Skip to content

Version 0.1.5

Choose a tag to compare

@rossarmstrong rossarmstrong released this 02 Jun 05:45
· 9 commits to main since this release
v0.1.5
727141a

Added

  • Unified cosine_similarity API: now automatically dispatches to SIMD or batch implementation based on batch_size.
  • New benchmark script for streaming real embedding datasets from Hugging Face Hub.
  • Benchmarks and compares cosine similarity performance across numpy, sklearn, and symrank APIs.
  • Enables large-scale, realistic evaluation of similarity search performance.

Changed

  • Updated README with clearer usage examples and parameter descriptions.

Fixed

  • Improved type hints and docstrings for all public APIs.
  • Cleaned up __init__.py to only expose the main APIs: cosine_similarity, cosine_similarity_batch, cosine_similarity_simd.
  • Consistent handling of batch_size=None and batch_size=0 (both now mean "no batching").