dim, heads=n_heads, dim_head=d_head, dropout=dropout,
+                              context_dim=context_dim if self.disable_self_attn else None, dtype=dtype, device=device, operations=operations)  # is a self-attention if not self.disable_self_attn
+        self.ff = FeedForward(inner_dim, dim_out=dim, dropout=dropout, glu=gated_ff, dtype=dtype, device=device, operations=operations)
+
+        if disable_temporal_crossattention:
+            if switch_temporal_ca_to_sa:
+                raise ValueError
+            else:
+                self.attn2 = None
+        else:
+            context_dim_attn2 = None
+            if not switch_temporal_ca_to_sa:
+                context_dim_attn2 = context_dim
+
+            self.attn2 = CrossAttention(query_dim=inner_dim, context_dim=context_dim_attn2,
+                                heads=n_heads, dim_head=d_head, dropout=dropout, dtype=dtype, device=device, operations=operations)  # is self-attn if context is none
+            self.norm2 = operations.LayerNorm(inner_dim, dtype=dtype, device=device)
+
+        self.norm1 = operations.LayerNorm(inner_dim, dtype=dtype, device=device)
+        self.norm3 = operations.LayerNorm(inner_dim, dtype=dtype, device=device)
+        self.checkpoint = checkpoint
+        self.n_heads = n_heads
+        self.d_head = d_head
+        self.switch_temporal_ca_to_sa = switch_temporal_ca_to_sa
+
+    def forward(self, x, context=None, transformer_options={}):
+        return checkpoint(self._forward, (x, context, transformer_options), self.parameters(), self.checkpoint)
+
+    def _forward(self, x, context=None, transformer_options={}):
+        extra_options = {}
+        block = transformer_options.get("block", None)
+        block_index = transformer_options.get("block_index", 0)
+        transformer_patches = {}
+        transformer_patches_replace = {}
+
+        for k in transformer_options:
+            if k == "patches":
+                transformer_patches = transformer_options[k]
+            elif k == "patches_replace":
+                transformer_patches_replace = transformer_options[k]
+            else:
+                extra_options[k] = transformer_options[k]
+
+        extra_options["n_heads"] = self.n_heads
+        extra_options["dim_head"] = self.d_head
+
+        if self.ff_in:
+            x_skip = x
+            x = self.ff_in(self.norm_in(x))
+            if self.is_res:
+                x += x_skip
+
+        n = self.norm1(x)
+        if self.disable_self_attn:
+            context_attn1 = context
+        else:
+            context_attn1 = None
+        value_attn1 = None
+
+        if "attn1_patch" in transformer_patches:
+            patch = transformer_patches["attn1_patch"]
+            if context_attn1 is None:
+                context_attn1 = n
+            value_attn1 = context_attn1
+            for p in patch:
+                n, context_attn1, value_attn1 = p(n, context_attn1, value_attn1, extra_options)
+
+        if block is not None:
+            transformer_block = (block[0], block[1], block_index)
+        else:
+            transformer_block = None
+        attn1_replace_patch = transformer_patches_replace.get("attn1", {})
+        block_attn1 = transformer_block
+        if block_attn1 not in attn1_replace_patch:
+            block_attn1 = block
+
+        if block_attn1 in attn1_replace_patch:
+            if context_attn1 is None:
+                context_attn1 = n
+                value_attn1 = n
+            n = self.attn1.to_q(n)
+            context_attn1 = self.attn1.to_k(context_attn1)
+            value_attn1 = self.attn1.to_v(value_attn1)
+            n = attn1_replace_patch[block_attn1](n, context_attn1, value_attn1, extra_options)
+            n = self.attn1.to_out(n)
+        else:
+            n = self.attn1(n, context=context_attn1, value=value_attn1)
+
+        if "attn1_output_patch" in transformer_patches:
+            patch = transformer_patches["attn1_output_patch"]
+            for p in patch:
+                n = p(n, extra_options)
+
+        x += n
+        if "middle_patch" in transformer_patches:
+            patch = transformer_patches["middle_patch"]
+            for p in patch:
+                x = p(x, extra_options)
+
+        if self.attn2 is not None:
+            n = self.norm2(x)
+            if self.switch_temporal_ca_to_sa:
+                context_attn2 = n
+            else:
+                context_attn2 = context
+            value_attn2 = None
+            if "attn2_patch" in transformer_patches:
+                patch = transformer_patches["attn2_patch"]
+                value_attn2 = context_attn2
+                for p in patch:
+                    n, context_attn2, value_attn2 = p(n, context_attn2, value_attn2, extra_options)
+
+            attn2_replace_patch = transformer_patches_replace.get("attn2", {})
+            block_attn2 = transformer_block
+            if block_attn2 not in attn2_replace_patch:
+                block_attn2 = block
+
+            if block_attn2 in attn2_replace_patch:
+                if value_attn2 is None:
+                    value_attn2 = context_attn2
+                n = self.attn2.to_q(n)
+                context_attn2 = self.attn2.to_k(context_attn2)
+                value_attn2 = self.attn2.to_v(value_attn2)
+                n = attn2_replace_patch[block_attn2](n, context_attn2, value_attn2, extra_options)
+                n = self.attn2.to_out(n)
+            else:
+                n = self.attn2(n, context=context_attn2, value=value_attn2)
+
+        if "attn2_output_patch" in transformer_patches:
+            patch = transformer_patches["attn2_output_patch"]
+            for p in patch:
+                n = p(n, extra_options)
+
+        x += n
+        if self.is_res:
+            x_skip = x
+        x = self.ff(self.norm3(x))
+        if self.is_res:
+            x += x_skip
+
+        return x
+
+
+class SpatialTransformer(nn.Module):
+    """
+    Transformer block for image-like data.
+    First, project the input (aka embedding)
+    and reshape to b, t, d.
+    Then apply standard transformer action.
+    Finally, reshape to image
+    NEW: use_linear for more efficiency instead of the 1x1 convs
+    """
+    def __init__(self, in_channels, n_heads, d_head,
+                 depth=1, dropout=0., context_dim=None,
+                 disable_self_attn=False, use_linear=False,
+                 use_checkpoint=True, dtype=None, device=None, operations=ops):
+        super().__init__()
+        if exists(context_dim) and not isinstance(context_dim, list):
+            context_dim = [context_dim] * depth
+        self.in_channels = in_channels
+        inner_dim = n_heads * d_head
+        self.norm = operations.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True, dtype=dtype, device=device)
+        if not use_linear:
+            self.proj_in = operations.Conv2d(in_channels,
+                                     inner_dim,
+                                     kernel_size=1,
+                                     stride=1,
+                                     padding=0, dtype=dtype, device=device)
+        else:
+            self.proj_in = operations.Linear(in_channels, inner_dim, dtype=dtype, device=device)
+
+        self.transformer_blocks = nn.ModuleList(
+            [BasicTransformerBlock(inner_dim, n_heads, d_head, dropout=dropout, context_dim=context_dim[d],
+                                   disable_self_attn=disable_self_attn, checkpoint=use_checkpoint, dtype=dtype, device=device, operations=operations)
+                for d in range(depth)]
+        )
+        if not use_linear:
+            self.proj_out = operations.Conv2d(inner_dim,in_channels,
+                                                  kernel_size=1,
+                                                  stride=1,
+                                                  padding=0, dtype=dtype, device=device)
+        else:
+            self.proj_out = operations.Linear(in_channels, inner_dim, dtype=dtype, device=device)
+        self.use_linear = use_linear
+
+    def forward(self, x, context=None, transformer_options={}):
+        # note: if no context is given, cross-attention defaults to self-attention
+        if not isinstance(context, list):
+            context = [context] * len(self.transformer_blocks)
+        b, c, h, w = x.shape
+        x_in = x
+        x = self.norm(x)
+        if not self.use_linear:
+            x = self.proj_in(x)
+        x = rearrange(x, 'b c h w -> b (h w) c').contiguous()
+        if self.use_linear:
+            x = self.proj_in(x)
+        for i, block in enumerate(self.transformer_blocks):
+            transformer_options["block_index"] = i
+            x = block(x, context=context[i], transformer_options=transformer_options)
+        if self.use_linear:
+            x = self.proj_out(x)
+        x = rearrange(x, 'b (h w) c -> b c h w', h=h, w=w).contiguous()
+        if not self.use_linear:
+            x = self.proj_out(x)
+        return x + x_in
+
+
+class SpatialVideoTransformer(SpatialTransformer):
+    def __init__(
+        self,
+        in_channels,
+        n_heads,
+        d_head,
+        depth=1,
+        dropout=0.0,
+        use_linear=False,
+        context_dim=None,
+        use_spatial_context=False,
+        timesteps=None,
+        merge_strategy: str = "fixed",
+        merge_factor: float = 0.5,
+        time_context_dim=None,
+        ff_in=False,
+        checkpoint=False,
+        time_depth=1,
+        disable_self_attn=False,
+        disable_temporal_crossattention=False,
+        max_time_embed_period: int = 10000,
+        dtype=None, device=None, operations=ops
+    ):
+        super().__init__(
+            in_channels,
+            n_heads,
+            d_head,
+            depth=depth,
+            dropout=dropout,
+            use_checkpoint=checkpoint,
+            context_dim=context_dim,
+            use_linear=use_linear,
+            disable_self_attn=disable_self_attn,
+            dtype=dtype, device=device, operations=operations
+        )
+        self.time_depth = time_depth
+        self.depth = depth
+        self.max_time_embed_period = max_time_embed_period
+
+        time_mix_d_head = d_head
+        n_time_mix_heads = n_heads
+
+        time_mix_inner_dim = int(time_mix_d_head * n_time_mix_heads)
+
+        inner_dim = n_heads * d_head
+        if use_spatial_context:
+            time_context_dim = context_dim
+
+        self.time_stack = nn.ModuleList(
+            [
+                BasicTransformerBlock(
+                    inner_dim,
+                    n_time_mix_heads,
+                    time_mix_d_head,
+                    dropout=dropout,
+                    context_dim=time_context_dim,
+                    # timesteps=timesteps,
+                    checkpoint=checkpoint,
+                    ff_in=ff_in,
+                    inner_dim=time_mix_inner_dim,
+                    disable_self_attn=disable_self_attn,
+                    disable_temporal_crossattention=disable_temporal_crossattention,
+                    dtype=dtype, device=device, operations=operations
+                )
+                for _ in range(self.depth)
+            ]
+        )
+
+        assert len(self.time_stack) == len(self.transformer_blocks)
+
+        self.use_spatial_context = use_spatial_context
+        self.in_channels = in_channels
+
+        time_embed_dim = self.in_channels * 4
+        self.time_pos_embed = nn.Sequential(
+            operations.Linear(self.in_channels, time_embed_dim, dtype=dtype, device=device),
+            nn.SiLU(),
+            operations.Linear(time_embed_dim, self.in_channels, dtype=dtype, device=device),
+        )
+
+        self.time_mixer = AlphaBlender(
+            alpha=merge_factor, merge_strategy=merge_strategy
+        )
+
+    def forward(
+        self,
+        x: torch.Tensor,
+        context: Optional[torch.Tensor] = None,
+        time_context: Optional[torch.Tensor] = None,
+        timesteps: Optional[int] = None,
+        image_only_indicator: Optional[torch.Tensor] = None,
+        transformer_options={}
+    ) -> torch.Tensor:
+        _, _, h, w = x.shape
+        x_in = x
+        spatial_context = None
+        if exists(context):
+            spatial_context = context
+
+        if self.use_spatial_context:
+            assert (
+                context.ndim == 3
+            ), f"n dims of spatial context should be 3 but are {context.ndim}"
+
+            if time_context is None:
+                time_context = context
+            time_context_first_timestep = time_context[::timesteps]
+            time_context = repeat(
+                time_context_first_timestep, "b ... -> (b n) ...", n=h * w
+            )
+        elif time_context is not None and not self.use_spatial_context:
+            time_context = repeat(time_context, "b ... -> (b n) ...", n=h * w)
+            if time_context.ndim == 2:
+                time_context = rearrange(time_context, "b c -> b 1 c")
+
+        x = self.norm(x)
+        if not self.use_linear:
+            x = self.proj_in(x)
+        x = rearrange(x, "b c h w -> b (h w) c")
+        if self.use_linear:
+            x = self.proj_in(x)
+
+        num_frames = torch.arange(timesteps, device=x.device)
+        num_frames = repeat(num_frames, "t -> b t", b=x.shape[0] // timesteps)
+        num_frames = rearrange(num_frames, "b t -> (b t)")
+        t_emb = timestep_embedding(num_frames, self.in_channels, repeat_only=False, max_period=self.max_time_embed_period).to(x.dtype)
+        emb = self.time_pos_embed(t_emb)
+        emb = emb[:, None, :]
+
+        for it_, (block, mix_block) in enumerate(
+            zip(self.transformer_blocks, self.time_stack)
+        ):
+            transformer_options["block_index"] = it_
+            x = block(
+                x,
+                context=spatial_context,
+                transformer_options=transformer_options,
+            )
+
+            x_mix = x
+            x_mix = x_mix + emb
+
+            B, S, C = x_mix.shape
+            x_mix = rearrange(x_mix, "(b t) s c -> (b s) t c", t=timesteps)
+            x_mix = mix_block(x_mix, context=time_context) #TODO: transformer_options
+            x_mix = rearrange(
+                x_mix, "(b s) t c -> (b t) s c", s=S, b=B // timesteps, c=C, t=timesteps
+            )
+
+            x = self.time_mixer(x_spatial=x, x_temporal=x_mix, image_only_indicator=image_only_indicator)
+
+        if self.use_linear:
+            x = self.proj_out(x)
+        x = rearrange(x, "b (h w) c -> b c h w", h=h, w=w)
+        if not self.use_linear:
+            x = self.proj_out(x)
+        out = x + x_in
+        return out
+
+

ldm_patched/ldm/modules/diffusionmodules/model.py

@@ -0,0 +1,650 @@
+# pytorch_diffusion + derived encoder decoder
+import math
+import torch
+import torch.nn as nn
+import numpy as np
+from einops import rearrange
+from typing import Optional, Any
+
+from ldm_patched.modules import model_management
+import ldm_patched.modules.ops
+ops = ldm_patched.modules.ops.disable_weight_init
+
+if model_management.xformers_enabled_vae():
+    import xformers
+    import xformers.ops
+
+def get_timestep_embedding(timesteps, embedding_dim):
+    """
+    This matches the implementation in Denoising Diffusion Probabilistic Models:
+    From Fairseq.
+    Build sinusoidal embeddings.
+    This matches the implementation in tensor2tensor, but differs slightly
+    from the description in Section 3.5 of "Attention Is All You Need".
+    """
+    assert len(timesteps.shape) == 1
+
+    half_dim = embedding_dim // 2
+    emb = math.log(10000) / (half_dim - 1)
+    emb = torch.exp(torch.arange(half_dim, dtype=torch.float32) * -emb)
+    emb = emb.to(device=timesteps.device)
+    emb = timesteps.float()[:, None] * emb[None, :]
+    emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1)
+    if embedding_dim % 2 == 1:  # zero pad
+        emb = torch.nn.functional.pad(emb, (0,1,0,0))
+    return emb
+
+
+def nonlinearity(x):
+    # swish
+    return x*torch.sigmoid(x)
+
+
+def Normalize(in_channels, num_groups=32):
+    return torch.nn.GroupNorm(num_groups=num_groups, num_channels=in_channels, eps=1e-6, affine=True)
+
+
+class Upsample(nn.Module):
+    def __init__(self, in_channels, with_conv):
+        super().__init__()
+        self.with_conv = with_conv
+        if self.with_conv:
+            self.conv = ops.Conv2d(in_channels,
+                                        in_channels,
+                                        kernel_size=3,
+                                        stride=1,
+                                        padding=1)
+
+    def forward(self, x):
+        try:
+            x = torch.nn.functional.interpolate(x, scale_factor=2.0, mode="nearest")
+        except: #operation not implemented for bf16
+            b, c, h, w = x.shape
+            out = torch.empty((b, c, h*2, w*2), dtype=x.dtype, layout=x.layout, device=x.device)
+            split = 8
+            l = out.shape[1] // split
+            for i in range(0, out.shape[1], l):
+                out[:,i:i+l] = torch.nn.functional.interpolate(x[:,i:i+l].to(torch.float32), scale_factor=2.0, mode="nearest").to(x.dtype)
+            del x
+            x = out
+
+        if self.with_conv:
+            x = self.conv(x)
+        return x
+
+
+class Downsample(nn.Module):
+    def __init__(self, in_channels, with_conv):
+        super().__init__()
+        self.with_conv = with_conv
+        if self.with_conv:
+            # no asymmetric padding in torch conv, must do it ourselves
+            self.conv = ops.Conv2d(in_channels,
+                                        in_channels,
+                                        kernel_size=3,
+                                        stride=2,
+                                        padding=0)
+
+    def forward(self, x):
+        if self.with_conv:
+            pad = (0,1,0,1)
+            x = torch.nn.functional.pad(x, pad, mode="constant", value=0)
+            x = self.conv(x)
+        else:
+            x = torch.nn.functional.avg_pool2d(x, kernel_size=2, stride=2)
+        return x
+
+
+class ResnetBlock(nn.Module):
+    def __init__(self, *, in_channels, out_channels=None, conv_shortcut=False,
+                 dropout, temb_channels=512):
+        super().__init__()
+        self.in_channels = in_channels
+        out_channels = in_channels if out_channels is None else out_channels
+        self.out_channels = out_channels
+        self.use_conv_shortcut = conv_shortcut
+
+        self.swish = torch.nn.SiLU(inplace=True)
+        self.norm1 = Normalize(in_channels)
+        self.conv1 = ops.Conv2d(in_channels,
+                                     out_channels,
+                                     kernel_size=3,
+                                     stride=1,
+                                     padding=1)
+        if temb_channels > 0:
+            self.temb_proj = ops.Linear(temb_channels,
+                                             out_channels)
+        self.norm2 = Normalize(out_channels)
+        self.dropout = torch.nn.Dropout(dropout, inplace=True)
+        self.conv2 = ops.Conv2d(out_channels,
+                                     out_channels,
+                                     kernel_size=3,
+                                     stride=1,
+                                     padding=1)
+        if self.in_channels != self.out_channels:
+            if self.use_conv_shortcut:
+                self.conv_shortcut = ops.Conv2d(in_channels,
+                                                     out_channels,
+                                                     kernel_size=3,
+                                                     stride=1,
+                                                     padding=1)
+            else:
+                self.nin_shortcut = ops.Conv2d(in_channels,
+                                                    out_channels,
+                                                    kernel_size=1,
+                                                    stride=1,
+                                                    padding=0)
+
+    def forward(self, x, temb):
+        h = x
+        h = self.norm1(h)
+        h = self.swish(h)
+        h = self.conv1(h)
+
+        if temb is not None:
+            h = h + self.temb_proj(self.swish(temb))[:,:,None,None]
+
+        h = self.norm2(h)
+        h = self.swish(h)
+        h = self.dropout(h)
+        h = self.conv2(h)
+
+        if self.in_channels != self.out_channels:
+            if self.use_conv_shortcut:
+                x = self.conv_shortcut(x)
+            else:
+                x = self.nin_shortcut(x)
+
+        return x+h
+
+def slice_attention(q, k, v):
+    r1 = torch.zeros_like(k, device=q.device)
+    scale = (int(q.shape[-1])**(-0.5))
+
+    mem_free_total = model_management.get_free_memory(q.device)
+
+    gb = 1024 ** 3
+    tensor_size = q.shape[0] * q.shape[1] * k.shape[2] * q.element_size()
+    modifier = 3 if q.element_size() == 2 else 2.5
+    mem_required = tensor_size * modifier
+    steps = 1
+
+    if mem_required > mem_free_total:
+        steps = 2**(math.ceil(math.log(mem_required / mem_free_total, 2)))
+
+    while True:
+        try:
+            slice_size = q.shape[1] // steps if (q.shape[1] % steps) == 0 else q.shape[1]
+            for i in range(0, q.shape[1], slice_size):
+                end = i + slice_size
+                s1 = torch.bmm(q[:, i:end], k) * scale
+
+                s2 = torch.nn.functional.softmax(s1, dim=2).permute(0,2,1)
+                del s1
+
+                r1[:, :, i:end] = torch.bmm(v, s2)
+                del s2
+            break
+        except model_management.OOM_EXCEPTION as e:
+            model_management.soft_empty_cache(True)
+            steps *= 2
+            if steps > 128:
+                raise e
+            print("out of memory error, increasing steps and trying again", steps)
+
+    return r1
+
+def normal_attention(q, k, v):
+    # compute attention
+    b,c,h,w = q.shape
+
+    q = q.reshape(b,c,h*w)
+    q = q.permute(0,2,1)   # b,hw,c
+    k = k.reshape(b,c,h*w) # b,c,hw
+    v = v.reshape(b,c,h*w)
+
+    r1 = slice_attention(q, k, v)
+    h_ = r1.reshape(b,c,h,w)
+    del r1
+    return h_
+
+def xformers_attention(q, k, v):
+    # compute attention
+    B, C, H, W = q.shape
+    q, k, v = map(
+        lambda t: t.view(B, C, -1).transpose(1, 2).contiguous(),
+        (q, k, v),
+    )
+
+    try:
+        out = xformers.ops.memory_efficient_attention(q, k, v, attn_bias=None)
+        out = out.transpose(1, 2).reshape(B, C, H, W)
+    except NotImplementedError as e:
+        out = slice_attention(q.view(B, -1, C), k.view(B, -1, C).transpose(1, 2), v.view(B, -1, C).transpose(1, 2)).reshape(B, C, H, W)
+    return out
+
+def pytorch_attention(q, k, v):
+    # compute attention
+    B, C, H, W = q.shape
+    q, k, v = map(
+        lambda t: t.view(B, 1, C, -1).transpose(2, 3).contiguous(),
+        (q, k, v),
+    )
+
+    try:
+        out = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=None, dropout_p=0.0, is_causal=False)
+        out = out.transpose(2, 3).reshape(B, C, H, W)
+    except model_management.OOM_EXCEPTION as e:
+        print("scaled_dot_product_attention OOMed: switched to slice attention")
+        out = slice_attention(q.view(B, -1, C), k.view(B, -1, C).transpose(1, 2), v.view(B, -1, C).transpose(1, 2)).reshape(B, C, H, W)
+    return out
+
+
+class AttnBlock(nn.Module):
+    def __init__(self, in_channels):
+        super().__init__()
+        self.in_channels = in_channels
+
+        self.norm = Normalize(in_channels)
+        self.q = ops.Conv2d(in_channels,
+                                 in_channels,
+                                 kernel_size=1,
+                                 stride=1,
+                                 padding=0)
+        self.k = ops.Conv2d(in_channels,
+                                 in_channels,
+                                 kernel_size=1,
+                                 stride=1,
+                                 padding=0)
+        self.v = ops.Conv2d(in_channels,
+                                 in_channels,
+                                 kernel_size=1,
+                                 stride=1,
+                                 padding=0)
+        self.proj_out = ops.Conv2d(in_channels,
+                                        in_channels,
+                                        kernel_size=1,
+                                        stride=1,
+                                        padding=0)
+
+        if model_management.xformers_enabled_vae():
+            print("Using xformers attention in VAE")
+            self.optimized_attention = xformers_attention
+        elif model_management.pytorch_attention_enabled():
+            print("Using pytorch attention in VAE")
+            self.optimized_attention = pytorch_attention
+        else:
+            print("Using split attention in VAE")
+            self.optimized_attention = normal_attention
+
+    def forward(self, x):
+        h_ = x
+        h_ = self.norm(h_)
+        q = self.q(h_)
+        k = self.k(h_)
+        v = self.v(h_)
+
+        h_ = self.optimized_attention(q, k, v)
+
+        h_ = self.proj_out(h_)
+
+        return x+h_
+
+
+def make_attn(in_channels, attn_type="vanilla", attn_kwargs=None):
+    return AttnBlock(in_channels)
+
+
+class Model(nn.Module):
+    def __init__(self, *, ch, out_ch, ch_mult=(1,2,4,8), num_res_blocks,
+                 attn_resolutions, dropout=0.0, resamp_with_conv=True, in_channels,
+                 resolution, use_timestep=True, use_linear_attn=False, attn_type="vanilla"):
+        super().__init__()
+        if use_linear_attn: attn_type = "linear"
+        self.ch = ch
+        self.temb_ch = self.ch*4
+        self.num_resolutions = len(ch_mult)
+        self.num_res_blocks = num_res_blocks
+        self.resolution = resolution
+        self.in_channels = in_channels
+
+        self.use_timestep = use_timestep
+        if self.use_timestep:
+            # timestep embedding
+            self.temb = nn.Module()
+            self.temb.dense = nn.ModuleList([
+                ops.Linear(self.ch,
+                                self.temb_ch),
+                ops.Linear(self.temb_ch,
+                                self.temb_ch),
+            ])
+
+        # downsampling
+        self.conv_in = ops.Conv2d(in_channels,
+                                       self.ch,
+                                       kernel_size=3,
+                                       stride=1,
+                                       padding=1)
+
+        curr_res = resolution
+        in_ch_mult = (1,)+tuple(ch_mult)
+        self.down = nn.ModuleList()
+        for i_level in range(self.num_resolutions):
+            block = nn.ModuleList()
+            attn = nn.ModuleList()
+            block_in = ch*in_ch_mult[i_level]
+            block_out = ch*ch_mult[i_level]
+            for i_block in range(self.num_res_blocks):
+                block.append(ResnetBlock(in_channels=block_in,
+                                         out_channels=block_out,
+                                         temb_channels=self.temb_ch,
+                                         dropout=dropout))
+                block_in = block_out
+                if curr_res in attn_resolutions:
+                    attn.append(make_attn(block_in, attn_type=attn_type))
+            down = nn.Module()
+            down.block = block
+            down.attn = attn
+            if i_level != self.num_resolutions-1:
+                down.downsample = Downsample(block_in, resamp_with_conv)
+                curr_res = curr_res // 2
+            self.down.append(down)
+
+        # middle
+        self.mid = nn.Module()
+        self.mid.block_1 = ResnetBlock(in_channels=block_in,
+                                       out_channels=block_in,
+                                       temb_channels=self.temb_ch,
+                                       dropout=dropout)
+        self.mid.attn_1 = make_attn(block_in, attn_type=attn_type)
+        self.mid.block_2 = ResnetBlock(in_channels=block_in,
+                                       out_channels=block_in,
+                                       temb_channels=self.temb_ch,
+                                       dropout=dropout)
+
+        # upsampling
+        self.up = nn.ModuleList()
+        for i_level in reversed(range(self.num_resolutions)):
+            block = nn.ModuleList()
+            attn = nn.ModuleList()
+            block_out = ch*ch_mult[i_level]
+            skip_in = ch*ch_mult[i_level]
+            for i_block in range(self.num_res_blocks+1):
+                if i_block == self.num_res_blocks:
+                    skip_in = ch*in_ch_mult[i_level]
+                block.append(ResnetBlock(in_channels=block_in+skip_in,
+                                         out_channels=block_out,
+                                         temb_channels=self.temb_ch,
+                                         dropout=dropout))
+                block_in = block_out
+                if curr_res in attn_resolutions:
+                    attn.append(make_attn(block_in, attn_type=attn_type))
+            up = nn.Module()
+            up.block = block
+            up.attn = attn
+            if i_level != 0:
+                up.upsample = Upsample(block_in, resamp_with_conv)
+                curr_res = curr_res * 2
+            self.up.insert(0, up) # prepend to get consistent order
+
+        # end
+        self.norm_out = Normalize(block_in)
+        self.conv_out = ops.Conv2d(block_in,
+                                        out_ch,
+                                        kernel_size=3,
+                                        stride=1,
+                                        padding=1)
+
+    def forward(self, x, t=None, context=None):
+        #assert x.shape[2] == x.shape[3] == self.resolution
+        if context is not None:
+            # assume aligned context, cat along channel axis
+            x = torch.cat((x, context), dim=1)
+        if self.use_timestep:
+            # timestep embedding
+            assert t is not None
+            temb = get_timestep_embedding(t, self.ch)
+            temb = self.temb.dense[0](temb)
+            temb = nonlinearity(temb)
+            temb = self.temb.dense[1](temb)
+        else:
+            temb = None
+
+        # downsampling
+        hs = [self.conv_in(x)]
+        for i_level in range(self.num_resolutions):
+            for i_block in range(self.num_res_blocks):
+                h = self.down[i_level].block[i_block](hs[-1], temb)
+                if len(self.down[i_level].attn) > 0:
+                    h = self.down[i_level].attn[i_block](h)
+                hs.append(h)
+            if i_level != self.num_resolutions-1:
+                hs.append(self.down[i_level].downsample(hs[-1]))
+
+        # middle
+        h = hs[-1]
+        h = self.mid.block_1(h, temb)
+        h = self.mid.attn_1(h)
+        h = self.mid.block_2(h, temb)
+
+        # upsampling
+        for i_level in reversed(range(self.num_resolutions)):
+            for i_block in range(self.num_res_blocks+1):
+                h = self.up[i_level].block[i_block](
+                    torch.cat([h, hs.pop()], dim=1), temb)
+                if len(self.up[i_level].attn) > 0:
+                    h = self.up[i_level].attn[i_block](h)
+            if i_level != 0:
+                h = self.up[i_level].upsample(h)
+
+        # end
+        h = self.norm_out(h)
+        h = nonlinearity(h)
+        h = self.conv_out(h)
+        return h
+
+    def get_last_layer(self):
+        return self.conv_out.weight
+
+
+class Encoder(nn.Module):
+    def __init__(self, *, ch, out_ch, ch_mult=(1,2,4,8), num_res_blocks,
+                 attn_resolutions, dropout=0.0, resamp_with_conv=True, in_channels,
+                 resolution, z_channels, double_z=True, use_linear_attn=False, attn_type="vanilla",
+                 **ignore_kwargs):
+        super().__init__()
+        if use_linear_attn: attn_type = "linear"
+        self.ch = ch
+        self.temb_ch = 0
+        self.num_resolutions = len(ch_mult)
+        self.num_res_blocks = num_res_blocks
+        self.resolution = resolution
+        self.in_channels = in_channels
+
+        # downsampling
+        self.conv_in = ops.Conv2d(in_channels,
+                                       self.ch,
+                                       kernel_size=3,
+                                       stride=1,
+                                       padding=1)
+
+        curr_res = resolution
+        in_ch_mult = (1,)+tuple(ch_mult)
+        self.in_ch_mult = in_ch_mult
+        self.down = nn.ModuleList()
+        for i_level in range(self.num_resolutions):
+            block = nn.ModuleList()
+            attn = nn.ModuleList()
+            block_in = ch*in_ch_mult[i_level]
+            block_out = ch*ch_mult[i_level]
+            for i_block in range(self.num_res_blocks):
+                block.append(ResnetBlock(in_channels=block_in,
+                                         out_channels=block_out,
+                                         temb_channels=self.temb_ch,
+                                         dropout=dropout))
+                block_in = block_out
+                if curr_res in attn_resolutions:
+                    attn.append(make_attn(block_in, attn_type=attn_type))
+            down = nn.Module()
+            down.block = block
+            down.attn = attn
+            if i_level != self.num_resolutions-1:
+                down.downsample = Downsample(block_in, resamp_with_conv)
+                curr_res = curr_res // 2
+            self.down.append(down)
+
+        # middle
+        self.mid = nn.Module()
+        self.mid.block_1 = ResnetBlock(in_channels=block_in,
+                                       out_channels=block_in,
+                                       temb_channels=self.temb_ch,
+                                       dropout=dropout)
+        self.mid.attn_1 = make_attn(block_in, attn_type=attn_type)
+        self.mid.block_2 = ResnetBlock(in_channels=block_in,
+                                       out_channels=block_in,
+                                       temb_channels=self.temb_ch,
+                                       dropout=dropout)
+
+        # end
+        self.norm_out = Normalize(block_in)
+        self.conv_out = ops.Conv2d(block_in,
+                                        2*z_channels if double_z else z_channels,
+                                        kernel_size=3,
+                                        stride=1,
+                                        padding=1)
+
+    def forward(self, x):
+        # timestep embedding
+        temb = None
+        # downsampling
+        h = self.conv_in(x)
+        for i_level in range(self.num_resolutions):
+            for i_block in range(self.num_res_blocks):
+                h = self.down[i_level].block[i_block](h, temb)
+                if len(self.down[i_level].attn) > 0:
+                    h = self.down[i_level].attn[i_block](h)
+            if i_level != self.num_resolutions-1:
+                h = self.down[i_level].downsample(h)
+
+        # middle
+        h = self.mid.block_1(h, temb)
+        h = self.mid.attn_1(h)
+        h = self.mid.block_2(h, temb)
+
+        # end
+        h = self.norm_out(h)
+        h = nonlinearity(h)
+        h = self.conv_out(h)
+        return h
+
+
+class Decoder(nn.Module):
+    def __init__(self, *, ch, out_ch, ch_mult=(1,2,4,8), num_res_blocks,
+                 attn_resolutions, dropout=0.0, resamp_with_conv=True, in_channels,
+                 resolution, z_channels, give_pre_end=False, tanh_out=False, use_linear_attn=False,
+                 conv_out_op=ops.Conv2d,
+                 resnet_op=ResnetBlock,
+                 attn_op=AttnBlock,
+                **ignorekwargs):
+        super().__init__()
+        if use_linear_attn: attn_type = "linear"
+        self.ch = ch
+        self.temb_ch = 0
+        self.num_resolutions = len(ch_mult)
+        self.num_res_blocks = num_res_blocks
+        self.resolution = resolution
+        self.in_channels = in_channels
+        self.give_pre_end = give_pre_end
+        self.tanh_out = tanh_out
+
+        # compute in_ch_mult, block_in and curr_res at lowest res
+        in_ch_mult = (1,)+tuple(ch_mult)
+        block_in = ch*ch_mult[self.num_resolutions-1]
+        curr_res = resolution // 2**(self.num_resolutions-1)
+        self.z_shape = (1,z_channels,curr_res,curr_res)
+        print("Working with z of shape {} = {} dimensions.".format(
+            self.z_shape, np.prod(self.z_shape)))
+
+        # z to block_in
+        self.conv_in = ops.Conv2d(z_channels,
+                                       block_in,
+                                       kernel_size=3,
+                                       stride=1,
+                                       padding=1)
+
+        # middle
+        self.mid = nn.Module()
+        self.mid.block_1 = resnet_op(in_channels=block_in,
+                                       out_channels=block_in,
+                                       temb_channels=self.temb_ch,
+                                       dropout=dropout)
+        self.mid.attn_1 = attn_op(block_in)
+        self.mid.block_2 = resnet_op(in_channels=block_in,
+                                       out_channels=block_in,
+                                       temb_channels=self.temb_ch,
+                                       dropout=dropout)
+
+        # upsampling
+        self.up = nn.ModuleList()
+        for i_level in reversed(range(self.num_resolutions)):
+            block = nn.ModuleList()
+            attn = nn.ModuleList()
+            block_out = ch*ch_mult[i_level]
+            for i_block in range(self.num_res_blocks+1):
+                block.append(resnet_op(in_channels=block_in,
+                                         out_channels=block_out,
+                                         temb_channels=self.temb_ch,
+                                         dropout=dropout))
+                block_in = block_out
+                if curr_res in attn_resolutions:
+                    attn.append(attn_op(block_in))
+            up = nn.Module()
+            up.block = block
+            up.attn = attn
+            if i_level != 0:
+                up.upsample = Upsample(block_in, resamp_with_conv)
+                curr_res = curr_res * 2
+            self.up.insert(0, up) # prepend to get consistent order
+
+        # end
+        self.norm_out = Normalize(block_in)
+        self.conv_out = conv_out_op(block_in,
+                                        out_ch,
+                                        kernel_size=3,
+                                        stride=1,
+                                        padding=1)
+
+    def forward(self, z, **kwargs):
+        #assert z.shape[1:] == self.z_shape[1:]
+        self.last_z_shape = z.shape
+
+        # timestep embedding
+        temb = None
+
+        # z to block_in
+        h = self.conv_in(z)
+
+        # middle
+        h = self.mid.block_1(h, temb, **kwargs)
+        h = self.mid.attn_1(h, **kwargs)
+        h = self.mid.block_2(h, temb, **kwargs)
+
+        # upsampling
+        for i_level in reversed(range(self.num_resolutions)):
+            for i_block in range(self.num_res_blocks+1):
+                h = self.up[i_level].block[i_block](h, temb, **kwargs)
+                if len(self.up[i_level].attn) > 0:
+                    h = self.up[i_level].attn[i_block](h, **kwargs)
+            if i_level != 0:
+                h = self.up[i_level].upsample(h)
+
+        # end
+        if self.give_pre_end:
+            return h
+
+        h = self.norm_out(h)
+        h = nonlinearity(h)
+        h = self.conv_out(h, **kwargs)
+        if self.tanh_out:
+            h = torch.tanh(h)
+        return h

ldm_patched/ldm/modules/diffusionmodules/openaimodel.py

@@ -0,0 +1,893 @@
+from abc import abstractmethod
+import math
+
+import numpy as np
+import torch as th
+import torch.nn as nn
+import torch.nn.functional as F
+from einops import rearrange
+from functools import partial
+
+from .util import (
+    checkpoint,
+    avg_pool_nd,
+    zero_module,
+    timestep_embedding,
+    AlphaBlender,
+)
+from ..attention import SpatialTransformer, SpatialVideoTransformer, default
+from ldm_patched.ldm.util import exists
+import ldm_patched.modules.ops
+ops = ldm_patched.modules.ops.disable_weight_init
+
+class TimestepBlock(nn.Module):
+    """
+    Any module where forward() takes timestep embeddings as a second argument.
+    """
+
+    @abstractmethod
+    def forward(self, x, emb):
+        """
+        Apply the module to `x` given `emb` timestep embeddings.
+        """
+
+#This is needed because accelerate makes a copy of transformer_options which breaks "transformer_index"
+def forward_timestep_embed(ts, x, emb, context=None, transformer_options={}, output_shape=None, time_context=None, num_video_frames=None, image_only_indicator=None):
+    for layer in ts:
+        if isinstance(layer, VideoResBlock):
+            x = layer(x, emb, num_video_frames, image_only_indicator)
+        elif isinstance(layer, TimestepBlock):
+            x = layer(x, emb)
+        elif isinstance(layer, SpatialVideoTransformer):
+            x = layer(x, context, time_context, num_video_frames, image_only_indicator, transformer_options)
+            if "transformer_index" in transformer_options:
+                transformer_options["transformer_index"] += 1
+        elif isinstance(layer, SpatialTransformer):
+            x = layer(x, context, transformer_options)
+            if "transformer_index" in transformer_options:
+                transformer_options["transformer_index"] += 1
+        elif isinstance(layer, Upsample):
+            x = layer(x, output_shape=output_shape)
+        else:
+            x = layer(x)
+    return x
+
+class TimestepEmbedSequential(nn.Sequential, TimestepBlock):
+    """
+    A sequential module that passes timestep embeddings to the children that
+    support it as an extra input.
+    """
+
+    def forward(self, *args, **kwargs):
+        return forward_timestep_embed(self, *args, **kwargs)
+
+class Upsample(nn.Module):
+    """
+    An upsampling layer with an optional convolution.
+    :param channels: channels in the inputs and outputs.
+    :param use_conv: a bool determining if a convolution is applied.
+    :param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then
+                 upsampling occurs in the inner-two dimensions.
+    """
+
+    def __init__(self, channels, use_conv, dims=2, out_channels=None, padding=1, dtype=None, device=None, operations=ops):
+        super().__init__()
+        self.channels = channels
+        self.out_channels = out_channels or channels
+        self.use_conv = use_conv
+        self.dims = dims
+        if use_conv:
+            self.conv = operations.conv_nd(dims, self.channels, self.out_channels, 3, padding=padding, dtype=dtype, device=device)
+
+    def forward(self, x, output_shape=None):
+        assert x.shape[1] == self.channels
+        if self.dims == 3:
+            shape = [x.shape[2], x.shape[3] * 2, x.shape[4] * 2]
+            if output_shape is not None:
+                shape[1] = output_shape[3]
+                shape[2] = output_shape[4]
+        else:
+            shape = [x.shape[2] * 2, x.shape[3] * 2]
+            if output_shape is not None:
+                shape[0] = output_shape[2]
+                shape[1] = output_shape[3]
+
+        x = F.interpolate(x, size=shape, mode="nearest")
+        if self.use_conv:
+            x = self.conv(x)
+        return x
+
+class Downsample(nn.Module):
+    """
+    A downsampling layer with an optional convolution.
+    :param channels: channels in the inputs and outputs.
+    :param use_conv: a bool determining if a convolution is applied.
+    :param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then
+                 downsampling occurs in the inner-two dimensions.
+    """
+
+    def __init__(self, channels, use_conv, dims=2, out_channels=None, padding=1, dtype=None, device=None, operations=ops):
+        super().__init__()
+        self.channels = channels
+        self.out_channels = out_channels or channels
+        self.use_conv = use_conv
+        self.dims = dims
+        stride = 2 if dims != 3 else (1, 2, 2)
+        if use_conv:
+            self.op = operations.conv_nd(
+                dims, self.channels, self.out_channels, 3, stride=stride, padding=padding, dtype=dtype, device=device
+            )
+        else:
+            assert self.channels == self.out_channels
+            self.op = avg_pool_nd(dims, kernel_size=stride, stride=stride)
+
+    def forward(self, x):
+        assert x.shape[1] == self.channels
+        return self.op(x)
+
+
+class ResBlock(TimestepBlock):
+    """
+    A residual block that can optionally change the number of channels.
+    :param channels: the number of input channels.
+    :param emb_channels: the number of timestep embedding channels.
+    :param dropout: the rate of dropout.
+    :param out_channels: if specified, the number of out channels.
+    :param use_conv: if True and out_channels is specified, use a spatial
+        convolution instead of a smaller 1x1 convolution to change the
+        channels in the skip connection.
+    :param dims: determines if the signal is 1D, 2D, or 3D.
+    :param use_checkpoint: if True, use gradient checkpointing on this module.
+    :param up: if True, use this block for upsampling.
+    :param down: if True, use this block for downsampling.
+    """
+
+    def __init__(
+        self,
+        channels,
+        emb_channels,
+        dropout,
+        out_channels=None,
+        use_conv=False,
+        use_scale_shift_norm=False,
+        dims=2,
+        use_checkpoint=False,
+        up=False,
+        down=False,
+        kernel_size=3,
+        exchange_temb_dims=False,
+        skip_t_emb=False,
+        dtype=None,
+        device=None,
+        operations=ops
+    ):
+        super().__init__()
+        self.channels = channels
+        self.emb_channels = emb_channels
+        self.dropout = dropout
+        self.out_channels = out_channels or channels
+        self.use_conv = use_conv
+        self.use_checkpoint = use_checkpoint
+        self.use_scale_shift_norm = use_scale_shift_norm
+        self.exchange_temb_dims = exchange_temb_dims
+
+        if isinstance(kernel_size, list):
+            padding = [k // 2 for k in kernel_size]
+        else:
+            padding = kernel_size // 2
+
+        self.in_layers = nn.Sequential(
+            operations.GroupNorm(32, channels, dtype=dtype, device=device),
+            nn.SiLU(),
+            operations.conv_nd(dims, channels, self.out_channels, kernel_size, padding=padding, dtype=dtype, device=device),
+        )
+
+        self.updown = up or down
+
+        if up:
+            self.h_upd = Upsample(channels, False, dims, dtype=dtype, device=device)
+            self.x_upd = Upsample(channels, False, dims, dtype=dtype, device=device)
+        elif down:
+            self.h_upd = Downsample(channels, False, dims, dtype=dtype, device=device)
+            self.x_upd = Downsample(channels, False, dims, dtype=dtype, device=device)
+        else:
+            self.h_upd = self.x_upd = nn.Identity()
+
+        self.skip_t_emb = skip_t_emb
+        if self.skip_t_emb:
+            self.emb_layers = None
+            self.exchange_temb_dims = False
+        else:
+            self.emb_layers = nn.Sequential(
+                nn.SiLU(),
+                operations.Linear(
+                    emb_channels,
+                    2 * self.out_channels if use_scale_shift_norm else self.out_channels, dtype=dtype, device=device
+                ),
+            )
+        self.out_layers = nn.Sequential(
+            operations.GroupNorm(32, self.out_channels, dtype=dtype, device=device),
+            nn.SiLU(),
+            nn.Dropout(p=dropout),
+            operations.conv_nd(dims, self.out_channels, self.out_channels, kernel_size, padding=padding, dtype=dtype, device=device)
+            ,
+        )
+
+        if self.out_channels == channels:
+            self.skip_connection = nn.Identity()
+        elif use_conv:
+            self.skip_connection = operations.conv_nd(
+                dims, channels, self.out_channels, kernel_size, padding=padding, dtype=dtype, device=device
+            )
+        else:
+            self.skip_connection = operations.conv_nd(dims, channels, self.out_channels, 1, dtype=dtype, device=device)
+
+    def forward(self, x, emb):
+        """
+        Apply the block to a Tensor, conditioned on a timestep embedding.
+        :param x: an [N x C x ...] Tensor of features.
+        :param emb: an [N x emb_channels] Tensor of timestep embeddings.
+        :return: an [N x C x ...] Tensor of outputs.
+        """
+        return checkpoint(
+            self._forward, (x, emb), self.parameters(), self.use_checkpoint
+        )
+
+
+    def _forward(self, x, emb):
+        if self.updown:
+            in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1]
+            h = in_rest(x)
+            h = self.h_upd(h)
+            x = self.x_upd(x)
+            h = in_conv(h)
+        else:
+            h = self.in_layers(x)
+
+        emb_out = None
+        if not self.skip_t_emb:
+            emb_out = self.emb_layers(emb).type(h.dtype)
+            while len(emb_out.shape) < len(h.shape):
+                emb_out = emb_out[..., None]
+        if self.use_scale_shift_norm:
+            out_norm, out_rest = self.out_layers[0], self.out_layers[1:]
+            h = out_norm(h)
+            if emb_out is not None:
+                scale, shift = th.chunk(emb_out, 2, dim=1)
+                h *= (1 + scale)
+                h += shift
+            h = out_rest(h)
+        else:
+            if emb_out is not None:
+                if self.exchange_temb_dims:
+                    emb_out = rearrange(emb_out, "b t c ... -> b c t ...")
+                h = h + emb_out
+            h = self.out_layers(h)
+        return self.skip_connection(x) + h
+
+
+class VideoResBlock(ResBlock):
+    def __init__(
+        self,
+        channels: int,
+        emb_channels: int,
+        dropout: float,
+        video_kernel_size=3,
+        merge_strategy: str = "fixed",
+        merge_factor: float = 0.5,
+        out_channels=None,
+        use_conv: bool = False,
+        use_scale_shift_norm: bool = False,
+        dims: int = 2,
+        use_checkpoint: bool = False,
+        up: bool = False,
+        down: bool = False,
+        dtype=None,
+        device=None,
+        operations=ops
+    ):
+        super().__init__(
+            channels,
+            emb_channels,
+            dropout,
+            out_channels=out_channels,
+            use_conv=use_conv,
+            use_scale_shift_norm=use_scale_shift_norm,
+            dims=dims,
+            use_checkpoint=use_checkpoint,
+            up=up,
+            down=down,
+            dtype=dtype,
+            device=device,
+            operations=operations
+        )
+
+        self.time_stack = ResBlock(
+            default(out_channels, channels),
+            emb_channels,
+            dropout=dropout,
+            dims=3,
+            out_channels=default(out_channels, channels),
+            use_scale_shift_norm=False,
+            use_conv=False,
+            up=False,
+            down=False,
+            kernel_size=video_kernel_size,
+            use_checkpoint=use_checkpoint,
+            exchange_temb_dims=True,
+            dtype=dtype,
+            device=device,
+            operations=operations
+        )
+        self.time_mixer = AlphaBlender(
+            alpha=merge_factor,
+            merge_strategy=merge_strategy,
+            rearrange_pattern="b t -> b 1 t 1 1",
+        )
+
+    def forward(
+        self,
+        x: th.Tensor,
+        emb: th.Tensor,
+        num_video_frames: int,
+        image_only_indicator = None,
+    ) -> th.Tensor:
+        x = super().forward(x, emb)
+
+        x_mix = rearrange(x, "(b t) c h w -> b c t h w", t=num_video_frames)
+        x = rearrange(x, "(b t) c h w -> b c t h w", t=num_video_frames)
+
+        x = self.time_stack(
+            x, rearrange(emb, "(b t) ... -> b t ...", t=num_video_frames)
+        )
+        x = self.time_mixer(
+            x_spatial=x_mix, x_temporal=x, image_only_indicator=image_only_indicator
+        )
+        x = rearrange(x, "b c t h w -> (b t) c h w")
+        return x
+
+
+class Timestep(nn.Module):
+    def __init__(self, dim):
+        super().__init__()
+        self.dim = dim
+
+    def forward(self, t):
+        return timestep_embedding(t, self.dim)
+
+def apply_control(h, control, name):
+    if control is not None and name in control and len(control[name]) > 0:
+        ctrl = control[name].pop()
+        if ctrl is not None:
+            try:
+                h += ctrl
+            except:
+                print("warning control could not be applied", h.shape, ctrl.shape)
+    return h
+
+class UNetModel(nn.Module):
+    """
+    The full UNet model with attention and timestep embedding.
+    :param in_channels: channels in the input Tensor.
+    :param model_channels: base channel count for the model.
+    :param out_channels: channels in the output Tensor.
+    :param num_res_blocks: number of residual blocks per downsample.
+    :param dropout: the dropout probability.
+    :param channel_mult: channel multiplier for each level of the UNet.
+    :param conv_resample: if True, use learned convolutions for upsampling and
+        downsampling.
+    :param dims: determines if the signal is 1D, 2D, or 3D.
+    :param num_classes: if specified (as an int), then this model will be
+        class-conditional with `num_classes` classes.
+    :param use_checkpoint: use gradient checkpointing to reduce memory usage.
+    :param num_heads: the number of attention heads in each attention layer.
+    :param num_heads_channels: if specified, ignore num_heads and instead use
+                               a fixed channel width per attention head.
+    :param num_heads_upsample: works with num_heads to set a different number
+                               of heads for upsampling. Deprecated.
+    :param use_scale_shift_norm: use a FiLM-like conditioning mechanism.
+    :param resblock_updown: use residual blocks for up/downsampling.
+    :param use_new_attention_order: use a different attention pattern for potentially
+                                    increased efficiency.
+    """
+
+    def __init__(
+        self,
+        image_size,
+        in_channels,
+        model_channels,
+        out_channels,
+        num_res_blocks,
+        dropout=0,
+        channel_mult=(1, 2, 4, 8),
+        conv_resample=True,
+        dims=2,
+        num_classes=None,
+        use_checkpoint=False,
+        dtype=th.float32,
+        num_heads=-1,
+        num_head_channels=-1,
+        num_heads_upsample=-1,
+        use_scale_shift_norm=False,
+        resblock_updown=False,
+        use_new_attention_order=False,
+        use_spatial_transformer=False,    # custom transformer support
+        transformer_depth=1,              # custom transformer support
+        context_dim=None,                 # custom transformer support
+        n_embed=None,                     # custom support for prediction of discrete ids into codebook of first stage vq model
+        legacy=True,
+        disable_self_attentions=None,
+        num_attention_blocks=None,
+        disable_middle_self_attn=False,
+        use_linear_in_transformer=False,
+        adm_in_channels=None,
+        transformer_depth_middle=None,
+        transformer_depth_output=None,
+        use_temporal_resblock=False,
+        use_temporal_attention=False,
+        time_context_dim=None,
+        extra_ff_mix_layer=False,
+        use_spatial_context=False,
+        merge_strategy=None,
+        merge_factor=0.0,
+        video_kernel_size=None,
+        disable_temporal_crossattention=False,
+        max_ddpm_temb_period=10000,
+        device=None,
+        operations=ops,
+    ):
+        super().__init__()
+        assert use_spatial_transformer == True, "use_spatial_transformer has to be true"
+        if use_spatial_transformer:
+            assert context_dim is not None, 'Fool!! You forgot to include the dimension of your cross-attention conditioning...'
+
+        if context_dim is not None:
+            assert use_spatial_transformer, 'Fool!! You forgot to use the spatial transformer for your cross-attention conditioning...'
+            # from omegaconf.listconfig import ListConfig
+            # if type(context_dim) == ListConfig:
+            #     context_dim = list(context_dim)
+
+        if num_heads_upsample == -1:
+            num_heads_upsample = num_heads
+
+        if num_heads == -1:
+            assert num_head_channels != -1, 'Either num_heads or num_head_channels has to be set'
+
+        if num_head_channels == -1:
+            assert num_heads != -1, 'Either num_heads or num_head_channels has to be set'
+
+        self.image_size = image_size
+        self.in_channels = in_channels
+        self.model_channels = model_channels
+        self.out_channels = out_channels
+
+        if isinstance(num_res_blocks, int):
+            self.num_res_blocks = len(channel_mult) * [num_res_blocks]
+        else:
+            if len(num_res_blocks) != len(channel_mult):
+                raise ValueError("provide num_res_blocks either as an int (globally constant) or "
+                                 "as a list/tuple (per-level) with the same length as channel_mult")
+            self.num_res_blocks = num_res_blocks
+
+        if disable_self_attentions is not None:
+            # should be a list of booleans, indicating whether to disable self-attention in TransformerBlocks or not
+            assert len(disable_self_attentions) == len(channel_mult)
+        if num_attention_blocks is not None:
+            assert len(num_attention_blocks) == len(self.num_res_blocks)
+
+        transformer_depth = transformer_depth[:]
+        transformer_depth_output = transformer_depth_output[:]
+
+        self.dropout = dropout
+        self.channel_mult = channel_mult
+        self.conv_resample = conv_resample
+        self.num_classes = num_classes
+        self.use_checkpoint = use_checkpoint
+        self.dtype = dtype
+        self.num_heads = num_heads
+        self.num_head_channels = num_head_channels
+        self.num_heads_upsample = num_heads_upsample
+        self.use_temporal_resblocks = use_temporal_resblock
+        self.predict_codebook_ids = n_embed is not None
+
+        self.default_num_video_frames = None
+        self.default_image_only_indicator = None
+
+        time_embed_dim = model_channels * 4
+        self.time_embed = nn.Sequential(
+            operations.Linear(model_channels, time_embed_dim, dtype=self.dtype, device=device),
+            nn.SiLU(),
+            operations.Linear(time_embed_dim, time_embed_dim, dtype=self.dtype, device=device),
+        )
+
+        if self.num_classes is not None:
+            if isinstance(self.num_classes, int):
+                self.label_emb = nn.Embedding(num_classes, time_embed_dim)
+            elif self.num_classes == "continuous":
+                print("setting up linear c_adm embedding layer")
+                self.label_emb = nn.Linear(1, time_embed_dim)
+            elif self.num_classes == "sequential":
+                assert adm_in_channels is not None
+                self.label_emb = nn.Sequential(
+                    nn.Sequential(
+                        operations.Linear(adm_in_channels, time_embed_dim, dtype=self.dtype, device=device),
+                        nn.SiLU(),
+                        operations.Linear(time_embed_dim, time_embed_dim, dtype=self.dtype, device=device),
+                    )
+                )
+            else:
+                raise ValueError()
+
+        self.input_blocks = nn.ModuleList(
+            [
+                TimestepEmbedSequential(
+                    operations.conv_nd(dims, in_channels, model_channels, 3, padding=1, dtype=self.dtype, device=device)
+                )
+            ]
+        )
+        self._feature_size = model_channels
+        input_block_chans = [model_channels]
+        ch = model_channels
+        ds = 1
+
+        def get_attention_layer(
+            ch,
+            num_heads,
+            dim_head,
+            depth=1,
+            context_dim=None,
+            use_checkpoint=False,
+            disable_self_attn=False,
+        ):
+            if use_temporal_attention:
+                return SpatialVideoTransformer(
+                    ch,
+                    num_heads,
+                    dim_head,
+                    depth=depth,
+                    context_dim=context_dim,
+                    time_context_dim=time_context_dim,
+                    dropout=dropout,
+                    ff_in=extra_ff_mix_layer,
+                    use_spatial_context=use_spatial_context,
+                    merge_strategy=merge_strategy,
+                    merge_factor=merge_factor,
+                    checkpoint=use_checkpoint,
+                    use_linear=use_linear_in_transformer,
+                    disable_self_attn=disable_self_attn,
+                    disable_temporal_crossattention=disable_temporal_crossattention,
+                    max_time_embed_period=max_ddpm_temb_period,
+                    dtype=self.dtype, device=device, operations=operations
+                )
+            else:
+                return SpatialTransformer(
+                                ch, num_heads, dim_head, depth=depth, context_dim=context_dim,
+                                disable_self_attn=disable_self_attn, use_linear=use_linear_in_transformer,
+                                use_checkpoint=use_checkpoint, dtype=self.dtype, device=device, operations=operations
+                            )
+
+        def get_resblock(
+            merge_factor,
+            merge_strategy,
+            video_kernel_size,
+            ch,
+            time_embed_dim,
+            dropout,
+            out_channels,
+            dims,
+            use_checkpoint,
+            use_scale_shift_norm,
+            down=False,
+            up=False,
+            dtype=None,
+            device=None,
+            operations=ops
+        ):
+            if self.use_temporal_resblocks:
+                return VideoResBlock(
+                    merge_factor=merge_factor,
+                    merge_strategy=merge_strategy,
+                    video_kernel_size=video_kernel_size,
+                    channels=ch,
+                    emb_channels=time_embed_dim,
+                    dropout=dropout,
+                    out_channels=out_channels,
+                    dims=dims,
+                    use_checkpoint=use_checkpoint,
+                    use_scale_shift_norm=use_scale_shift_norm,
+                    down=down,
+                    up=up,
+                    dtype=dtype,
+                    device=device,
+                    operations=operations
+                )
+            else:
+                return ResBlock(
+                    channels=ch,
+                    emb_channels=time_embed_dim,
+                    dropout=dropout,
+                    out_channels=out_channels,
+                    use_checkpoint=use_checkpoint,
+                    dims=dims,
+                    use_scale_shift_norm=use_scale_shift_norm,
+                    down=down,
+                    up=up,
+                    dtype=dtype,
+                    device=device,
+                    operations=operations
+                )
+
+        for level, mult in enumerate(channel_mult):
+            for nr in range(self.num_res_blocks[level]):
+                layers = [
+                    get_resblock(
+                        merge_factor=merge_factor,
+                        merge_strategy=merge_strategy,
+                        video_kernel_size=video_kernel_size,
+                        ch=ch,
+                        time_embed_dim=time_embed_dim,
+                        dropout=dropout,
+                        out_channels=mult * model_channels,
+                        dims=dims,
+                        use_checkpoint=use_checkpoint,
+                        use_scale_shift_norm=use_scale_shift_norm,
+                        dtype=self.dtype,
+                        device=device,
+                        operations=operations,
+                    )
+                ]
+                ch = mult * model_channels
+                num_transformers = transformer_depth.pop(0)
+                if num_transformers > 0:
+                    if num_head_channels == -1:
+                        dim_head = ch // num_heads
+                    else:
+                        num_heads = ch // num_head_channels
+                        dim_head = num_head_channels
+                    if legacy:
+                        #num_heads = 1
+                        dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
+                    if exists(disable_self_attentions):
+                        disabled_sa = disable_self_attentions[level]
+                    else:
+                        disabled_sa = False
+
+                    if not exists(num_attention_blocks) or nr < num_attention_blocks[level]:
+                        layers.append(get_attention_layer(
+                                ch, num_heads, dim_head, depth=num_transformers, context_dim=context_dim,
+                                disable_self_attn=disabled_sa, use_checkpoint=use_checkpoint)
+                        )
+                self.input_blocks.append(TimestepEmbedSequential(*layers))
+                self._feature_size += ch
+                input_block_chans.append(ch)
+            if level != len(channel_mult) - 1:
+                out_ch = ch
+                self.input_blocks.append(
+                    TimestepEmbedSequential(
+                        get_resblock(
+                            merge_factor=merge_factor,
+                            merge_strategy=merge_strategy,
+                            video_kernel_size=video_kernel_size,
+                            ch=ch,
+                            time_embed_dim=time_embed_dim,
+                            dropout=dropout,
+                            out_channels=out_ch,
+                            dims=dims,
+                            use_checkpoint=use_checkpoint,
+                            use_scale_shift_norm=use_scale_shift_norm,
+                            down=True,
+                            dtype=self.dtype,
+                            device=device,
+                            operations=operations
+                        )
+                        if resblock_updown
+                        else Downsample(
+                            ch, conv_resample, dims=dims, out_channels=out_ch, dtype=self.dtype, device=device, operations=operations
+                        )
+                    )
+                )
+                ch = out_ch
+                input_block_chans.append(ch)
+                ds *= 2
+                self._feature_size += ch
+
+        if num_head_channels == -1:
+            dim_head = ch // num_heads
+        else:
+            num_heads = ch // num_head_channels
+            dim_head = num_head_channels
+        if legacy:
+            #num_heads = 1
+            dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
+        mid_block = [
+            get_resblock(
+                merge_factor=merge_factor,
+                merge_strategy=merge_strategy,
+                video_kernel_size=video_kernel_size,
+                ch=ch,
+                time_embed_dim=time_embed_dim,
+                dropout=dropout,
+                out_channels=None,
+                dims=dims,
+                use_checkpoint=use_checkpoint,
+                use_scale_shift_norm=use_scale_shift_norm,
+                dtype=self.dtype,
+                device=device,
+                operations=operations
+            )]
+        if transformer_depth_middle >= 0:
+            mid_block += [get_attention_layer(  # always uses a self-attn
+                            ch, num_heads, dim_head, depth=transformer_depth_middle, context_dim=context_dim,
+                            disable_self_attn=disable_middle_self_attn, use_checkpoint=use_checkpoint
+                        ),
+            get_resblock(
+                merge_factor=merge_factor,
+                merge_strategy=merge_strategy,
+                video_kernel_size=video_kernel_size,
+                ch=ch,
+                time_embed_dim=time_embed_dim,
+                dropout=dropout,
+                out_channels=None,
+                dims=dims,
+                use_checkpoint=use_checkpoint,
+                use_scale_shift_norm=use_scale_shift_norm,
+                dtype=self.dtype,
+                device=device,
+                operations=operations
+            )]
+        self.middle_block = TimestepEmbedSequential(*mid_block)
+        self._feature_size += ch
+
+        self.output_blocks = nn.ModuleList([])
+        for level, mult in list(enumerate(channel_mult))[::-1]:
+            for i in range(self.num_res_blocks[level] + 1):
+                ich = input_block_chans.pop()
+                layers = [
+                    get_resblock(
+                        merge_factor=merge_factor,
+                        merge_strategy=merge_strategy,
+                        video_kernel_size=video_kernel_size,
+                        ch=ch + ich,
+                        time_embed_dim=time_embed_dim,
+                        dropout=dropout,
+                        out_channels=model_channels * mult,
+                        dims=dims,
+                        use_checkpoint=use_checkpoint,
+                        use_scale_shift_norm=use_scale_shift_norm,
+                        dtype=self.dtype,
+                        device=device,
+                        operations=operations
+                    )
+                ]
+                ch = model_channels * mult
+                num_transformers = transformer_depth_output.pop()
+                if num_transformers > 0:
+                    if num_head_channels == -1:
+                        dim_head = ch // num_heads
+                    else:
+                        num_heads = ch // num_head_channels
+                        dim_head = num_head_channels
+                    if legacy:
+                        #num_heads = 1
+                        dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
+                    if exists(disable_self_attentions):
+                        disabled_sa = disable_self_attentions[level]
+                    else:
+                        disabled_sa = False
+
+                    if not exists(num_attention_blocks) or i < num_attention_blocks[level]:
+                        layers.append(
+                            get_attention_layer(
+                                ch, num_heads, dim_head, depth=num_transformers, context_dim=context_dim,
+                                disable_self_attn=disabled_sa, use_checkpoint=use_checkpoint
+                            )
+                        )
+                if level and i == self.num_res_blocks[level]:
+                    out_ch = ch
+                    layers.append(
+                        get_resblock(
+                            merge_factor=merge_factor,
+                            merge_strategy=merge_strategy,
+                            video_kernel_size=video_kernel_size,
+                            ch=ch,
+                            time_embed_dim=time_embed_dim,
+                            dropout=dropout,
+                            out_channels=out_ch,
+                            dims=dims,
+                            use_checkpoint=use_checkpoint,
+                            use_scale_shift_norm=use_scale_shift_norm,
+                            up=True,
+                            dtype=self.dtype,
+                            device=device,
+                            operations=operations
+                        )
+                        if resblock_updown
+                        else Upsample(ch, conv_resample, dims=dims, out_channels=out_ch, dtype=self.dtype, device=device, operations=operations)
+                    )
+                    ds //= 2
+                self.output_blocks.append(TimestepEmbedSequential(*layers))
+                self._feature_size += ch
+
+        self.out = nn.Sequential(
+            operations.GroupNorm(32, ch, dtype=self.dtype, device=device),
+            nn.SiLU(),
+            zero_module(operations.conv_nd(dims, model_channels, out_channels, 3, padding=1, dtype=self.dtype, device=device)),
+        )
+        if self.predict_codebook_ids:
+            self.id_predictor = nn.Sequential(
+            operations.GroupNorm(32, ch, dtype=self.dtype, device=device),
+            operations.conv_nd(dims, model_channels, n_embed, 1, dtype=self.dtype, device=device),
+            #nn.LogSoftmax(dim=1)  # change to cross_entropy and produce non-normalized logits
+        )
+
+    def forward(self, x, timesteps=None, context=None, y=None, control=None, transformer_options={}, **kwargs):
+        """
+        Apply the model to an input batch.
+        :param x: an [N x C x ...] Tensor of inputs.
+        :param timesteps: a 1-D batch of timesteps.
+        :param context: conditioning plugged in via crossattn
+        :param y: an [N] Tensor of labels, if class-conditional.
+        :return: an [N x C x ...] Tensor of outputs.
+        """
+        transformer_options["original_shape"] = list(x.shape)
+        transformer_options["transformer_index"] = 0
+        transformer_patches = transformer_options.get("patches", {})
+
+        num_video_frames = kwargs.get("num_video_frames", self.default_num_video_frames)
+        image_only_indicator = kwargs.get("image_only_indicator", self.default_image_only_indicator)
+        time_context = kwargs.get("time_context", None)
+
+        assert (y is not None) == (
+            self.num_classes is not None
+        ), "must specify y if and only if the model is class-conditional"
+        hs = []
+        t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False).to(x.dtype)
+        emb = self.time_embed(t_emb)
+
+        if self.num_classes is not None:
+            assert y.shape[0] == x.shape[0]
+            emb = emb + self.label_emb(y)
+
+        h = x
+        for id, module in enumerate(self.input_blocks):
+            transformer_options["block"] = ("input", id)
+            h = forward_timestep_embed(module, h, emb, context, transformer_options, time_context=time_context, num_video_frames=num_video_frames, image_only_indicator=image_only_indicator)
+            h = apply_control(h, control, 'input')
+            if "input_block_patch" in transformer_patches:
+                patch = transformer_patches["input_block_patch"]
+                for p in patch:
+                    h = p(h, transformer_options)
+
+            hs.append(h)
+            if "input_block_patch_after_skip" in transformer_patches:
+                patch = transformer_patches["input_block_patch_after_skip"]
+                for p in patch:
+                    h = p(h, transformer_options)
+
+        transformer_options["block"] = ("middle", 0)
+        h = forward_timestep_embed(self.middle_block, h, emb, context, transformer_options, time_context=time_context, num_video_frames=num_video_frames, image_only_indicator=image_only_indicator)
+        h = apply_control(h, control, 'middle')
+
+
+        for id, module in enumerate(self.output_blocks):
+            transformer_options["block"] = ("output", id)
+            hsp = hs.pop()
+            hsp = apply_control(hsp, control, 'output')
+
+            if "output_block_patch" in transformer_patches:
+                patch = transformer_patches["output_block_patch"]
+                for p in patch:
+                    h, hsp = p(h, hsp, transformer_options)
+
+            h = th.cat([h, hsp], dim=1)
+            del hsp
+            if len(hs) > 0:
+                output_shape = hs[-1].shape
+            else:
+                output_shape = None
+            h = forward_timestep_embed(module, h, emb, context, transformer_options, output_shape, time_context=time_context, num_video_frames=num_video_frames, image_only_indicator=image_only_indicator)
+        h = h.type(x.dtype)
+        if self.predict_codebook_ids:
+            return self.id_predictor(h)
+        else:
+            return self.out(h)

ldm_patched/ldm/modules/diffusionmodules/upscaling.py

@@ -0,0 +1,81 @@
+import torch
+import torch.nn as nn
+import numpy as np
+from functools import partial
+
+from .util import extract_into_tensor, make_beta_schedule
+from ldm_patched.ldm.util import default
+
+
+class AbstractLowScaleModel(nn.Module):
+    # for concatenating a downsampled image to the latent representation
+    def __init__(self, noise_schedule_config=None):
+        super(AbstractLowScaleModel, self).__init__()
+        if noise_schedule_config is not None:
+            self.register_schedule(**noise_schedule_config)
+
+    def register_schedule(self, beta_schedule="linear", timesteps=1000,
+                          linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3):
+        betas = make_beta_schedule(beta_schedule, timesteps, linear_start=linear_start, linear_end=linear_end,
+                                   cosine_s=cosine_s)
+        alphas = 1. - betas
+        alphas_cumprod = np.cumprod(alphas, axis=0)
+        alphas_cumprod_prev = np.append(1., alphas_cumprod[:-1])
+
+        timesteps, = betas.shape
+        self.num_timesteps = int(timesteps)
+        self.linear_start = linear_start
+        self.linear_end = linear_end
+        assert alphas_cumprod.shape[0] == self.num_timesteps, 'alphas have to be defined for each timestep'
+
+        to_torch = partial(torch.tensor, dtype=torch.float32)
+
+        self.register_buffer('betas', to_torch(betas))
+        self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod))
+        self.register_buffer('alphas_cumprod_prev', to_torch(alphas_cumprod_prev))
+
+        # calculations for diffusion q(x_t | x_{t-1}) and others
+        self.register_buffer('sqrt_alphas_cumprod', to_torch(np.sqrt(alphas_cumprod)))
+        self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod)))
+        self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod)))
+        self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod)))
+        self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod - 1)))
+
+    def q_sample(self, x_start, t, noise=None):
+        noise = default(noise, lambda: torch.randn_like(x_start))
+        return (extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start +
+                extract_into_tensor(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape) * noise)
+
+    def forward(self, x):
+        return x, None
+
+    def decode(self, x):
+        return x
+
+
+class SimpleImageConcat(AbstractLowScaleModel):
+    # no noise level conditioning
+    def __init__(self):
+        super(SimpleImageConcat, self).__init__(noise_schedule_config=None)
+        self.max_noise_level = 0
+
+    def forward(self, x):
+        # fix to constant noise level
+        return x, torch.zeros(x.shape[0], device=x.device).long()
+
+
+class ImageConcatWithNoiseAugmentation(AbstractLowScaleModel):
+    def __init__(self, noise_schedule_config, max_noise_level=1000, to_cuda=False):
+        super().__init__(noise_schedule_config=noise_schedule_config)
+        self.max_noise_level = max_noise_level
+
+    def forward(self, x, noise_level=None):
+        if noise_level is None:
+            noise_level = torch.randint(0, self.max_noise_level, (x.shape[0],), device=x.device).long()
+        else:
+            assert isinstance(noise_level, torch.Tensor)
+        z = self.q_sample(x, noise_level)
+        return z, noise_level
+
+
+