Datasets:

infosense
/

msmarco-reduced-trecrag

Modalities:

Formats:

Languages:

Size:

Libraries:

Dataset card Data Studio Files Files and versions

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

MSMARCO Reduced TREC RAG 2024 Dataset

This dataset was created from the MS MARCO V2.1 Segmented Corpus distributed by the TREC RAG Track.
It includes only the documents referenced in the TREC RAG 2024 QRELS, plus additional randomly selected documents to bring the total size to approximately 1 GB.
This reduced collection was prepared for the 5480 Large Language Models class homework assignment.

Contents

collection: contains the 597,131 human-readable segments.
- Contains 9 files with 65,536 lines and a final file with 7,307 lines.
index: contains the non-human-readable FAISS index derived from collection.
- The collection was constructed using all-MiniLM-L12-v2 from sentence-transformers. The same model should be used for querying the index.

License and Usage

This dataset is derived from the MS MARCO V2.1 Segmented Corpus provided by the TREC RAG Track.
Original content © Microsoft.
Distribution and use are permitted for non-commercial research and educational purposes only.
Redistribution or use for commercial purposes is prohibited.

Downloads last month: 4