Qwen3-VL-8B GRPO RLVR checkpoints from a token-dropout exploration study. OMR ppexplore=winner (0.714); video ~0.485 dead-heat.
Nguyen Quang Trung
ngqtrung
AI & ML interests
None yet
Recent Activity
updated a dataset about 22 hours ago
ngqtrung/videoqa-samples-videos published a dataset about 22 hours ago
ngqtrung/videoqa-samples-videos updated a Space about 22 hours ago
ngqtrung/videoqa-samplesOrganizations
Qwen3-VL-8B RLVR — Models (v1)
Qwen3-VL-8B GRPO RLVR checkpoints from a token-dropout exploration study. OMR ppexplore=winner (0.714); video ~0.485 dead-heat.
Qwen3-VL-8B RLVR — Datasets (v1)
Curated SFT + GRPO RL datasets (video MC-QA, OMR math-image, OpenMMReasoner-RL, Vero) for Qwen3-VL-8B post-training.
models 12
ngqtrung/video-8b-grpo-ppexplore-n16k8
Image-Text-to-Text • 9B • Updated • 26
ngqtrung/video-8b-grpo-ppexplore
Image-Text-to-Text • 9B • Updated • 21
ngqtrung/video-8b-grpo-sft770
Image-Text-to-Text • 9B • Updated • 28
ngqtrung/video-8b-grpo-base
Image-Text-to-Text • 9B • Updated • 28
ngqtrung/omr-8b-grpo-base
Image-Text-to-Text • 9B • Updated • 28
ngqtrung/omr-8b-grpo-ppexplore
Image-Text-to-Text • 9B • Updated • 28
ngqtrung/verify-tool
Updated
ngqtrung/Qwen3-Omni-Thinker-30B-Instruct
Image-Text-to-Text • 32B • Updated • 3
ngqtrung/Qwen3-Omni-Thinker-30B-Thinking
Image-Text-to-Text • 32B • Updated • 3
ngqtrung/Qwen2.5-Omni-Thinker-7B
Image-Text-to-Text • 9B • Updated • 4
datasets 48
ngqtrung/videoqa-samples-videos
Viewer • Updated • 101 • 69
ngqtrung/vero-rl
Viewer • Updated • 95.1k • 213
ngqtrung/openmmreasoner-rl-74k
Updated • 82
ngqtrung/omr-grpo-val
Viewer • Updated • 15k • 44
ngqtrung/omr-grpo-train
Viewer • Updated • 75k • 61
ngqtrung/videorl-video-val
Viewer • Updated • 11.3k • 116
ngqtrung/videorl-video-rl-train
Viewer • Updated • 212k • 87
ngqtrung/ommr-sft-recipe
Updated • 33
ngqtrung/vmar-sft-distill-raw
Viewer • Updated • 506k • 47
ngqtrung/vmar-realgold-eval
Viewer • Updated • 4.24k • 47