Team Ai
Apppublic

commoncrawl/commonlid

sourceHugging Faceapache-2.0updated 2mo agoView on Hugging Face
5likes
App README

CommonLID Leaderboard

Live results table for the CommonLID benchmark and its commonlid_nano slice. Models are ranked by macro F1; click any row for per-language detail.

How the data flows

  1. 1.Each run writes <dataset>/<model>/{summary.json, predictions.jsonl} to data/results/ via commonlid run ….
  2. 2.The contents of data/results/ are uploaded to `commoncrawl/commonlid-results` with huggingface-cli upload.
  3. 3.This Space loads the dataset on boot and renders one tab per benchmark.

Local preview

bash
uv sync --extra leaderboard
uv run commonlid leaderboard serve --local-dir ./data/results

Pinning a specific results revision

Set COMMONLID_RESULTS_REVISION (or --revision on the CLI) to lock the Space to a specific commit on the results dataset.