[V1] Support VLMs with fine-grained scheduling (#9871)

Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Co-authored-by: Roger Wang <ywang@roblox.com>

vllm/model_executor/models/opt.py

@@ -360,6 +360,9 @@ class OPTForCausalLM(nn.Module, SupportsPP):
         self.make_empty_intermediate_tensors = (
             self.model.make_empty_intermediate_tensors)
 
+    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,
@@ -367,9 +370,11 @@ class OPTForCausalLM(nn.Module, SupportsPP):
         kv_caches: List[torch.Tensor],
         attn_metadata: AttentionMetadata,
         intermediate_tensors: Optional[IntermediateTensors] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
     ) -> Union[torch.Tensor, IntermediateTensors]:
         hidden_states = self.model(input_ids, positions, kv_caches,
-                                   attn_metadata, intermediate_tensors)
+                                   attn_metadata, intermediate_tensors,
+                                   inputs_embeds)
         return hidden_states
 
     def compute_logits(