Qdrant/PubMed-MV

PubMed-Multi-Vector

Overview

This dataset is a large slice of pre-embedded PubMed abstracts -- embedded with BGE-M3. It is designed to benchmark hybrid retrieval methods with corpus variables held constant. It generates dense, sparse, and ColBERT-style multi-vector representations over the exact same text corpus, accumulating over 8.37 billion multi-vector tokens across nearly 35 TB of data.

The dataset includes:

• 23.9M dense embeddings
• 23.9M sparse embeddings
• 8.37B multi-vector token embeddings
• Exact top-1000 ground truth for dense, sparse, and filtered retrieval

Dataset Contents

Each row carries the document's dense and sparse embeddings, and all eight FineWeb source fields are retained verbatim:

Field	Type	Notes
id	string	Unique document identifier
title	string	Article title
content	string	Main article text or abstract
contents	string	Alternative or full document text
PMID	int64	PubMed Unique Identifier
source_file_name	string	Origin source file path
source_row_number	int64	Line/row index within the source file
bge_m3_dense_embedding	list<float>	BGE-M3 dense vector representation
bge_m3_sparse_embedding	struct<indices: list<uint32>, values: list<float>>	BGE-M3 sparse token weights (lexical indices and values)
bge_m3_multivector_embedding	list<list<float>>	BGE-M3 multi-vector representation (ColBERT-style multi-token embeddings)

Embeddings

Each document is represented using BAAI/bge-m3.

• Dense: 1024 dimensions; Vectors are unit-norm. Compare with cosine, or equivalently dot product; do not re-normalize.
• Sparse: BM25-style sparse embeddings for hybrid-retrieval.
• Multivector: ColBERT-style dense multi-vectr embeddings for advaced reranking and rescoring workloads.

Intended Uses

PubMed-Multi-Vector is intended for large-scale evaluation of:

• dense vector search
• sparse vector search
• multi vector search
• hybrid retrieval
• vector database ingestion and indexing
• approximate nearest-neighbor algorithms
• distributed retrieval systems

It is also useful for work that needs a very large embedded corpus rather than a retrieval benchmark: quantization and compression studies, index-build cost and memory-footprint measurement, sharding and routing strategies, embedding-space analysis at web scale, and cost/recall trade-off curves where 10B vectors make the trade-off visible in a way 1M-vector benchmarks do not.

Licensing

We release this dataset under the Apache 2.0 license

Acknowledgments

This dataset would not exist without the upstream work it is built on:

• HuggingFace
• SkyPilot