lightonai/

ms-marco-en-bge-gemma

This dataset contains the MS MARCO dataset with negatives mined using ColBERT and then scored by bge-reranker-v2-gemma. It can be used to train a retrieval model using knowledge distillation, for example using PyLate.

To fine-tune a model using knowledge distillation loss we will need three distinct file:

Datasetsfrom datasets import load_dataset

train = load_dataset( "lightonai/ms-marco-en-gemma", "train"… See the full description on the dataset page: https://huggingface.co/datasets/lightonai/ms-marco-en-bge-gemma.