[Bugfix] Qwen3-VL(-MoE): pass architectures to with_hf_config for pipeline parallelism (#43272)

Signed-off-by: Xonder <xonder@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>

vllm/model_executor/models/qwen3_vl.py

@@ -1892,7 +1892,10 @@ class Qwen3VLForConditionalGeneration(
 
         with self._mark_language_model(vllm_config):
             self.language_model = Qwen3LLMForCausalLM(
-                vllm_config=vllm_config.with_hf_config(config.text_config),
+                vllm_config=vllm_config.with_hf_config(
+                    config.text_config,
+                    architectures=["Qwen3ForCausalLM"],
+                ),
                 prefix=maybe_prefix(prefix, "language_model"),
             )
 

vllm/model_executor/models/qwen3_vl_moe.py

@@ -275,7 +275,10 @@ class Qwen3VLMoeForConditionalGeneration(
 
         with self._mark_language_model(vllm_config):
             self.language_model = Qwen3MoeLLMForCausalLM(
-                vllm_config=vllm_config.with_hf_config(config.text_config),
+                vllm_config=vllm_config.with_hf_config(
+                    config.text_config,
+                    architectures=["Qwen3MoeForCausalLM"],
+                ),
                 prefix=maybe_prefix(prefix, "language_model"),
             )