xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 03:26:12 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Fix MRoPE Errors in the Qwen-VL Model When Processing Pure Text (#18526)

Browse Source 
Co-authored-by: 松灵 <wpf272043@alibaba-inc.com>
Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>
Co-authored-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
燃 2025-05-22 20:22:53 +08:00 committed by GitHub
parent fa72f9a812
commit f6037d1907
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 4 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
vllm/worker/model_runner.py
View File

@ -729,7 +729,10 @@ class ModelInputForGPUBuilder(ModelRunnerInputBuilderBase[ModelInputForGPU]):
mm_kwargs, placeholder_maps = MultiModalPlaceholderMap.from_seq_group(
seq_group_metadata,
range(positions[0], positions[0] + len(positions)))
if not mm_kwargs:
# M-RoPE requires mrope_positions even for plain text; return early
# when mm_kwargs is empty only if inter_data.is_prompt is False.
if not mm_kwargs and not inter_data.is_prompt:
return
inter_data.multi_modal_kwargs = mm_kwargs
@ -741,12 +744,6 @@ class ModelInputForGPUBuilder(ModelRunnerInputBuilderBase[ModelInputForGPU]):
video_grid_thw = mm_kwargs.get("video_grid_thw", None)
audio_feature_lengths = mm_kwargs.get("audio_feature_lengths",
None)
assert (
image_grid_thw is not None or video_grid_thw is not None
or audio_feature_lengths is not None), (
"mrope embedding type requires multi-modal input mapper "
"returns 'image_grid_thw' or 'video_grid_thw' or "
"'audio_feature_lengths'.")
second_per_grid_ts = mm_kwargs.get("second_per_grid_ts", None)
use_audio_in_video = mm_kwargs.get("use_audio_in_video", False)