Update app.py

app.py

@@ -71,14 +71,14 @@ def safe_load_pretrained_model(model_path, model_base=None, model_name=None, **k
     #         tokenizer.pad_token_id = 0
     #         print('[INFO] Set pad_token_id to 0 (default)')
 
-    # # Also ensure pad_token is set (Llama 3 specific) - CRITICAL for output
-    # if tokenizer.pad_token is None:
-    #     if tokenizer.eos_token is not None:
-    #         tokenizer.pad_token = tokenizer.eos_token
-    #         print(f'[INFO] Set pad_token to eos_token: {tokenizer.eos_token}')
-    #     else:
-    #         tokenizer.add_special_tokens({'pad_token': '[PAD]'})
-    #         print('[INFO] Added [PAD] token to tokenizer')
+    # Also ensure pad_token is set (Llama 3 specific) - CRITICAL for output
+    if tokenizer.pad_token is None:
+        if tokenizer.eos_token is not None:
+            tokenizer.pad_token = 128001
+            print(f'[INFO] Set pad_token to eos_token: {tokenizer.pad_token}')
+        else:
+            tokenizer.add_special_tokens({'pad_token': '[PAD]'})
+            print('[INFO] Added [PAD] token to tokenizer')
 
     # Set padding_side to left for Llama 3 (prevents empty generation)
     if hasattr(tokenizer, 'padding_side'):
@@ -130,7 +130,7 @@ def safe_load_pretrained_model(model_path, model_base=None, model_name=None, **k
 
     return tokenizer, model, image_processor, context_len
 
-# 将 builder 的加载函数替换为安全版
+
 if _original_load_pretrained_model is not None:
     builder.load_pretrained_model = safe_load_pretrained_model