February 2025

Multivector Search ready and Table data preview available in Cloud UI

Features

  • Multivector Search is now live: documents can be stored as contextualized vector lists. Fast multi-vector queries are supported at scale, powered by our XTR optimization.
  • Drop_index added to SDK: users can remove unused or outdated indexes from your tables.
  • Explore Your Data at a Glance: preview sample data from any table with a single click. lancedb-cloud
  • Search by Project/Table in Cloud UI: allow users to quickly locate the desired project/table.lancedb-cloud

Bug fix

  • FTS stability fix: Resolved a crash in Full Text Search (FTS) during flat-mode searches.
  • prefilter parameter enforcement: Fixed a bug where the prefilter parameter was not honored in FTS queries.
  • Vector index bounds error: Addressed an out-of-bounds indexing issue during vector index creation.
  • distance_range() compatibility: Fixed errors when performing vector searches with distance_range() on unindexed rows.
  • Error messaging improvements: Replaced generic HTTP 500 errors with detailed, actionable error messages for easier debugging.
January 2025

Support Hamming Distance and GPU based indexing ready

Features

  • Support Hamming distance and binary vector: Added hamming as a distance metric (joining l2, cosine, dot) for binary vector similarity search.
  • GPU-Accelerated IVF-PQ indexing: Build IVF-PQ indexes 10x faster. enterprise
  • AWS Graviton 4 & Google Axion build optimizations: ARM64 SIMD acceleration cuts query costs. enterprise
  • float16 Vector Index Supports: reduce storage size while maintaining search quality.
  • Self-Serve Cloud Onboarding: new workflow-based UI guides users for smooth experience. lancedb-cloud

Bug fix

  • list_indices and index_stats now always fetch the latest version of the table by default unless a specific version is explicitly provided.
  • Error message fix: Improved clarity for cases where create_index is called to create a vector index on tables with fewer than 256 rows.
  • TypeScript SDK fixes: Resolved an issue where 1createTable()failed to correctly save embeddings and formergeInsert` not utilizing saved embeddings. lancedb#2065
  • Multi-vector schema inference: Addressed an issue where the vector column could not be inferred for multi-vector indexes. lancedb#2026
  • Hybrid search consistency: Fixed a discrepancy where hybrid search returned distance values inconsistent with standalone vector search. lancedb#2061
December 2024

Performant SQL queries at scale and more cost-effective vector search

Features

  • Run SQL with massive datasets: added Apache Arrow flight-SQL protocol to run SQL queries with billions of data and return in seconds. enterprise.
  • Accelerate vector search: added our Quantinized-IVF algorithm and other optimization techniques to improve QPS/core. enterprise
  • Azure Stack Router Deployment: route traffic efficiently to serve low query latency. enterprise
  • Distance range filtering. filter query results using distance_range() to return search results with a lowerbound, upperbound or a range [lance#3326].
  • Full-Text Search(FTS) indexing options: configure tokenizers, stopword lists and more at FTS index creation.

Bug fix

  • Full-text search parameters: Fixed an issue where full-text search index configurations were not applied correctly. lancedb#1928
  • Float16 vector queries: Addressed a bug preventing the use of lists of Float16 values in vector queries. lancedb#1931
  • Versioned checkout API: Resolved inconsistencies in the checkout method when specifying the version parameter. lancedb#1988
  • Table recreation error: Fixed an issue where dropping and recreating a table with the same name resulted in a table creation error.