0.3.0

This release fixes a few bugs, add some safety checks and fixes an issue with storing many nodes in the same location. The latter should improve recall for low-dimension data.

What's Changed

• Build on ARM by @cevian in #102
• Fix handling of many graph nodes at the same location by @cevian in #111
• Fix hot update handling by @cevian in #117
• Panic if AVX or FMA is missing by @cevian in #119

New Contributors

• @mcadariu made their first contribution in #89

Full Changelog: 0.2.0...0.3.0

0.2.0

This is a major release

This release contains major algorithmic improvements to make the index much more performant. The (now default) SBQ mode allows for very fast retrieval over a compressed index. Please see the accompanying benchmark posts for more information.

• improved DiskANN algorithm including better stopping criteria and other algorithmic changes
• A new quantization algorithm called Statistical Binary Quantization (SBQ) for improved accuracy on quantized retrieval. This is SOTA for binary quantization.
• Rescoring as part of index retrieval -- which greatly improves accuracy for partitioned tables.
• Deprecating PQ
• New extension name vectorscale and project name pgvectorscale
• New index name diskann

0.0.2

Prepare for release 0.0.2

Release 0.0.1

Prepare 0.0.1

0.0.1-dev.1

First release for dev