v0.1.7 v0.1.8

medium risk

minitap-ai/mobile-useMar 23, 2026 → Mar 23, 20260 days between releases

What's New

Reranking speed increased 3.3x via single-pass and multi-threading.

New Features

  • Single rerank pass logic to eliminate redundant ONNX session loads.
  • ORT format support with automatic fallback to .onnx.
  • Zero-copy memory-mapped file access for model loading.
  • Multi-threaded inference utilizing up to 4 cores.

Improvements

  • Reranking performance improved 3.3x across all packages.
  • Faster deserialization using pre-optimized flatbuffer models.
  • Reduced redundant ONNX session initialization overhead.

Upgrade Notes

No configuration changes are required. The library automatically handles the new ORT format with fallback to .onnx. Internal processing logic has changed to a single rerank pass.

The Two Releases

v0.1.7

v0.1.7

Mar 23, 2026· 7 months ago

Replaces HuggingFace tokenizers with `tokie` to improve speed and reduce dependencies.

Full release notes
v0.1.8

v0.1.8

Mar 23, 2026· 7 months ago

Significantly improves reranking performance through multi-threading and optimized model formats.

Full release notes