The model is initialized from the
ColBERTv1.0-bert-based-spanish-mmarcoES checkpoint and trained using the ColBERTv2 style of training.
It was trained on 1 Tesla L4 GPU with 24GBs of memory with 20k warmup steps warmup using a batch size of 64 and the AdamW optimizer with a constant learning rate of 1e-05.
Total training time was around 80 hours.
The model is fine-tuned on the French version of the
mMARCO dataset, a multi-lingual machine-translated version of the MS MARCO dataset.
The model is evaluated on the smaller development set of mMARCO-fr, which consists of 6,980 queries for a corpus of 8.8M candidate passages. We report the mean reciprocal rank (MRR) and recall at various cut-offs (R@k).
