LvW-bench / bench /README.md
Lucasgitguy-co's picture
LvW-bench 0.1.0: tasks, eval.yaml, leaderboard, seed results
dfcc32b verified
|
Raw History Blame Contribute Delete
713 Bytes
# bench/
- `score.py`: rescores an `lvwb` predictions file with the same rules as the Go scorer.
No model is needed: `python bench/score.py <result>.predictions.jsonl` (add `--data DIR`
to use local parquet or JSONL files instead of downloading the dataset).
## Submitting a result
1. Run the model with `lvwb run <model>` on the `standard` profile.
2. Open a discussion on this dataset, titled `Result: <model>`. Attach the result `.json`
and its `.predictions.jsonl`, and say whether the model is general or a specialist.
3. To show the score on the model's own page, open a PR on the model repo that adds the
output of `lvwb export <result.json> --eval-results` as `.eval_results/lvw-bench.yaml`.