github weaviate/weaviate v1.24.0
v1.24.0 - High Frequency Updates, HNSW Binary Quantization, Support for Multiple Vectors per Class, Durability Improvements, Japanese & Chinese Tokenizers, Improved NotEqual Operator, Telemetry

latest releases: v1.25.1, v1.24.14, v1.23.15...
2 months ago

Breaking Changes

none

New Features

High Frequency Updates

Weaviate now has reinforced support for high-frequency updates, enabling tens of millions of updates per day without degrading performance. This improvement involved optimizing the handling of object properties and vector updates. Specifically, updates that do not affect the vector index avoid unnecessary processing, and updates that change vectors are processed much more efficiently. These enhancements are critical for maintaining high performance and scalability in environments with heavy update loads​.

HNSW Binary Quantization

Introducing binary quantization support for the HNSW (Hierarchical Navigable Small World) vector index, marking a significant advancement in the efficiency and performance of vector search operations. This new feature enables the conversion of vectors into compact binary formats, drastically reducing the memory footprint of the vector index while maintaining high search accuracy. The introduction of binary quantization support is a game-changer for applications with large-scale datasets, where memory efficiency is paramount. By leveraging binary quantization, you can now achieve faster search speeds and lower memory usage, making it easier to scale applications without compromising on performance.

Support for Multiple Vectors per Class

Hey Weaviate community 👋 we heard you! The most upvoted feature on the public roadmap is here!

Support for multiple vectors per class means enhancing Weaviate's flexibility and applicability for complex data models. This feature allows for more nuanced data representation, supporting diverse and multifaceted search and machine learning use cases. By accommodating multiple vectors per class, Weaviate enables richer data indexing and retrieval strategies, improving accuracy and efficiency in search queries and data analysis tasks.

  • Introduce class multi-vector support by @antas-marcin in #4180
  • Multiple Vectors: gRPC Batch API vectors by @antas-marcin in #4246
  • Multiple Vectors: Validation of none vectorizer modules when multiple vectorizer configuration is present by @antas-marcin in #4248
  • Return all named vectors if vectors is true by @dirkkul in #4253
  • Aggregate with named vectors by @dirkkul in #4254
  • Support for VectorConfig update by @aliszka in #4266
  • Summed metrics for target vectors by @aliszka in #4264
  • Target vectors marshalling by @aliszka in #4271
  • Use Async Queue(s) and VectorIndex(es) only relevant for configured legacy vector or target vectors by @aliszka in #4279
  • Hide legacy vector index config when target vectors configured by @aliszka in #4283
  • Handle named vectors when checking certainy/distance in queries by @tsmith023 in #4285

Japanese & Chinese Tokenizers

Durability Improvements

Improved NotEqual Operator

Telemetry

Other

Fixes

Performance Improvements

Testing Improvements

  • Cleanup gha caches after PR merge by @moogacs in #4162
  • Move transformers acceptance tests to ubuntu-latest-4-cores by @moogacs in #4173
  • Assert eventually gcs backup bucket creation by @moogacs in #4176
  • Assert eventually cloud backup bucket creation by @moogacs in #4178
  • Fix flaky test with named vectors return order by @dirkkul in #4255
  • Add backup tests with named vectors by @dirkkul in #4281
  • Improved class update tests by @aliszka in #4286
  • Migration of compression tests to acceptance package by @aliszka in #3923
  • Speed acceptance tests and gha by modularizing the running test and parallelize them by @moogacs in #4017
  • Change module acceptance tests to run on ubuntu-latest by @antas-marcin in #4045
  • Add debug weaviate build and related script to use remote delve by @reyreaud-l in #4051

New Contributors

Full Changelog: v1.23.10...v1.24.0

Don't miss a new weaviate release

NewReleases is sending notifications on new releases.