pruna

The juiciest bits 🧃

New algorithms joining the garden 🌱

We added a fresh batch of algorithm support this release, and this one deserves an extra round of applause: all of these new algorithm additions came from external contributors! Huge thank you to @krishjp, @rensortino, and @kschwethelm for helping expand what Pruna can optimize and for bringing more community-built improvements into the repo

• feat: integrate Llama.cpp and enhance engine stability for cross-platform usage by @krishjp in https://github.com/PrunaAI/pruna/pull/616
• feat: token merging for image classification by @rensortino in https://github.com/PrunaAI/pruna/pull/537
• feat: integrate KVPress for KV cache compression (#366) by @kschwethelm in https://github.com/PrunaAI/pruna/pull/623

Compatibility updates and pruning some old branches: ✂️

Transformers 5 compatibility is here, TorchAO support now goes up to 0.17.0, and stable-fast wheel handling has been updated for newer Torch versions. At the same time, we cleaned up outdated or unreachable dependencies to keep installs smoother and more reliable:

• feat: compatibility with transformers 5 by @sdiazlor in https://github.com/PrunaAI/pruna/pull/685
• build: bump max torchao from 0.16.0 to 0.17.0 by @gsprochette in https://github.com/PrunaAI/pruna/pull/692
• fix: drop ipex support because their index is unreachable by @gsprochette in https://github.com/PrunaAI/pruna/pull/667
• fix: drop outdated whisper-s2t by @gsprochette in https://github.com/PrunaAI/pruna/pull/689

Some Bug Fixing 🐞 and Maintenance

A lot of work went into making Pruna more stable across backends, datasets, compilation flows, and test environments:

• fix: torch compile default to inductor on cpu by @gsprochette in https://github.com/PrunaAI/pruna/pull/673
• fix: torch compile bugs hqq by @gsprochette in https://github.com/PrunaAI/pruna/pull/668
• fix: patch torchmetrics (transformers 5) by @sdiazlor in https://github.com/PrunaAI/pruna/pull/683
• fix: update moe tiny model for testing (transformers 5) by @sdiazlor in https://github.com/PrunaAI/pruna/pull/684
• fix: stop auto decoding for audio datasets by @begumcig in https://github.com/PrunaAI/pruna/pull/686
• style: move type ignore to its correct place by @gsprochette in https://github.com/PrunaAI/pruna/pull/681
• test: fix breaking nightly tests for ONEIG dataset by @begumcig in https://github.com/PrunaAI/pruna/pull/687
• fix: add kernel version by @sdiazlor in https://github.com/PrunaAI/pruna/pull/675
• fix: device casting for llama cpp by @begumcig in https://github.com/PrunaAI/pruna/pull/693
• fix: stop audio decoding by @begumcig in https://github.com/PrunaAI/pruna/pull/694
• fix: switch torch_dynamic and quanto saves to save_before_apply by @gsprochette in https://github.com/PrunaAI/pruna/pull/696

We’ve also continued improving the developer experience around CI, dependency handling, pre-commit checks, and test stability:

• ci: add no extra flag by @begumcig in https://github.com/PrunaAI/pruna/pull/653
• ci: automatically assign new reviewers by @sdiazlor in https://github.com/PrunaAI/pruna/pull/620
• build: add dependencies cooldown by @gsprochette in https://github.com/PrunaAI/pruna/pull/660
• fix: ruff check pre-commit fails due to "Failed to parse pyproject.toml" by @psyberck in https://github.com/PrunaAI/pruna/pull/661
• build: restrict dependency cooldown to pypi index by @gsprochette in https://github.com/PrunaAI/pruna/pull/672
• build: switch back to unsafe-best-match for easy torch install by @gsprochette in https://github.com/PrunaAI/pruna/pull/674
• ci: add stale issue handling by @begumcig in https://github.com/PrunaAI/pruna/pull/682
• refactor: add requires stable fast mark by @begumcig in https://github.com/PrunaAI/pruna/pull/688
• ci: change uv sync to uv pip install by @gsprochette in https://github.com/PrunaAI/pruna/pull/691
• docs: remove reference to pruna[full] by @gsprochette in https://github.com/PrunaAI/pruna/pull/695

Documentation updates 📚

We cleaned up and expanded contribution docs so it’s easier to understand how to add extras, manage dependencies, and contribute safely:

• docs: add contribution guidlines for extras by @begumcig in https://github.com/PrunaAI/pruna/pull/654
• docs: removed first prune from PR template by @minettekaum in https://github.com/PrunaAI/pruna/pull/652
• docs: update with AI Contribution Policy by @sdiazlor in https://github.com/PrunaAI/pruna/pull/621
• docs: improve documentation for adding dependencies by @gsprochette in https://github.com/PrunaAI/pruna/pull/659

New Faces in the Garden 👩‍🌾

• @kschwethelm made their first contribution in https://github.com/PrunaAI/pruna/pull/623
• @psyberck made their first contribution in https://github.com/PrunaAI/pruna/pull/661
• @krishjp made their first contribution in https://github.com/PrunaAI/pruna/pull/616

Full Changelog: https://github.com/PrunaAI/pruna/compare/v0.3.3...v0.3.4