Merge branch 'v0.8.4-dev-wm' into 'v0.8.4-dev'

[fix]解决无法启动并行报错

See merge request dcutoolkit/deeplearing/vllm!100

vllm/engine/arg_utils.py

@@ -1122,7 +1122,6 @@ class EngineArgs:
         target_parallel_config: ParallelConfig,
         enable_chunked_prefill: bool,
         disable_log_stats: bool,
-        num_speculative_heads: Optional[int],
     ) -> Optional["SpeculativeConfig"]:
         """Initializes and returns a SpeculativeConfig object based on
         `speculative_config`.
@@ -1143,7 +1142,6 @@ class EngineArgs:
             "target_parallel_config": target_parallel_config,
             "enable_chunked_prefill": enable_chunked_prefill,
             "disable_log_stats": disable_log_stats,
-            "num_speculative_heads": num_speculative_heads,
         })
         speculative_config = SpeculativeConfig.from_dict(
             self.speculative_config)
@@ -1246,8 +1244,7 @@ class EngineArgs:
             target_model_config=model_config,
             target_parallel_config=parallel_config,
             enable_chunked_prefill=self.enable_chunked_prefill,
-            disable_log_stats=self.disable_log_stats,    
-            num_speculative_heads=self.num_speculative_heads
+            disable_log_stats=self.disable_log_stats,
         )
 
         # Reminder: Please update docs/source/features/compatibility_matrix.md