xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-24 18:45:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Fix Qwen2.5-VL quantized model weights loading (#23512)

Browse Source 
Signed-off-by: Zifei Tong <zifeitong@gmail.com>
This commit is contained in:
zifeitong 2025-08-24 19:40:22 -07:00 committed by GitHub
parent 39971db3aa
commit a71e4765cc
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 5 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/model_executor/models/qwen2_5_vl.py
View File

@ -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={