eduagarcia/open_pt_llm_leaderboard
๐ Open Portuguese LLM Leaderboard โ archived
This leaderboard is archived: submissions are closed, the evaluation cluster is off and no new result will be added. The Space is now a static React app that ships the final results with it โ it no longer downloads anything from the Hugging Face dataset repos at runtime.
Layout
Regenerating the snapshot
The compiler reads the dataset snapshots the gradio app used (eval-queue/, eval-results/, dynamic-info/), so download them first if they are not on disk:
set -a && . ./.env && set +a # HF token + repo ids
HF_HOME=. uv run python build_static_data.pyIt rewrites data/leaderboard.csv.gz (~120 KiB) and data/metadata.json.
data/leaderboard.csv.gz is binary, so the Hub stores it in Xet/LFS (*.gz is already in .gitattributes) โ make sure git lfs install has been run before committing a new snapshot, otherwise the push is rejected. A fresh clone needs git lfs pull before npm run build; the build aborts if the file is still a pointer.
Front-end
npm install
npm run dev # http://localhost:5173
npm run build # -> dist/ (what the Space serves)The Space serves the committed dist/ directly (sdk: static + app_file: dist/index.html). app_build_command is intentionally not used: the Hub evaluates the config at push time, before the build ref exists, and that race left the Space stuck in a configuration error. Run `npm run build` and commit `dist/` together with any change to the app or to `data/`.
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
