Release v0.4.0

- Add CURPress - CUR decomposition-based KV cache compression (#150)
- Add Compactor press for enhanced compression capabilities (#143)
- Add decoding press functionality for compression during decoding (#139)
- Add AIME25 and Math500 benchmark datasets for evaluation (#142)
- Add post_init_from_model hook to BasePress for model-specific initialization (#163)

- Move tests to GPU for faster CI (#132)
- Improve needle-in-haystack test (#133)
- Update README and documentation (#162)
- Update docstrings (#159)
- Update decoding notebook (#156)
- Move utils, clean and fix imports (#160)

- Fix LongBench-v2 benchmark (#161)
- Fix kvzip press access to past_key_values
- Fix ComposedPress (#148)
- Fix imports (#144)

Transformers compatibility (#115)

Migration to uv (#108)

Fix trandformers<4.54.0

Fix failing tests (#94)

add Alessio to authors (#92)

Support Qwen3 and Gemma3 (#81)

Add FinchPress (#69)

Add QFilterPress (#54)

Add DuoAttentionPress (#50)

* Add DuoAttentionPress

* Fix tests and compression_ratio

* Address feedback

* Update plot

* Update version