# How to update benchmark results

The results are stored as JSON files for each model in a Huggingface dataset located [here](https://huggingface.co/datasets/pulsia/fr-bench-pdf2md-results/tree/main).

The format of the files is detailed in the notebook `format_results.ipynb`.

## Steps to update

1. Pull the dataset in a `data` folder with:

```
git clone https://huggingface.co/datasets/pulsia/fr-bench-pdf2md-results
```
2. Put your results CSV in this folder.
3. Run all cells of the `format_results.ipynb` notebook.
4. Commit the changes in the dataset repo and push on main (or use the HF PR system).
5. The results will automatically propagate in the HF leaderboard space [here](https://huggingface.co/spaces/pulsia/fr-bench-pdf2md).

Username for pushing: pulsia
Password: create a token with write access.