[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>
Alexander Crettenand committed
Sep 10, 2026 at 08:49 UTC
7d8d71e9897e44f0b148cec117291387ea26ed33
2 files changed
+8
-2
vllm/model_executor/models/qwen3_vl.py
+4
-1
@@ -1892,7 +1892,10 @@ class Qwen3VLForConditionalGeneration(
1892
1893
with self._mark_language_model(vllm_config):
1894
self.language_model = Qwen3LLMForCausalLM(
1895
- vllm_config=vllm_config.with_hf_config(config.text_config),
1895
+ vllm_config=vllm_config.with_hf_config(
1896
+ config.text_config,
1897
+ architectures=["Qwen3ForCausalLM"],
1898
+ ),
1899
prefix=maybe_prefix(prefix, "language_model"),
1900
)
1901
vllm/model_executor/models/qwen3_vl_moe.py
+4
-1
@@ -275,7 +275,10 @@ class Qwen3VLMoeForConditionalGeneration(
275
276
with self._mark_language_model(vllm_config):
277
self.language_model = Qwen3MoeLLMForCausalLM(
278
- vllm_config=vllm_config.with_hf_config(config.text_config),
278
+ vllm_config=vllm_config.with_hf_config(
279
+ config.text_config,
280
+ architectures=["Qwen3MoeForCausalLM"],
281
+ ),
282
prefix=maybe_prefix(prefix, "language_model"),
283
)
284