v0.9.1

Changelog

Features 🏎️

• #805 - run the minishlab potion models with numpy instead of ONNX, 1.9x to 2.6x faster on documents with unchanged outputs by @joein
• #806 - speed up embedding a single text, such as a query, with the potion models by @joein
• #808 - add an mmap option for the potion models to memory-map the embedding matrix: it loads instantly and lets parallel workers share one copy by @joein

Fixes 🔧

• #794 - raise a ValueError when a custom model with PoolingType.DISABLED returns per-token output instead of sentence embeddings by @joein
• #795 - reject custom cross-encoder names that differ from a registered one only by case, which silently resolved to the earlier model by @wToTw123
• #830 - allow Pillow 13 by @svaningelgem

Other

• #819 - switch the project to uv by @tbung
• fastembed-gpu 0.9.1 is released alongside, with the same changes.