Add LaTeX OCR: multimodal env on the new Task API + dataset streaming - #1003

Open

adithya-s-k wants to merge 17 commits into

huggingface:mainfrom

adithya-s-k:add-latex-ocr-multimodal-streaming-env

Commits on Jul 23, 2026

Add GRPO + LaTeX-OCR tutorial notebook to examples
adithya-s-k
committed
grpo_latex_ocr tutorial: verbose cell output + example previews in eval
adithya-s-k
committed
grpo_latex_ocr tutorial: USE_TRACKIO toggle, live eval avg, 500 train tasks
adithya-s-k
committed
grpo_latex_ocr tutorial: fast eval (KV cache on, gradient checkpointing off during eval)
adithya-s-k
committed
grpo_latex_ocr tutorial: pin local env to materialize mode + cap rows for fast startup
adithya-s-k
committed
latex_ocr_env: address review feedback + add docs stub
adithya-s-k
committed
latex_ocr_env: more review fixes (split-total 0 vs unknown, strict split, stream restart, consistent task_id, validate fallback)
adithya-s-k
committed
docs: add latex_ocr env to the toctree
adithya-s-k
committed
grpo_latex_ocr tutorial: install gradio in the local-serve cell (env server imports its UI)
adithya-s-k
committed
Reward: raise exact-match weight to 0.4 (edit 0.6 / exact 0.4), env-var tunable
adithya-s-k
committed
latex_ocr: length guard against whitespace-padding reward hack
adithya-s-k
committed

Commits on Jul 24, 2026

grpo_latex_ocr example: add Results section with 4-model training curves
adithya-s-k
committed