iHore/vLLM
8
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[Bugfix] Qwen3-VL/Qwen-Omni: honor max_pixels/min_pixels for video prompts (#49015)

Browse Source
This commit is contained in:
SYLAR
2026-07-18 10:29:11 -07:00
committed by GitHub
parent a287eb163f
commit 7c2acd38b7
2 changed files with 30 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
vllm/model_executor/models/qwen2_5_omni_thinker.py
+17
View File
@@ -509,6 +509,23 @@ class Qwen2_5OmniThinkerMultiModalProcessor(
**mm_kwargs,
)
merged = self.info.ctx.get_merged_mm_kwargs(mm_kwargs)
if mm_data.get("videos") and (
merged.keys() & {"size", "min_pixels", "max_pixels"}
):
mm_kwargs = dict(mm_kwargs)
video_size = dict(self.info.get_hf_processor().video_processor.size)
size_override = merged.get("size")
if size_override is not None:
video_size = video_size | size_override
min_pixels = merged.get("min_pixels")
if min_pixels is not None:
video_size["shortest_edge"] = min_pixels
max_pixels = merged.get("max_pixels")
if max_pixels is not None:
video_size["longest_edge"] = max_pixels
mm_kwargs["size"] = video_size
hf_inputs = super()._call_hf_processor(
prompt=prompt,
mm_data=mm_data,
vllm/model_executor/models/qwen3_vl.py
+13
View File
@@ -1272,6 +1272,19 @@ class Qwen3VLMultiModalProcessor(BaseMultiModalProcessor[Qwen3VLProcessingInfo])
# NOTE: a copy of is created to update do_sample_frames,
# otherwise mm_hash for the object will be incorrect.
video_mm_kwargs = dict(**mm_kwargs)
merged = self.info.ctx.get_merged_mm_kwargs(mm_kwargs)
if merged.keys() & {"size", "min_pixels", "max_pixels"}:
video_size = dict(self.info.get_video_processor().size)
size_override = merged.get("size")
if size_override is not None:
video_size = video_size | size_override
min_pixels = merged.get("min_pixels")
if min_pixels is not None:
video_size["shortest_edge"] = min_pixels
max_pixels = merged.get("max_pixels")
if max_pixels is not None:
video_size["longest_edge"] = max_pixels
video_mm_kwargs["size"] = video_size
sampled_fps = video_mm_kwargs.get("fps")
if is_list_of(sampled_fps, float):
video_mm_kwargs["fps"] = sampled_fps[item_idx]