github alibaba/zvec v0.6.0
on GitHub

one day ago

🚀 Zvec v0.6.0 Release Notes

📦 New Features

  • Turbo Quantizer Abstraction: Introduced a pluggable Quantizer abstraction in the Turbo module with a uniform interface that future quantizers (int8 uniform, int8 record, PQ, RaBitQ, …) can implement, decoupling quantization logic from index builders and searchers. Includes the first concrete Fp32Quantizer implementation and scalar FP32 distance kernels. (#546)
  • Random Rotation for INT8/INT4 Quantization: Added an optional random rotation feature to INT8/INT4 quantization, distributing variance evenly across vector dimensions to reduce quantization error and improve recall. On cohere-1m: HNSW INT8 recall improved from 0.9285 → 0.9397, Flat INT8 from 0.9695 → 0.9881; INT4 recall improved dramatically (HNSW 0.2114 → 0.7117). (#483)
  • Group-By Search: Zvec now supports group-by search (a.k.a. group-by deduplication) across the core interface, algorithm, and DB layers. Users can retrieve top-K results per group instead of globally, with support for fetch_vector, is_linear, and bf_pks query modes on Flat, HNSW, HNSW-RaBitQ, and sparse indexes. Group-by is now also exposed in the Python API. (#527, #561)
  • Zero-Copy Python Vector Query: Replaced serialize_vector memcpy with VectorViewClause that points directly at the numpy buffer, eliminating redundant memory copies for Python dense vector queries. (#517)
  • UTF-8 Tokenizer & ASCII Folding: Integrated utf8proc 2.11.3 to upgrade the FTS text analysis pipeline with full Unicode support — Unicode-aware word boundary detection, codepoint-aware lowercase conversion, and a new ASCII folding token filter that converts Unicode characters to ASCII equivalents. (#515)
  • Standard Tokenizer (UAX #29): Replaced the previous general-category tokenizer with Unicode 17 UAX #29 word-boundary tables generated from official Unicode data, implementing Lucene-style token selection for alphanumeric, numeric, ideographic, hiragana, katakana, hangul, southeast Asian, regional indicator, and common emoji sequences. (#547)
  • Stemmer Token Filter: Added a stemmer token filter for FTS based on Snowball 3.1.1, reducing words to their root form with support for 34+ languages configurable via stemmer_lang in extra_params. (#513)
  • DiskANN C API: Added complete DiskANN C API interfaces including index params setter/getters, query params with CRUD operations, and query wiring for vector query / group-by vector query / sub-query, following existing HNSW/FTS patterns. (#495)

🚀 Performance Improvements

  • Faster FTS Conjunction Queries: Added block-max skip and score early-exit to ConjunctionIterator — skip entire non-competitive blocks (128 docs) by checking block-max score upper bounds, and short-circuit score accumulation when remaining upper bound cannot beat the threshold. Benchmark (500k docs, quora dataset): AND queries 22–38% faster, phrase queries 33% faster. (#441)
  • IVF Index Optimization: Miscellaneous optimizations in IVF index internals. (#528)

🐞 Bug Fixes

  • FTS Correctness: Fixed segment stats not written on reopen, zero-match filter semantics, global doc id gaps during compaction, and snapshot read-only indexes without flush. (#566, #584, #588, #569)
  • Index Stability: Fixed lost-wakeup race in DiskANN and HNSW sparse builder progress loops, IVF nprobe selected list scanning, and miscellaneous IVF index bugs. (#530, #544, #524)
  • SQL Engine: Fixed misused group_count / group_topk in the SQL engine. (#531)
  • Collection: Fixed LOCK file to open read-only when collection is read-only, and preserved the live writing segment across CreateIndex/DropIndex DDL failures so persisted-segment tasks no longer corrupt the active segment. (#541, #574)
  • Python Bindings: Added query validation during execution to catch invalid queries early. (#538)
  • Build Fixes: Propagated compiler settings to Arrow ExternalProject, propagated CMAKE_C_COMPILER to LZ4, fixed C API duplicated & mismatched definitions, and accepted Xcode 26.5's renamed libtool version string in iOS Arrow build. (#542, #551, #525, #581)
  • Misc Fixes: Fixed mismatched doc strings in .pyi stubs, clarified create path existence error, and fixed README examples to use Query. (#529, #559, #558)

🔧 Refactoring & Cleanup

  • API Cleanup: Removed QueryParams::set_type() and clarified multi-query wording in docs. (#518, #590)
  • Code Quality: Enabled targeted bugprone clang-tidy checks, ignored dirty submodule worktrees, and fixed typos in IVF searcher tests. (#500, #523, #526)
  • DiskANN I/O Decoupling: Replaced the separate DiskANN plugin .so and hard build-time dependency on libaio-dev with a thread-safe singleton that dlopens libaio.so at runtime and caches syscall pointers. Falls back to synchronous pread() with a warning when libaio is absent, simplifying the build and deployment story. (#532)

🤖 CI & Build Improvements

  • Build Hardening: Declared and enforced 64-bit-only Python support. (#576)
  • CI Scheduling: RISC-V build now runs weekly instead of per-PR. (#577)
  • Dependency Bumps: actions/cache 5→6. (#537)

📝 Documentation

  • Updated README, fixed WeChat QR code link, and clarified multi-query wording. (#589, #575, #590)

👋 New Contributors

Thanks to all first-time contributors in this release:

Full Changelog: v0.5.1...v0.6.0

Check out latest releases or
releases around alibaba/zvec v0.6.0

Don't miss a new zvec release

NewReleases is sending notifications on new releases.

Get notifications