[4/N] Initialize MM components in context managers (M-P) (#32663)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

vllm/model_executor/models/qwen3_vl_moe.py

@@ -453,14 +453,14 @@ class Qwen3VLMoeForConditionalGeneration(
                 ]
 
         with self._mark_language_model(vllm_config):
-            self.language_model = Qwen3MoeLLMForCausalLM(
+            self.language_model = language_model = Qwen3MoeLLMForCausalLM(
                 vllm_config=vllm_config, prefix=maybe_prefix(prefix, "language_model")
             )
 
             # Whether to include the gate_up_proj mapping is determined by
             # the language model.
             self.packed_modules_mapping = (
-            self.packed_modules_mapping | self.language_model.packed_modules_mapping
+                self.packed_modules_mapping | language_model.packed_modules_mapping
             )
 
         self.make_empty_intermediate_tensors = (

vllm/model_executor/models/skyworkr1v.py

@@ -908,7 +908,6 @@ class SkyworkR1VChatModel(nn.Module, SupportsMultiModal, SupportsPP):
         **kwargs: object,
     ) -> IntermediateTensors:
         if intermediate_tensors is not None:
-            input_ids = None
             inputs_embeds = None
 
         forward_kwargs = {

vllm/model_executor/models/step3_vl.py

@@ -1104,14 +1104,6 @@ class Step3VLForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsPP)
     ) -> torch.Tensor | IntermediateTensors:
         if intermediate_tensors is not None:
             inputs_embeds = None
-        elif inputs_embeds is None:
-            vision_embeddings = self.embed_multimodal(**kwargs)
-            inputs_embeds = self.embed_input_ids(
-                input_ids,
-                vision_embeddings,
-                is_multimodal=input_ids == self.config.image_token_id,
-            )
-            input_ids = None
 
         hidden_states = self.language_model(
             input_ids, positions, intermediate_tensors, inputs_embeds=inputs_embeds

vllm/model_executor/models/tarsier.py

@@ -597,14 +597,7 @@ class TarsierForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsPP)
     ) -> torch.Tensor | IntermediateTensors:
         if intermediate_tensors is not None:
             inputs_embeds = None
-        elif inputs_embeds is None:
-            vision_embeddings = self.embed_multimodal(**kwargs)
-            inputs_embeds = self.embed_input_ids(
-                input_ids,
-                vision_embeddings,
-                is_multimodal=input_ids == self.config.image_token_index,
-            )
-            input_ids = None
+
         hidden_states = self.language_model.model(
             input_ids=input_ids,
             positions=positions,