fix vllm eos_token (#973)

Author: Jintao-Huang

swift/llm/deploy.py

@@ -7,6 +7,7 @@
 from typing import List, Optional, Union
 
 import json
+import torch
 from fastapi import FastAPI, Request
 from fastapi.responses import JSONResponse, StreamingResponse
 from modelscope import GenerationConfig

swift/llm/utils/vllm_utils.py

@@ -134,6 +134,7 @@ def __init__(
         n: int = 1,
         length_penalty: float = 1.,
         stop: Optional[List[str]] = None,
+        skip_special_tokens: bool = False,
         **kwargs,
     ) -> None:
         # The parameter design is similar to transformers.GenerationConfig.
@@ -162,6 +163,7 @@ def __init__(
         kwargs['n'] = n
         kwargs['length_penalty'] = length_penalty
         kwargs['stop'] = stop
+        kwargs['skip_special_tokens'] = skip_special_tokens
         parameters = inspect.signature(SamplingParams.__init__).parameters
         for k in kwargs.copy().keys():
             if k not in parameters: