[Bug Fix] Qwen3.5-nvfp4 MTP Speculative Decoding Weight Shape Mismatch

vllm/model_executor/models/qwen3_5_mtp.py

@@ -13,7 +13,7 @@
 from vllm.distributed.parallel_state import get_pp_group
 from vllm.logger import init_logger
 from vllm.model_executor.layers.fused_moe import FusedMoE
-from vllm.model_executor.layers.linear import ColumnParallelLinear
+from vllm.model_executor.layers.linear import ReplicatedLinear
 from vllm.model_executor.layers.logits_processor import LogitsProcessor
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead,
@@ -59,7 +59,7 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
         super().__init__()
 
         model_config = vllm_config.model_config
-        quant_config = vllm_config.quant_config
+        quant_config = vllm_config.quant_config  # noqa: F841
 
         config: Qwen3_5TextConfig | Qwen3_5MoeTextConfig = model_config.hf_text_config
 
@@ -75,13 +75,12 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
             config.hidden_size,
         )
 
-        self.fc = ColumnParallelLinear(
+        self.fc = ReplicatedLinear(
             self.config.hidden_size * 2,
             self.config.hidden_size,
-            gather_output=True,
             bias=False,
-            return_bias=False,
-            quant_config=quant_config,
+            # Not quantizing the MTP layer.
+            quant_config=None,
             prefix=f"{prefix}.fc",
         )
 
@@ -125,7 +124,7 @@ def forward(
             inputs_embeds = self.pre_fc_norm_embedding(inputs_embeds)
             hidden_states = self.pre_fc_norm_hidden(hidden_states)
             hidden_states = torch.cat([inputs_embeds, hidden_states], dim=-1)
-            hidden_states = self.fc(hidden_states)
+            hidden_states, _ = self.fc(hidden_states)
             residual = None
         else:
             assert intermediate_tensors is not None