Skip to content

v0.6.0: Optimized kernels and upgraded dependencies for enhanced performance in PyTorch 2.4+

Choose a tag to compare

View all tags
@zhangyuxuann zhangyuxuann released this 19 Aug 07:35
· 106 commits to main since this release
v0.6.0
a40f8f6

What's Changed

  1. Optimized the custom LayerNorm kernel, further boosting end-to-end inference and training speed.
  2. Integrated a custom Triton-based implementation of the TriangleAttention operator (triattention), improving computational efficiency.
  3. Integrated the cuEquivariance operator from NVIDIA/cuEquivariance to accelerate equivariant operations, with notable efficiency gains in the TriangleAttention and TriangleMultiplication modules.
  4. Upgraded the container image and dependencies to resolve efficiency bottlenecks in PyTorch 2.4 and later versions; Supported Biotite 1.2 and above.

New Contributors

Full Changelog: v0.5.5...v0.6.0

Contributors

chaitjo
Assets 2
Loading