SAVERouter v0.1.0

Initial public release of SAVERouter.

Highlights:

• Fixed-K sparse supervision acquisition for economical LLM routing.
• Hierarchical capability estimation with query-level refinement.
• SA-BEP and SA-CR metrics for supervision-amortized evaluation.
• Frozen paper profiles for LLMRouterBench, Mixinstruct, MMR-Bench, and RouterBench.
• Download-free smoke test, unit tests, and reproducibility CLI.

Install:
python -m pip install saverouter
saverouter smoke-test

PyPI: https://pypi.org/project/saverouter/
Project page: https://lamda-model-reuse.github.io/SaveRouter/
Paper: https://arxiv.org/abs/2609.37402
Reproduction instructions: https://github.com/LAMDA-Model-Reuse/SaveRouter#reproducing-the-experiments

The attached wheel and source distribution are built and validated automatically from this release tag.