Downloads · 30 days
0
logan7000/mllm-open-r1-gt-qwen25vl-7b-old-full
mllm-open-r1-gt-qwen25vl-7b-old-full is a machine learning model from logan7000. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Recipe: old (beta 0, K 8, T 1.0, maxcompletionlength 1024, lr 1e-6, cosinewithminlr 0.1, warmupratio 0.03, bnpo, scalerewards group, effective batch 64 = 8 prompts/step, 1 epoch), identical to the big-tier reference s…
Downloads · 30 days
0
Access
Public
Updated Sep 5, 2026
Repo size
33.2 GB
Likes
0
Public
Click a slice to open those files.
.safetensors33.2 GB · 100%
From the Hugging Face model README
Recipe: old (beta 0, K 8, T 1.0, max_completion_length 1024, lr 1e-6, cosine_with_min_lr 0.1, warmup_ratio 0.03, bnpo, scale_rewards group, effective batch 64 = 8 prompts/step, 1 epoch), identical to the big-tier reference script examples/openr1_qwen25vl7b_gt_old.sh. best/ = best-by-val (MathVista-150 holdout) step 300; endpoint/ = checkpoint-961 (end of 1 epoch); training/ = train.log, trainer_state_best, trainer_state_endpoint, best_metric.
Eval protocol for the paper tables: v2 (T=0, top_p 0.95, 16k tokens, boxed prompt, mathruler rule + Qwen2.5-32B judge), endpoint selection, AVG4 over MathVista/MathVision/MathVerse/We-Math. Local source: /weka/scratch/jhu/dssg2026-ext-rghani1/yyang331/mllm-repro-out/mllm-co-grpo-dp/openr1_qwen25vl7b_gt_old Trained 2026-09-03..09-05 on the JHU a100 cluster.