PyTorch 2.10 released

PyTorch, a popular machine learning framework, has been updated to version 2.10.

Among the main innovations we can note:

• AMD ROCm Support:

  • Introduced support for grouped GEMM.
  • Improved ROCm support for Windows.
  • Added new GFX1150/GFX1151 (AI 300 series) GPU models to hipblaslt support lists.
  • Enhanced functionality such as support for scaled_mm v2 and AOTriton scaled_dot_product_attention.

• Intel GPU Support:

  • New Torch XPU APIs for Intel have been implemented.
  • Support for additional ATen operators.
  • Performance optimization for Intel GPUs.

• NVIDIA CUDA Support:

  • Advanced capabilities for writing template kernels.
  • Improved support for CUDA 13.

• Python 3.14 support for torch.compile(), as well as experimental support for building Python 3.14 without a global interpreter lock (free-threaded).

• Reduced core startup overhead thanks to horizontal fusion of combo cores in Torch Inductor.

The full list of changes is available at github.

Ready-made versions compiled for different GPUs, OS and languages ​​(Python/C++/Java) are listed at Pytorch website

Source: linux.org.ru