pgvector, an open-source PostgreSQL extension that provides vector similarity search capabilities, has released v0.7.0. This new release includes many new functional and performance features for supporting vector similarity search workloads in PostgreSQL.

This latest version of pgvector adds new vector types, including halfvec (2-byte floats; indexing up to 4,000 dimensions) and sparsevec (indexing up to 1,000 nonzero dimensions), and includes indexing support for binary vectors using the bit type (indexing up to 64,000 dimensions). Additionally, this release adds support for quantizing vectors using expression indexes, including from 4-byte to 2-byte floats and binary quantization using binary_quantize function. pgvector 0.7.0 also adds new distance functions, including hamming_distance and jaccard_distance for bit vectors, and now supports HNSW indexing for L1 distance operations. pgvector 0.7.0 also includes additional support for SIMD with CPU dispatching for Linux x86-64 architectures.

For more information, please see the CHANGELOG for 0.7.0:

https://github.com/pgvector/pgvector/blob/master/CHANGELOG.md#070-2024-04-29

For more information about pgvector, including how to get started, please visit the project repository on GitHub:

https://github.com/pgvector/pgvector