Skip to content

Latest commit

 

History

History
310 lines (266 loc) · 14.7 KB

CHANGELOG.md

File metadata and controls

310 lines (266 loc) · 14.7 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project DOES NOT adhere to Semantic Versioning at the moment.

We try to indicate most contributions here with the contributor names who are not part of the Facebook Faiss team. Feel free to add entries here if you submit a PR.

Unreleased

Changed

  • Previously, when moving indices to GPU with coarse quantizers that were not implemented on GPU, the cloner would silently fallback to CPU. This version will now throw an exception instead and the calling code would need to explicitly allow fallback to CPU by setting a flag in cloner config.

1.8.0 - 2024-02-27

Added

  • Added a new conda package faiss-gpu-raft alongside faiss-cpu and faiss-gpu
  • Integrated IVF-Flat and IVF-PQ implementations in faiss-gpu-raft from RAFT by Nvidia [thanks Corey Nolet and Tarang Jain]
  • Added a context parameter to InvertedLists and InvertedListsIterator
  • Added Faiss on Rocksdb demo to showing how inverted lists can be persisted in a key-value store
  • Introduced Offline IVF framework powered by Faiss big batch search
  • Added SIMD NEON Optimization for QT_FP16 in Scalar Quantizer. [thanks Naveen Tatikonda]
  • Generalized ResultHandler and supported range search for HNSW and FastScan
  • Introduced avx512 optimization mode and FAISS_OPT_LEVEL env variable [thanks Alexandr Ghuzva]
  • Added search parameters for IndexRefine::search() and IndexRefineFlat::search()
  • Supported large two-level clustering
  • Added support for Python 3.11 and 3.12
  • Added support for CUDA 12

Changed

  • Used the benchmark to find Pareto optimal indices. Intentionally limited to IVF(Flat|HNSW),PQ|SQ indices
  • Splitted off RQ encoding steps to another file
  • Supported better NaN handling
  • HNSW speedup + Distance 4 points [thanks Alexandr Ghuzva]

Fixed

  • Fixed DeviceVector reallocations in Faiss GPU
  • Used efSearch from params if provided in HNSW search
  • Fixed warp synchronous behavior in Faiss GPU CUDA 12

1.7.4 - 2023-04-12

Added

  • Added big batch IVF search for conducting efficient search with big batches of queries
  • Checkpointing in big batch search support
  • Precomputed centroids support
  • Support for iterable inverted lists for eg. key value stores
  • 64-bit indexing arithmetic support in FAISS GPU
  • IndexIVFShards now handle IVF indexes with a common quantizer
  • Jaccard distance support
  • CodePacker for non-contiguous code layouts
  • Approximate evaluation of top-k distances for ResidualQuantizer and IndexBinaryFlat
  • Added support for 12-bit PQ / IVFPQ fine quantizer decoders for standalone vector codecs (faiss/cppcontrib)
  • Conda packages for osx-arm64 (Apple M1) and linux-aarch64 (ARM64) architectures
  • Support for Python 3.10

Removed

  • CUDA 10 is no longer supported in precompiled packages
  • Removed Python 3.7 support for precompiled packages
  • Removed constraint for using fine quantizer with no greater than 8 bits for IVFPQ, for example, now it is possible to use IVF256,PQ10x12 for a CPU index

Changed

  • Various performance optimizations for PQ / IVFPQ for AVX2 and ARM for training (fused distance+nearest kernel), search (faster kernels for distance_to_code() and scan_list_*()) and vector encoding
  • A magnitude faster CPU code for LSQ/PLSQ training and vector encoding (reworked code)
  • Performance improvements for Hamming Code computations for AVX2 and ARM (reworked code)
  • Improved auto-vectorization support for IP and L2 distance computations (better handling of pragmas)
  • Improved ResidualQuantizer vector encoding (pooling memory allocations, avoid r/w to a temporary buffer)

Fixed

  • HSNW bug fixed which improves the recall rate! Special thanks to zh Wang @hhy3 for this.
  • Faiss GPU IVF large query batch fix
  • Faiss + Torch fixes, re-enable k = 2048
  • Fix the number of distance computations to match max_codes parameter
  • Fix decoding of large fast_scan blocks

1.7.3 - 2022-11-3

Added

  • Added sparse k-means routines and moved the generic kmeans to contrib
  • Added FlatDistanceComputer for all FlatCodes indexes
  • Support for fast accumulation of 4-bit LSQ and RQ
  • Added product additive quantization
  • Support per-query search parameters for many indexes + filtering by ids
  • write_VectorTransform and read_vectorTransform were added to the public API (by @AbdelrahmanElmeniawy)
  • Support for IDMap2 in index_factory by adding "IDMap2" to prefix or suffix of the input String (by @AbdelrahmanElmeniawy)
  • Support for merging all IndexFlatCodes descendants (by @AbdelrahmanElmeniawy)
  • Remove and merge features for IndexFastScan (by @AbdelrahmanElmeniawy)
  • Performance improvements: 1) specialized the AVX2 pieces of code speeding up certain hotspots, 2) specialized kernels for vector codecs (this can be found in faiss/cppcontrib)

Fixed

  • Fixed memory leak in OnDiskInvertedLists::do_mmap when the file is not closed (by @AbdelrahmanElmeniawy)
  • LSH correctly throws error for metric types other than METRIC_L2 (by @AbdelrahmanElmeniawy)

1.7.2 - 2021-12-15

Added

  • Support LSQ on GPU (by @KinglittleQ)
  • Support for exact 1D kmeans (by @KinglittleQ)

1.7.1 - 2021-05-27

Added

  • Support for building C bindings through the FAISS_ENABLE_C_API CMake option.
  • Serializing the indexes with the python pickle module
  • Support for the NNDescent k-NN graph building method (by @KinglittleQ)
  • Support for the NSG graph indexing method (by @KinglittleQ)
  • Residual quantizers: support as codec and unoptimized search
  • Support for 4-bit PQ implementation for ARM (by @vorj, @n-miyamoto-fixstars, @LWisteria, and @matsui528)
  • Implementation of Local Search Quantization (by @KinglittleQ)

Changed

  • The order of xb an xq was different between faiss.knn and faiss.knn_gpu. Also the metric argument was called distance_type.
  • The typed vectors (LongVector, LongLongVector, etc.) of the SWIG interface have been deprecated. They have been replaced with Int32Vector, Int64Vector, etc. (by h-vetinari)

Fixed

  • Fixed a bug causing kNN search functions for IndexBinaryHash and IndexBinaryMultiHash to return results in a random order.
  • Copy constructor of AlignedTable had a bug leading to crashes when cloning IVFPQ indices.

1.7.0 - 2021-01-27

1.6.5 - 2020-11-22

1.6.4 - 2020-10-12

Added

  • Arbitrary dimensions per sub-quantizer now allowed for GpuIndexIVFPQ.
  • Brute-force kNN on GPU (bfKnn) now accepts int32 indices.
  • Nightly conda builds now available (for CPU).
  • Faiss is now supported on Windows.

1.6.3 - 2020-03-24

Added

Changed

  • Replaced obj table in Clustering object: now it is a ClusteringIterationStats structure that contains additional statistics.

Removed

  • Removed support for useFloat16Accumulator for accumulators on GPU (all accumulations are now done in float32, regardless of whether float16 or float32 input data is used).

Fixed

  • Some python3 fixes in benchmarks.
  • Fixed GpuCloner (some fields were not copied, default to no precomputed tables with IndexIVFPQ).
  • Fixed support for new pytorch versions.
  • Serialization bug with alternative distances.
  • Removed test on multiple-of-4 dimensions when switching between blas and AVX implementations.

1.6.2 - 2020-03-10

1.6.1 - 2019-12-04

1.6.0 - 2019-09-24

Added

  • Faiss as a codec: We introduce a new API within Faiss to encode fixed-size vectors into fixed-size codes. The encoding is lossy and the tradeoff between compression and reconstruction accuracy can be adjusted.
  • ScalarQuantizer support for GPU, see gpu/GpuIndexIVFScalarQuantizer.h. This is particularly useful as GPU memory is often less abundant than CPU.
  • Added easy-to-use serialization functions for indexes to byte arrays in Python (faiss.serialize_index, faiss.deserialize_index).
  • The Python KMeans object can be used to use the GPU directly, just add gpu=True to the constuctor see gpu/test/test_gpu_index.py test TestGPUKmeans.

Changed

  • Change in the code layout: many C++ sources are now in subdirectories impl/ and utils/.

1.5.3 - 2019-06-24

Added

Changed

  • Throw python exception for OOM (facebookresearch#758).
  • Make DistanceComputer available for all random access indexes.
  • Gradually moving from long to uint64_t for portability.

Fixed

1.5.2 - 2019-05-28

Added

  • Support for searching several inverted lists in parallel (parallel_mode != 0).
  • Better support for PQ codes where nbit != 8 or 16.
  • IVFSpectralHash implementation: spectral hash codes inside an IVF.
  • 6-bit per component scalar quantizer (4 and 8 bit were already supported).
  • Combinations of inverted lists: HStackInvertedLists and VStackInvertedLists.
  • Configurable number of threads for OnDiskInvertedLists prefetching (including 0=no prefetch).
  • More test and demo code compatible with Python 3 (print with parentheses).

Changed

  • License was changed from BSD+Patents to MIT.
  • Exceptions raised in sub-indexes of IndexShards and IndexReplicas are now propagated.
  • Refactored benchmark code: data loading is now in a single file.

1.5.1 - 2019-04-05

Added

  • MatrixStats object, which reports useful statistics about a dataset.
  • Option to round coordinates during k-means optimization.
  • An alternative option for search in HNSW.
  • Support for range search in IVFScalarQuantizer.
  • Support for direct uint_8 codec in ScalarQuantizer.
  • Better support for PQ code assignment with external index.
  • Support for IMI2x16 (4B virtual centroids).
  • Support for k = 2048 search on GPU (instead of 1024).
  • Support for renaming an ondisk invertedlists.
  • Support for nterrupting computations with interrupt signal (ctrl-C) in python.
  • Simplified build system (with --with-cuda/--with-cuda-arch options).

Changed

  • Moved stats() and imbalance_factor() from IndexIVF to InvertedLists object.
  • Renamed IndexProxy to IndexReplicas.
  • Most CUDA mem alloc failures now throw exceptions instead of terminating on an assertion.
  • Updated example Dockerfile.
  • Conda packages now depend on the cudatoolkit packages, which fixes some interferences with pytorch. Consequentially, faiss-gpu should now be installed by conda install -c pytorch faiss-gpu cudatoolkit=10.0.

1.5.0 - 2018-12-19

Added

  • New GpuIndexBinaryFlat index.
  • New IndexBinaryHNSW index.

1.4.0 - 2018-08-30

Added

  • Automatic tracking of C++ references in Python.
  • Support for non-intel platforms, some functions optimized for ARM.
  • Support for overriding nprobe for concurrent searches.
  • Support for floating-point quantizers in binary indices.

Fixed

  • No more segfaults due to Python's GC.
  • GpuIndexIVFFlat issues for float32 with 64 / 128 dims.
  • Sharding of flat indexes on GPU with index_cpu_to_gpu_multiple.

1.3.0 - 2018-07-10

Added

  • Support for binary indexes (IndexBinaryFlat, IndexBinaryIVF).
  • Support fp16 encoding in scalar quantizer.
  • Support for deduplication in IndexIVFFlat.
  • Support for index serialization.

Fixed

  • MMAP bug for normal indices.
  • Propagation of io_flags in read func.
  • k-selection for CUDA 9.
  • Race condition in OnDiskInvertedLists.

1.2.1 - 2018-02-28

Added

  • Support for on-disk storage of IndexIVF data.
  • C bindings.
  • Extended tutorial to GPU indices.