v0.5.17

This release adds newer Qwen models to the candidate list and includes general reasoning models in the math profile.

It also fixes crashes when reading non-UTF-8 caches, respects the configured Apple GPU memory limit, and corrects bandwidth lookup for Ada laptop workstation GPUs.

Ranking helpers have been split into smaller modules, and the README star-history chart works again.

Update with uv tool upgrade whichllm or pip install --upgrade whichllm.

Full changelog