Version 8.0-M2

This version is going to be bundled with RediSearch 8.0-M2 release (version and tag names are aligned with RediSearch ones).

New Features

• Introduce INT8 vector type [MOD-8198] (#560) (#571)
• Introduce UINT8 vector type - [MOD-8230, MOD-8408] (#584)
• Add benchmarks for INT8 [MOD-8204] (#583)

Arichtetural Improvements

• Deprecate the initial capacity parameter for creating a new index (#546)
• Introduce abstract raw vectors data layer in the vector index base class to separate data from index component [MOD-7496] (#523)
• Introduce IndexComputer component to separate computation logic from vectors data and index. [MOD-6738] (#535)

Contributors

We'd like to thank all the contributors who worked on this release!

@GuyAv46, @alonre24, @lerman25 and @meiravgri

Version 0.8.1

Changes

• Handle pending jobs after deleting in-place in the tiered index (#534) (#538)
• Add AVX512VL flag to FP16 flag to enhance performance. (#522)

Contributors

We'd like to thank all the contributors who worked on this release!

@alonre24, @meiravgri and @nafraf

Version 0.8.0

This is the first 0.8 official release!

Changes

• Better dimensions selection for spaces micro-benchmarks (#502)
• Enhance benchmark files download and add dispatcher to benchmarks flow (#486)
• GitHub Actions - update action versions (#488)
• Introduce AVX512_FP16 optimization for distance calculation (#477)
• Introduce AVX512_BF16 optimization for distance calculation (#472)

Contributors

We'd like to thank all the contributors who worked on this release!

@GuyAv46, @alonre24, and @meiravgri

Version 0.8-M2

Changes

• Support bfloat16 and float16 data types (#455)

Contributors

We'd like to thank all the contributors who worked on this release!

@GuyAv46, @alonre24 and @meiravgri

Version 0.8-M1

Changes

• Fix spellcheck flow (#441)
• Improve updatable heap - [MOD-5439] (#434)
• Create SECURITY.md (#433)
• Dump neighbors in HNSW debug command [MOD-6496] (#432)
• Update README (#427)
• GHA - Use Merge queue - [MOD-6478] (#422)
• Fix flaky test_tiered_hnsw:bufferLimitAsync (#425)
• Bumped GitHub spellcheck action to latest version: 0.35.0 (#424)

🐛 Bug Fixes

• Fix hardware optimization compilation - [MOD-6567] (#438)
• Fix label deletion from an unordered map - [MOD-6710] (#435)

Contributors

We'd like to thank all the contributors who worked on this release!

@GuyAv46, @alonre24 and @jonasbn

Version 0.7.1

Changes

• Add an API to allow external lock/unlock for shared ownership to reduce locks overhead in ad-hoc BF mode (#419)
• Optimize implementation for insert and delete vectors in HNSW (#416) (#406)
• Refactor query result API - (#412)

Contributors

We'd like to thank all the contributors who worked on this release!

@GuyAv46 and @alonre24

Version 0.7.0

Changes

• HNSW refactor - data blocks (#397) - HNSW is now using data blocks to store vectors and metadata instead of array
• Tired index (#327) - Using FLAT as a frontend index to allow long async background writes (insert, update, delete) for the backend index

Contributors

We'd like to thank all the contributors who worked on this release!

@DvirDukhan, @GuyAv46, @alonre24, @chayim and @meiravgri

Version 0.7-M3

Changes

HNSW index to use data blocks (#397) - move HNSW storage layer to data blocks

Version 0.7-M2

Second milestone release for 0.7

Changes

• Feature hnsw tiered index (#327) - Allows async multi-writers indexing and deletion via garbage collection (repair and swap jobs) in HNSW. Improved index factory API, including tiered index factory. Changed info structure and API for general-purpose info sections.

Version 0.6.1

🐛 Bug Fixes

#346 - Fixed a bug in the deletion logic of a multi-value index.