We are happy to announce the support of pgvector 0.7.0, a powerful extension for PostgreSQL that enables efficient and scalable vector search capabilities. This release marks a significant milestone in the development of pgvector, which has been gaining popularity among developers and data scientists alike. Here some insights for pgvector 0.7.0 version:
pgvector 0.7.0 is now available in all RDS PostgreSQL
Pgvector 0.7.0 introduces two new vector types: halfvec and sparsevec. Halfvec allows for indexing up to 4,000 dimensions using 2-byte floats, while sparsevec enables indexing up to 1,000 nonzero dimensions. Additionally, the release includes indexing support for binary vectors using the bit type, which can handle up to 64,000 dimensions.
The new version also adds support for quantizing vectors using expression indexes. This feature enables users to convert vectors from 4-byte to 2-byte floats and even binary quantization using the binary_quantize function. This improvement enables more efficient storage and querying of vectors.
Pgvector 0.7.0 introduces two new distance functions: hamming_distance and jaccard_distance, specifically designed for bit vectors. The release also includes support for HNSW (Hierarchical Navigable Small World) indexing for L1 distance operations, further enhancing the performance of vector similarity searches.
To further optimize performance, pgvector 0.7.0 includes additional support for SIMD (Single Instruction, Multiple Data) and CPU dispatching for Linux x86-64 architectures. This feature enables the extension to take advantage of the CPU's capabilities, resulting in faster query execution times.
For more information see RDS PostgreSQL pgvector installation section in our Help Center.