[Bugfix] Merge MM embeddings by index instead of token IDs (#16229)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Signed-off-by: NickLucche <nlucches@redhat.com>
Signed-off-by: Roger Wang <hey@rogerw.io>
Co-authored-by: NickLucche <nlucches@redhat.com>
Co-authored-by: Roger Wang <hey@rogerw.io>

vllm/model_executor/models/deepseek_mtp.py

@@ -101,6 +101,9 @@ class DeepSeekMultiTokenPredictor(nn.Module):
         )
         self.logits_processor = LogitsProcessor(config.vocab_size)
 
+    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.embed_tokens(input_ids)
+
     def forward(
         self,
         input_ids: torch.Tensor,
@@ -142,6 +145,9 @@ class DeepSeekMTP(nn.Module, SupportsPP):
                                                  prefix=maybe_prefix(
                                                      prefix, "model"))
 
+    def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
+        return self.model.get_input_embeddings(input_ids)
+
     def forward(
         self,
         input_ids: torch.Tensor,