improve gpt2

improve gpt2

fooocus_version.py

@@ -1 +1 @@
-version = '2.1.742'
+version = '2.1.743'

modules/expansion.py

@@ -1,7 +1,7 @@
 import torch
-
 import fcbh.model_management as model_management
 
+from transformers.generation.logits_process import LogitsProcessorList
 from transformers import AutoTokenizer, AutoModelForCausalLM, set_seed
 from modules.path import fooocus_expansion_path
 from fcbh.model_patcher import ModelPatcher
@@ -16,6 +16,14 @@ fooocus_magic_split = [
 ]
 dangrous_patterns = '[]【】()（）|:：'
 
+black_list = ['art', 'digital', 'Ġpaint', 'painting', 'drawing', 'draw', 'drawn',
+              'concept', 'illustration', 'illustrated', 'illustrate',
+              'face', 'eye', 'eyes', 'hand', 'hands',
+              'monster', 'artistic', 'oil', 'brush',
+              'artwork', 'artworks']
+
+black_list += ['Ġ' + k for k in black_list]
+
 
 def safe_str(x):
     x = str(x)
@@ -33,6 +41,15 @@ def remove_pattern(x, pattern):
 class FooocusExpansion:
     def __init__(self):
         self.tokenizer = AutoTokenizer.from_pretrained(fooocus_expansion_path)
+        self.vocab = self.tokenizer.vocab
+        self.logits_bias = torch.zeros((1, len(self.vocab)), dtype=torch.float32)
+        self.logits_bias[0, self.tokenizer.eos_token_id] = - 16.0
+        # test_198 = self.tokenizer('\n', return_tensors="pt")
+        self.logits_bias[0, 198] = - 1024.0
+        for k, v in self.vocab.items():
+            if k in black_list:
+                self.logits_bias[0, v] = - 1024.0
+
         self.model = AutoModelForCausalLM.from_pretrained(fooocus_expansion_path)
         self.model.eval()
 
@@ -52,6 +69,10 @@ class FooocusExpansion:
         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}.')
 
+    def logits_processor(self, input_ids, scores):
+        self.logits_bias = self.logits_bias.to(scores)
+        return scores + self.logits_bias
+
     def __call__(self, prompt, seed):
         if self.patcher.current_device != self.patcher.load_device:
             print('Fooocus Expansion loaded by itself.')
@@ -66,12 +87,18 @@ class FooocusExpansion:
         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 = 77 + 77 * int(float(current_token_length) / 77.0)
+
+        logits_processor = LogitsProcessorList([self.logits_processor])
+
         # https://huggingface.co/blog/introducing-csearch
         # https://huggingface.co/docs/transformers/generation_strategies
         features = self.model.generate(**tokenized_kwargs,
                                        num_beams=1,
-                                       max_new_tokens=256,
+                                       max_new_tokens=max_token_length - current_token_length,
-                                       do_sample=True)
+                                       do_sample=True,
+                                       logits_processor=logits_processor)
 
         response = self.tokenizer.batch_decode(features, skip_special_tokens=True)
         result = response[0][len(origin):]

update_log.md

@@ -1,3 +1,7 @@
+# 2.1.743
+
+* Improved GPT2 by removing some tokens that may corrupt styles.
+
 # 2.1.741
 
 Style Updates: