LaTeX OCR Env
📐
Transcribe math images to LaTeX and receive a score
LaTeX OCR environment, model, dataset, and five-run training comparison across four models, including unstable and stabilized Gemma.
Transcribe math images to LaTeX and receive a score
Note Live OpenEnv environment: formula images, indexed/streaming tasks, and server-side OCR rewards.
Note Published Qwen3.5 GRPO checkpoint from the LaTeX OCR experiments; its original model ID is preserved.
Note Upstream image/LaTeX dataset used by the environment. Maintained by unsloth.
Cross-model RL training, evaluation, and run history
Note Five comparison runs: Qwen3-VL, Qwen3.5, GLM-OCR, and both unstable/stabilized Gemma; plus 12 earlier segments and overnight checkpoint results. Reproduction guide and recovered source included.