0.1.0 — Unreleased
Initial release candidate.
- Load supported SentenceTransformers module pipelines from Hugging Face or local checkpoints.
- Encode dense, SPLADE/static sparse, and multi-vector representations through Nx and Bumblebee.
- Preserve saved prompts and sequence limits, with runtime prompt overrides and compilation buckets.
- Provide chunked search, embedding-level mining, similarity, quantization, and Matryoshka truncation utilities.
- Compare supported inference paths against SentenceTransformers 6.1.0 using generated checkpoints and public-model integration tests.
See the parity guide for the tested subset, API differences, and initial-release limitations.