[V1][Speculative Decoding] Fix DeepSeek MTP (#20022)

Signed-off-by: cjackal <44624812+cjackal@users.noreply.github.com>

vllm/model_executor/models/deepseek_mtp.py

@@ -52,11 +52,6 @@ class DeepSeekMultiTokenPredictorLayer(nn.Module):
         quant_config: Optional[QuantizationConfig] = None,
     ) -> None:
         super().__init__()
-        self.embed_tokens = VocabParallelEmbedding(
-            config.vocab_size,
-            config.hidden_size,
-        )
-
         self.enorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
         self.hnorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
         self.eh_proj = nn.Linear(config.hidden_size * 2,
@@ -74,8 +69,6 @@ class DeepSeekMultiTokenPredictorLayer(nn.Module):
         inputs_embeds: Optional[torch.Tensor] = None,
         spec_step_index: int = 0,
     ) -> torch.Tensor:
-        if inputs_embeds is None:
-            inputs_embeds = self.embed_tokens(input_ids)
         assert inputs_embeds is not None
         # masking inputs at position 0, as not needed by MTP
         inputs_embeds[positions == 0] = 0
@@ -112,7 +105,10 @@ class DeepSeekMultiTokenPredictor(nn.Module):
             for idx in range(self.mtp_start_layer_idx,
                              self.mtp_start_layer_idx + self.num_mtp_layers)
         })
-
+        self.embed_tokens = VocabParallelEmbedding(
+            config.vocab_size,
+            config.hidden_size,
+        )
         self.logits_processor = LogitsProcessor(config.vocab_size)
 
     def forward(
@@ -123,6 +119,8 @@ class DeepSeekMultiTokenPredictor(nn.Module):
         inputs_embeds: Optional[torch.Tensor] = None,
         spec_step_idx: int = 0,
     ) -> torch.Tensor:
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
         current_step_idx = (spec_step_idx % self.num_mtp_layers)
         return self.layers[str(self.mtp_start_layer_idx + current_step_idx)](
             input_ids,
@@ -242,6 +240,12 @@ class DeepSeekMTP(nn.Module, SupportsPP):
                     if name.endswith(".bias") and name not in params_dict:
                         continue
 
+                    # According to DeepSeek-V3 Technical Report, MTP modules
+                    # shares embedding layer. We only load the first weights.
+                    if (spec_layer != self.model.mtp_start_layer_idx
+                            and ".layers" not in name):
+                        continue
+
                     param = params_dict[name]
                     weight_loader = getattr(param, "weight_loader",
                                             default_weight_loader)
@@ -253,17 +257,25 @@ class DeepSeekMTP(nn.Module, SupportsPP):
         """
         Rewrite the weight name to match the format of the original model.
         Add .mtp_block for modules in transformer layer block for spec layer
+        and rename shared layer weights to be top level.
         """
         spec_layer_weight_names = [
             "embed_tokens", "enorm", "hnorm", "eh_proj", "shared_head"
         ]
+        shared_weight_names = ["embed_tokens"]
         spec_layer_weight = False
+        shared_weight = False
         for weight_name in spec_layer_weight_names:
             if weight_name in name:
                 spec_layer_weight = True
+                if weight_name in shared_weight_names:
+                    shared_weight = True
                 break
         if not spec_layer_weight:
             # treat rest weights as weights for transformer layer block
             name = name.replace(f"model.layers.{spec_layer}.",
                                 f"model.layers.{spec_layer}.mtp_block.")
+        elif shared_weight:
+            # treat shared weights as top level weights
+            name = name.replace(f"model.layers.{spec_layer}.", "model.")
         return name

vllm/v1/spec_decode/eagle.py

@@ -148,7 +148,7 @@ class EagleProposer:
             assert self.runner is not None
 
             # FIXME: need to consider multiple kv_cache_groups
-            attn_metadata = self.runner.attn_metadata_builder.build(
+            attn_metadata = self.runner.attn_metadata_builders[0].build(
                 common_prefix_len=0,
                 common_attn_metadata=common_attn_metadata,
             )