ngqtrung/omr-8b-grpo-ppexplore
Image-Text-to-Text • 9B • Updated • 30
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.
-
-
ngqtrung/omr-8b-grpo-base
Image-Text-to-Text • 9B • Updated • 30 -
ngqtrung/video-8b-grpo-base
Image-Text-to-Text • 9B • Updated • 30 -
ngqtrung/video-8b-grpo-sft770
Image-Text-to-Text • 9B • Updated • 30 -
ngqtrung/video-8b-grpo-ppexplore
Image-Text-to-Text • 9B • Updated • 24 -
ngqtrung/video-8b-grpo-ppexplore-n16k8
Image-Text-to-Text • 9B • Updated • 28