From cec275efcef62a9fb6ea5c3445572dddf9736206 Mon Sep 17 00:00:00 2001 From: GuanH <60228748+GuanH@users.noreply.github.com> Date: Sat, 15 Nov 2025 02:44:27 +0800 Subject: [PATCH] [Bugfix] resolve Qwen3-VL GPTQModel quantized model loading failure (#28663) Signed-off-by: GuanH Signed-off-by: Isotr0py Co-authored-by: Isotr0py --- vllm/model_executor/models/qwen3_vl.py | 4 +++- vllm/model_executor/models/utils.py | 5 +++-- 2 files changed, 6 insertions(+), 3 deletions(-) diff --git a/vllm/model_executor/models/qwen3_vl.py b/vllm/model_executor/models/qwen3_vl.py index faeb9f81d9..f1c020ab58 100644 --- a/vllm/model_executor/models/qwen3_vl.py +++ b/vllm/model_executor/models/qwen3_vl.py @@ -1138,7 +1138,9 @@ class Qwen3LLMForCausalLM(Qwen3ForCausalLM): self.config = config self.quant_config = quant_config - self.model = Qwen3LLMModel(vllm_config=vllm_config, prefix=prefix) + self.model = Qwen3LLMModel( + vllm_config=vllm_config, prefix=maybe_prefix(prefix, "model") + ) if get_pp_group().is_last_rank: if config.tie_word_embeddings: diff --git a/vllm/model_executor/models/utils.py b/vllm/model_executor/models/utils.py index f14b79f288..e5663c8a05 100644 --- a/vllm/model_executor/models/utils.py +++ b/vllm/model_executor/models/utils.py @@ -117,9 +117,10 @@ class AutoWeightsLoader: environment variable `VLLM_LOGGING_LEVEL=DEBUG`. """ - # Models trained using early version ColossalAI - # may include these tensors in checkpoint. Skip them. + # Models trained using early version ColossalAI or quantized by + # GPTQModel may include these tensors in checkpoint. Skip them. ROTARY_EMBEDS_UNUSED_WEIGHTS = [ + "rotary_pos_emb.inv_freq", "rotary_emb.inv_freq", "rotary_emb.cos_cached", "rotary_emb.sin_cached",