From 7d8d71e9897e44f0b148cec117291387ea26ed33 Mon Sep 17 00:00:00 2001 From: Alexander Crettenand Date: Thu, 10 Sep 2026 08:49:35 -0500 Subject: [PATCH] [Bugfix] Qwen3-VL(-MoE): pass architectures to with_hf_config for pipeline parallelism (#43272) Signed-off-by: Xonder Co-authored-by: Claude --- vllm/model_executor/models/qwen3_vl.py | 5 ++++- vllm/model_executor/models/qwen3_vl_moe.py | 5 ++++- 2 files changed, 8 insertions(+), 2 deletions(-) diff --git a/vllm/model_executor/models/qwen3_vl.py b/vllm/model_executor/models/qwen3_vl.py index 912e02e10d..d99cfe7aa2 100644 --- a/vllm/model_executor/models/qwen3_vl.py +++ b/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"), ) diff --git a/vllm/model_executor/models/qwen3_vl_moe.py b/vllm/model_executor/models/qwen3_vl_moe.py index c7cc440b8c..193298a186 100644 --- a/vllm/model_executor/models/qwen3_vl_moe.py +++ b/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"), )