xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-25 01:45:32 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] fix Qwen2.5-Omni processor output mapping (#23058)

Browse Source 
Signed-off-by: double7 <33449816+DoubleVII@users.noreply.github.com>
Co-authored-by: 杨森 <yangsen.double7@bytedance.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
This commit is contained in:
double7 2025-08-18 13:09:11 +08:00 committed by GitHub
parent 7be3a59d8e
commit 9f1c642254
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 5 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/model_executor/models/qwen2_5_omni_thinker.py
View File

@ -88,6 +88,11 @@ def _qwen2_5_omni_thinker_field_config(hf_inputs: Mapping[str, torch.Tensor]):
video_grid_thw = hf_inputs.get("video_grid_thw", torch.empty((0, 3)))
video_grid_sizes = video_grid_thw.prod(-1)
# vllm use `second_per_grid_ts` to compute multimodal rotary embedding
video_second_per_grid = hf_inputs.get("video_second_per_grid", None)
if video_second_per_grid is not None:
hf_inputs["second_per_grid_ts"] = video_second_per_grid
return dict(
input_audio_features=MultiModalFieldConfig.flat_from_sizes(
"audio", audio_feature_lengths, dim=1),