[V1] Refactor model executable interface for all text-only language models (#10374)

Signed-off-by: Roger Wang <ywang@roblox.com>

vllm/model_executor/models/qwen2_moe.py

@@ -344,6 +344,9 @@ class Qwen2MoeModel(nn.Module):
             make_empty_intermediate_tensors_factory(
                 ["hidden_states", "residual"], config.hidden_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,
@@ -351,9 +354,13 @@ class Qwen2MoeModel(nn.Module):
         kv_caches: List[torch.Tensor],
         attn_metadata: AttentionMetadata,
         intermediate_tensors: Optional[IntermediateTensors] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
     ) -> Union[torch.Tensor, IntermediateTensors]:
         if get_pp_group().is_first_rank:
-            hidden_states = self.embed_tokens(input_ids)
+            if inputs_embeds is not None:
+                hidden_states = inputs_embeds
+            else:
+                hidden_states = self.get_input_embeddings(input_ids)
             residual = None
         else:
             assert intermediate_tensors is not None
@@ -395,6 +402,9 @@ class Qwen2MoeForCausalLM(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,
@@ -402,9 +412,11 @@ class Qwen2MoeForCausalLM(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(