Skip to content

Release 2.2.2

Latest
Latest

Choose a tag to compare

View all tags
@markus583 markus583 released this 25 Sep 15:44
2.2.2
3ef928c
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

What's Changed

  • Faster PyTorch inference (#178): SaT.optimize() and WtP.optimize() compile a PyTorch checkpoint with torch.compile (TorchInductor by default). Call it after to() / half(). Not available for ONNX (ort_providers).
  • NVIDIA AITune (#178): optimize(backend="aitune") benchmarks CUDA backends and picks one. Optional extra: pip install wtpsplit[aitune] --extra-index-url https://pypi.nvidia.com.
  • LoRA training imports on transformers 4.29.2 (#180): fall back to is_torch_tpu_available and transformers.deepspeed when the newer names are missing.
  • WtP Canine on transformers 5: LACanineModel no longer calls the removed get_extended_attention_mask.
  • CI: pin Ruff to the pre-0.16 rule set so the lint job does not enable 413 new rules.

Full Changelog: 2.2.1...2.2.2

Assets 2
Loading