[Bugfix] [SpecDecode] Default speculative_draft_tensor_parallel_size to 1 when...

[Bugfix] [SpecDecode] Default speculative_draft_tensor_parallel_size to 1 when using MLPSpeculator (#7105)
Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com>

vllm/config.py

@@ -1068,7 +1068,7 @@ class SpeculativeConfig:
             draft_parallel_config = (
                 SpeculativeConfig.create_draft_parallel_config(
                     target_parallel_config,
-                    speculative_draft_tensor_parallel_size))
+                    speculative_draft_tensor_parallel_size, draft_hf_config))
 
         if num_speculative_tokens is None:
             raise ValueError(
@@ -1136,13 +1136,21 @@ class SpeculativeConfig:
     @staticmethod
     def create_draft_parallel_config(
         target_parallel_config: ParallelConfig,
-        speculative_draft_tensor_parallel_size: Optional[int]
+        speculative_draft_tensor_parallel_size: Optional[int],
+        draft_hf_config: PretrainedConfig,
     ) -> ParallelConfig:
         """Create a parallel config for use by the draft worker.
 
         This is mostly a copy of the target parallel config, except the tp_size.
         """
         if speculative_draft_tensor_parallel_size is None:
+            if draft_hf_config.model_type == "mlp_speculator":
+                speculative_draft_tensor_parallel_size = 1
+                if target_parallel_config.tensor_parallel_size > 1:
+                    logger.warning(
+                        "MLPSpeculator cannot currently be run with tp>1; "
+                        "setting speculative_draft_tensor_parallel_size=1")
+            else:
                 speculative_draft_tensor_parallel_size = \
                     target_parallel_config.tensor_parallel_size
         elif speculative_draft_tensor_parallel_size != 1: