Azuremis/

mlx7-two-tower-data

This repository contains datasets used for training Two-Tower (Dual Encoder) models for document retrieval.

The datasets provided here are structured for training dual encoder models with various sampling strategies:

classic_triplets: 48.2 MB intra_query_neg: 47.6 MB multi_pos_multi_neg: 126.5 MB

classic_triplets.parquet: Standard triplet format with (query, positive_document, negative_document)… See the full description on the dataset page: https://huggingface.co/datasets/Azuremis/mlx7-two-tower-data.