From a71e4765cc0c1534f2a8891aaf628e1751f6df07 Mon Sep 17 00:00:00 2001 From: zifeitong Date: Sun, 24 Aug 2025 19:40:22 -0700 Subject: [PATCH] [Bugfix] Fix Qwen2.5-VL quantized model weights loading (#23512) Signed-off-by: Zifei Tong --- vllm/model_executor/models/qwen2_5_vl.py | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/vllm/model_executor/models/qwen2_5_vl.py b/vllm/model_executor/models/qwen2_5_vl.py index 811ecffcc1e49..0f11636ce3bd3 100644 --- a/vllm/model_executor/models/qwen2_5_vl.py +++ b/vllm/model_executor/models/qwen2_5_vl.py @@ -135,7 +135,7 @@ class Qwen2_5_VLVideoPixelInputs(TypedDict): second_per_grid_ts: torch.Tensor """ - The video time interval (in seconds) for each grid along the temporal + The video time interval (in seconds) for each grid along the temporal dimension in the 3D position IDs. Returned when `videos` is not `None`. """ @@ -852,6 +852,10 @@ class Qwen2_5_VLForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsLoRA, SupportsPP, SupportsQuant): + packed_modules_mapping = { + "gate_up_proj": ["gate_proj", "up_proj"], + } + # To ensure correct weight loading and mapping. hf_to_vllm_mapper = WeightsMapper( orig_to_new_prefix={