v0.2.1

0.2.1 (2023-05-17)

Make

• Build (8429fe6)

v0.2.0

0.2.0 (2023-05-17)

Add

• HashIndex for Python (62a97a9)
• SetsIndex and FAISS IVFPQ example in Py (a7a6525)
• sorted_buffer_gt to replace one PQ (e6f7fcc)
• Automatic thread selection for Go/Java/Rust (63ab338)
• Conan recipe for C++ users (0085346)
• Heap copy/move constructors (c82f500)
• LICENSE (889f550)
• Logging more metadata on benchmarks (6d9dcb5)
• Memory-mapping files in Python (519f1c7)
• Native f16 conversion (3fb4a04)
• Non-bitwise Hamming distance (98fed63)
• Objective-C and Swift bindings (154c809)
• PyTest with 77 configurations (7c8268b)
• Serial code for f8 quantization (0ecdd1f)
• Single search and add operations in Python (e07b04d)
• Standalone server prototype (8c8c2df)
• Support custom accumulator in default metrics (737d919)
• User-defined metrics in Python (0a10aac)
• Wolfram Interface (5134713)

AdD

• Quarter precision benchmarks (1b5c7cd)

Build

• Bump version (e3c82fc)
• Conditional SIMD for compatibility (b259baf)
• Disabling warning on Release builds (9e9e8c3)
• Drop gradle scripts. (8d296c9)
• Drop NOMINMAX (d58f684)
• More options for CMake builds (3bddddf)
• Remove binding.gyp to avoid conflicts with semantic-versioning. (eaca40f)
• Remove launch.json version update (e2d9d46)
• Set default wolfram install dir (1b0055c)

Chore

• Formatting (96cd3ec)

Docs

• Add LAION and Deep1B download links (fdc35e1)
• Add links (d016909)
• Added images (785a2b8)
• Fully reorganized README (aa10801)
• Listing 1B datasets (0cda068)
• Markdown formatting (0a1fce3)
• Reordered intro (7a8e3db)
• Simplified Python examples (eaede82)
• Update benchmarking instructions (21bd286)
• Update Intro (97d3dc6)
• Update python docs (6e71ece)
• Update subtitle (4480e62)
• Wording (f976b13)

Fix

• auto_index_gt mutability (d5423ba)
• npm build on windows (043bcad)
• SetsIndex.add call, add clear() (2e6f39f)
• Compilation errors with Python (9469659)
• Counting number of tasks per thread (7cd807d)
• File compatibility checks (46db7bb)
• Forking and memory alignment bug (a7343ff)
• Guarding Float16 Arm-specific functions (2fa3730)
• IndexSearch to set return value. (647c5e1)
• Inverting angular distance (3e8f18f), closes #13
• IP error compensation with normalization (33d2343)
• Minimal alignment on windows. (6b5e582)
• Multi-threaded scheduling (803efa1)
• Name resolution in Python test (d222a19)
• Negative selection on filtering (049258b)
• Quantization and casting (9077f9d)
• Removed heap.max() function for Windows (84467b0)
• Saving whole vector to disk (6833bd8)
• Support pure-python submodules. (15eb5eb)
• User defined functions and OpenMP (0dbc327)

Make

• Bump version (7a24b82)
• Bump version (9a810e2)
• Bump version (8524077)
• Bump version (bab111d)
• Enable building python for windows on github (1baee9a)
• Fix Windows Build. (3aef873)
• Link Python with OpenMP on Linux (3355140)
• Objective-C interface availability (37f03db)
• Require recent MacOS for aligned_alloc (7210677)
• Revert failing Windows builds (5261f31)
• Support Java build on Windows (4811b81)
• Support python build on windows. (aeabc6f)
• Use .releaserc for semantic-release ([a963a3b](https://github.com/un...

v0.1.8: C++11, Python, JavaScript, Rust, Java

We are excited to share a single-header C++11 vector search engine with everyone! It's fast, reaching 300'000 operations per second on 64-core machines, and supports f16_t and integer quantization, as well as 1T point indexes out of the box! Give it a try!