update

1. gradio_demo.py 262行：model.generate(**kwargs) → model.generate(**kwargs, pad_token_id=tokenizer.eos_token_id)
2. gradio_demo.py 88行：tokenizer = LlamaTokenizer.from_pretrained(args.tokenizer_path) → tokenizer = LlamaTokenizer.from_pretrained(args.tokenizer_path,legacy=False)

gradio_demo.py

@@ -85,7 +85,7 @@ def setup():
         args.tokenizer_path = args.lora_model
         if args.lora_model is None:
             args.tokenizer_path = args.base_model
-    tokenizer = LlamaTokenizer.from_pretrained(args.tokenizer_path)
+    tokenizer = LlamaTokenizer.from_pretrained(args.tokenizer_path,legacy=False)
 
     base_model = LlamaForCausalLM.from_pretrained(
         args.base_model,
@@ -259,7 +259,7 @@ def predict(
             kwargs['stopping_criteria'] = [Stream(callback_func=callback)]
         clear_torch_cache()
         with torch.no_grad():
-            model.generate(**kwargs)
+            model.generate(**kwargs, pad_token_id=tokenizer.eos_token_id)
 
     def generate_with_streaming(**kwargs):
         return Iteratorize(generate_with_callback, kwargs, callback=None)