• Upgrade to transformers v5 and fix broken tests associated with this upgrade #180

• Minor updates to DMSPress and KVzapPress #177
• Speedup of the CI/CD #177

• Rename ThresholdPress to DMSPress (#174)

✨ New Features

• KVzapPress - a fast approximation of KVzip for prefill and decoding compression (https://arxiv.org/abs/2601.07891). Comes with KVzap training and evaluation utilities (#171)
• ThresholdPress - adaptive compression using score thresholds instead of fixed compression ratios (#171)

📈 Improvements

• Update KVzipPress with improvements and evaluation registry support (#172)
• Rename compress-question to query-aware in evaluation config (#168)
• Refactor ObservedAttentionPress for cleaner implementation (#166)
• Add leaderboard generation script (#171)

🐛 Bug Fixes

• Fix empty context handling in pipeline (#165)

🚀 Release v0.4.0

✨ New Features

• CURPress - Value-Guided KV Compression for LLMs via Approximated CUR Decomposition (#150)
• CompactorPress - Compactor: Calibrated Query-Agnostic KV Cache Compression with Approximate Leverage Scores (#143)
• Decoding Press Functionality - Support for KV cache compression during the decoding phase (#139)
• AIME25 & Math500 Benchmarks - New evaluation datasets for mathematical reasoning tasks (#142)
• post_init_from_model Hook - Add model-specific initialization support in BasePress (#163)

📈 Improvements

• Moved tests to GPU for faster CI execution (#132)
• Improved needle-in-haystack test coverage (#133)
• Updated README and documentation for clarity (#162)
• Enhanced docstrings throughout the codebase (#159)
• Updated decoding notebook with latest examples (#156)
• Code cleanup: moved utilities, cleaned imports (#160)

🐛 Bug Fixes

• Fixed LongBench-v2 benchmark evaluation (#161)
• Fixed kvzip press access to past_key_values
• Fixed ComposedPress behavior (#148)
• Fixed import issues (#144)

📦 Installation

pip install kvpress==0.4.0

📚 Full Changelog

v0.3.0...v0.4.0

What's Changed

• refactor: optimized covariance transform in ExpectedAttentionPress by @neuralsorcerer in #111
• fix ruler integration tests by @maxjeblick in #113
• fix typo by @neuralsorcerer in #116
• Add needle in haystack test by @alessiodevoto in #121
• fix masked_key_indices by @maxjeblick in #122
• Add copy-pr-bot settings by @maxjeblick in #123
• Add Github runner by @maxjeblick in #124
• evaluation README.md command error and logging error #127 by @wzp-0815 in #128
• add gpu runner by @maxjeblick in #125
• Upgrade expected attention with support for more models by @alessiodevoto in #126
• Add Expected Attention with Stats by @alessiodevoto in #120
• ⚠️ Transformers compatibility by @maxjeblick in #115 ---> this is a breaking change (the KV caching machinery changed in HF transformers and we adjusted KVPress accordingly)

New Contributors

• @neuralsorcerer made their first contribution in #111
• @wzp-0815 made their first contribution in #128

Full Changelog: v0.2.10...v0.3.0

What's Changed

• Migration to uv by @alessiodevoto in #108

Full Changelog: v0.2.9...v0.2.10

What's Changed

• Refactor evaluation by @alessiodevoto in #96
• Fix QFilters and DuotAttention when used with wrapper presses by @alessiodevoto in #97
• Add HuggingFace leaderboard by @alessiodevoto in #98
• Fix links in benchmarks directory by @alessiodevoto in #101
• Add KVzipPress by @Janghyun1230 in #93
• Test head-wise compression by @alessiodevoto in #103
• run backbone model only for prefill by @giulio98 in #100
• Transformers compatibility + evaluation by @alessiodevoto in #105

Full Changelog: v0.2.8...v0.2.9

What's Changed

🐛 Bug Fixes

• Fix failing tests by @maxjeblick in #94
  Reverts changes to CriticalKVPress performed in #90 that caused the press to initialize incorrectly. The PR also fixes some test logic.

Full Changelog: v0.2.7...v0.2.8

What's Changed

🐛 Bug Fixes

• Fix FinchPress for Qwen models family by @alessiodevoto in #82
  Resolved compatibility issues with Qwen model architecture in FinchPress compression

✨ New Features

• Add KeyDiffPress and BlockPress by @figuremout in #86
  Introduces new compression methods based on key difference analysis
• Fix for Qwen with Yarn by @giulio98 in #85
  Enable Yarn scaling in FinchPress and KeyRerotationPress

📚 Documentation & Maintenance

• Improve documentation by @maxjeblick in #90
  Add docstrings to all presses, with their corresponding parameters and paper reference.
• Add @alessiodevoto's to authors by @maxjeblick in #92 🚀

Full Changelog: v0.2.6...v0.2.7