Merge branch 'main' into trl-dev

swift/llm/argument/infer_args.py

@@ -78,6 +78,7 @@ class VllmArguments:
     use_async_engine: bool = True
     data_parallel_size: int = 1
     log_level: Literal['critical', 'error', 'warning', 'info', 'debug', 'trace'] = 'info'
+    vllm_quantization: Optional[str] = None
 
     def __post_init__(self):
         self.limit_mm_per_prompt = ModelArguments.parse_to_dict(self.limit_mm_per_prompt)