remove redundant SpeculativeConfig

vllm/engine/arg_utils.py

@@ -766,23 +766,6 @@ class EngineArgs:
         lora_group.add_argument("--default-mm-loras",
                                 **lora_kwargs["default_mm_loras"])
 
-        # Speculative arguments
-        speculative_group = parser.add_argument_group(
-            title="SpeculativeConfig",
-            description=SpeculativeConfig.__doc__,
-        )
-        speculative_group.add_argument(
-            "--speculative-config",
-            type=json.loads,
-            default=None,
-            help="The configurations for speculative decoding. Should be a "
-            "JSON string.")
-        parser.add_argument(
-            '--num-speculative-heads',
-            type=int,
-            default=EngineArgs.num_speculative_heads,
-            help='The number of speculative heads to sample from '
-                 'the draft model in speculative decoding.')
 
         # Observability arguments
         observability_kwargs = get_kwargs(ObservabilityConfig)
@@ -874,6 +857,13 @@ class EngineArgs:
                                 **vllm_kwargs["compilation_config"])
         vllm_group.add_argument("--additional-config",
                                 **vllm_kwargs["additional_config"])
+        
+        parser.add_argument(
+            '--num-speculative-heads',
+            type=int,
+            default=EngineArgs.num_speculative_heads,
+            help='The number of speculative heads to sample from '
+                 'the draft model in speculative decoding.')
 
         # Other arguments
         parser.add_argument('--disable-log-stats',