[inference] fix stop token for object detection (#6624)

* fix stop token

* update minicpm data pipeline

* fix npu qlora examples

Former-commit-id: 844919fadaa8a61dfae47020971ea80730b2346f

src/llamafactory/chat/hf_engine.py

@@ -133,7 +133,7 @@ class HuggingfaceEngine(BaseEngine):
                 if repetition_penalty is not None
                 else generating_args["repetition_penalty"],
                 length_penalty=length_penalty if length_penalty is not None else generating_args["length_penalty"],
-                eos_token_id=[tokenizer.eos_token_id] + tokenizer.additional_special_tokens_ids,
+                eos_token_id=template.get_stop_token_ids(tokenizer),
                 pad_token_id=tokenizer.pad_token_id,
             )
         )

src/llamafactory/chat/vllm_engine.py

@@ -168,7 +168,7 @@ class VllmEngine(BaseEngine):
             top_p=(top_p if top_p is not None else self.generating_args["top_p"]) or 1.0,  # top_p must > 0
             top_k=top_k if top_k is not None else self.generating_args["top_k"],
             stop=stop,
-            stop_token_ids=[self.tokenizer.eos_token_id] + self.tokenizer.additional_special_tokens_ids,
+            stop_token_ids=self.template.get_stop_token_ids(self.tokenizer),
             max_tokens=max_tokens,
             skip_special_tokens=self.generating_args["skip_special_tokens"],
         )