GRPO training datasets for open-world image classification (Oxford-IIIT Pet): direct and structured-reasoning variants. From my lmms-owc work.
Juan CM
jucamohedano
AI & ML interests
AI Systems MSc at Trento 🚀🤖
Recent Activity
liked a Space 3 days ago
FineEnvs/multi-harness-rl liked a model 4 days ago
EvanOLeary/laguna-xs2-dense-k8-kernelmix updated a model 9 days ago
jucamohedano/rlm-laguna-xs2-dense-3b-v0.1Organizations
GRPO LoRA fine-tunes of Qwen3-VL-4B on OVEN (verl), from my MSc thesis. Mostly negative results across reward, prompt, and data axes.
-
jucamohedano/qwen3-vl-4b-oven-grpo-traversal
Image-Text-to-Text • Updated • 10 -
jucamohedano/qwen3-vl-4b-oven-grpo-aggregation
Image-Text-to-Text • Updated • 8 -
jucamohedano/qwen3-vl-4b-oven-grpo-aggregation-prelim
Image-Text-to-Text • Updated • 10 -
jucamohedano/qwen3-vl-4b-oven-grpo-traversal-structured-exact
Image-Text-to-Text • Updated • 11
GRPO training datasets for open-world image classification (Oxford-IIIT Pet): direct and structured-reasoning variants. From my lmms-owc work.
Qwen3-VL (2B/4B/8B/32B) rollouts + LM-judge verdicts on OVEN, from my MSc thesis.
GRPO LoRA fine-tunes of Qwen3-VL-4B on OVEN (verl), from my MSc thesis. Mostly negative results across reward, prompt, and data axes.
-
jucamohedano/qwen3-vl-4b-oven-grpo-traversal
Image-Text-to-Text • Updated • 10 -
jucamohedano/qwen3-vl-4b-oven-grpo-aggregation
Image-Text-to-Text • Updated • 8 -
jucamohedano/qwen3-vl-4b-oven-grpo-aggregation-prelim
Image-Text-to-Text • Updated • 10 -
jucamohedano/qwen3-vl-4b-oven-grpo-traversal-structured-exact
Image-Text-to-Text • Updated • 11