biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Model] EVS support for nano_nemotron_vl (#26269)

Browse Source 
Signed-off-by: Tomer Asida <57313761+tomeras91@users.noreply.github.com>
Signed-off-by: tomeras91 <57313761+tomeras91@users.noreply.github.com>
Signed-off-by: Eugene Khvedchenia <ekhvedchenia@nvidia.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: Eugene Khvedchenia <ekhvedchenia@nvidia.com>
This commit is contained in:
tomeras91
2025-10-06 19:23:37 +03:00
committed by GitHub
parent fc679696f8
commit b8f603cebe
3 changed files with 225 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
vllm/model_executor/models/qwen2_5_vl.py
View File

@@ -1017,9 +1017,13 @@ class Qwen2_5_VLMultiModalProcessor(Qwen2VLMultiModalProcessor):
and video_pruning_rate is not None
and video_pruning_rate > 0.0
):
T, H, W = map(int, grid_thw)
tokens_per_frame = (H // image_processor.merge_size) * (
W // image_processor.merge_size
)
num_tokens = compute_retained_tokens_count(
grid_thw,
image_processor.merge_size,
tokens_per_frame,
T,
video_pruning_rate,
)
# End of EVS-specific code