Repository navigation

All pull requests

Labels (9)Milestones (0)New pull request

Open

(49)

Closed

(110)

fix(utils): decode local file URIs before opening images
#2140zoechen281 opened Oct 17, 2026
fix(utils): honor EXIF image orientation before resizing
#2139zoechen281 opened Oct 13, 2026
fix(eval): grade MathVision answers without executing input text
#2133Afloat16 opened Oct 2, 2026
fix(utils): composite palette and grayscale transparency onto white
#2132Afloat16 opened Oct 2, 2026
fix(utils): preserve source timing and padding indices for video frame lists
#2131Afloat16 opened Oct 2, 2026
Preserve per-video timing in fine-tuning position IDs
#2130Afloat16 opened Oct 2, 2026
Apply projector training flags to every DeepStack merger
#2129Afloat16 opened Oct 2, 2026
fix(train): honor vision learning rate without a projector override
#2127Afloat16 opened Oct 2, 2026
Honor language-model freezing for the output head
#2126Afloat16 opened Oct 2, 2026
fix(eval): score empty ODinW detections instead of dropping metrics
#2125Afloat16 opened Oct 2, 2026
fix: honor dtype and attention parameters in packed FlashAttention
#2124Afloat16 opened Oct 1, 2026
Keep a subtitle that starts on the frame
#2123SashaMIT opened Sep 28, 20261 task done
fix(qwen-vl-utils): smart_resize can return a zero-sized dimension (regression of #1145)
#2121Linxiushen opened Sep 22, 2026
Fix stale qwen-vl-utils project links
#2119yousizaitianqiong opened Sep 4, 2026
[doc] video timestamps for pre-sampled frames
#2117as-obaid opened Aug 26, 2026
Apply configured video pixel limits with SizeDict
#2116primorLee opened Aug 22, 2026
Honor configured model_max_length in finetune collation
#2115primorLee opened Aug 21, 2026
Fix torchvision video frame indices offset
#2091Travor278 opened May 3, 2026
Update attribute visual and language_model to qwen3-vl architecture
#2073padmalcom opened Mar 24, 2026
fix: avoid double resize when fetch_video receives image list
#2045HT-Yuan opened Feb 27, 2026
1
fix bug in qwen-vl-finetune/qwenvl/data/data_processor.py:
#2017Singularity0104 opened Feb 1, 2026
Fix Web Demo dependencies and runtime crashes
#1977study8677 opened Jan 6, 2026
Drop decord as a choice to read videos due to instability
#18913manifold opened Dec 3, 2025
Refactor return type of process_vision_info function
#1853maludwig opened Nov 25, 2025
Fix process_vision_info return type hint to support 2-tuple unpacking
#1852xuing opened Nov 25, 2025

Pull requests