frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Multimodal] Improve max video embedding length estimation in V1 (#24312)

Browse Source 
Signed-off-by: Roger Wang <hey@rogerw.me>
Co-authored-by: Roger Wang <hey@rogerw.me>
This commit is contained in:
Roger Wang
2025-09-06 02:33:19 -07:00
committed by GitHub
parent 305a1cc0d2
commit eddaafc1c7
2 changed files with 2 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@ -216,12 +216,9 @@ class LlavaOnevisionProcessingInfo(LlavaNextProcessingInfo):
seq_len: int,
mm_counts: Mapping[str, int],
) -> int:
max_images = mm_counts.get("image", 0)
max_videos = mm_counts.get("video", 0)
max_image_tokens = self.get_max_image_tokens() * max_images
max_total_frames = self._get_max_video_frames(seq_len -
max_image_tokens)
max_total_frames = self._get_max_video_frames(seq_len)
max_frames_per_video = min(max_total_frames // max(max_videos, 1),
_MAX_FRAMES_PER_VIDEO)

5
vllm/model_executor/models/qwen2_vl.py
View File

@ -915,12 +915,9 @@ class Qwen2VLProcessingInfo(BaseProcessingInfo):
seq_len: int,
mm_counts: Mapping[str, int],
) -> int:
max_images = mm_counts.get("image", 0)
max_videos = mm_counts.get("video", 0)
max_image_tokens = self.get_max_image_tokens() * max_images
max_total_frames = self._get_max_video_frames(seq_len -
max_image_tokens)
max_total_frames = self._get_max_video_frames(seq_len)
max_frames_per_video = min(max_total_frames // max(max_videos, 1),
_MAX_FRAMES_PER_VIDEO)