After installing PyTorch, a user will be able to import functorch and use functorch without needing to install another package. Previously, functorch was released out-of-tree in a separate package. Timely deprecating older CUDA versions allows us to proceed with introducing the latest CUDA version as they are introduced by Nvidia®, and hence allows support for C++17 in PyTorch and new NVIDIA Open GPU Kernel Modules. Additional improvements include accelerated add+matmul linear algebra kernels for sizes commonly used in Transformer models and Nested Tensors is now enabled by default. The BetterTransformer feature set supports fastpath execution for common Transformer models during Inference out-of-the-box, without the need to modify the model. We want to sincerely thank our dedicated community for your contributions. This release is composed of over 3,749 commits and 467 contributors since 1.12.1. Beta includes improved support for Apple M1 chips and functorch, a library that offers composable vmap (vectorization) and autodiff transforms, being included in-tree with the PyTorch release. We deprecated CUDA 10.2 and 11.3 and completed migration of CUDA 11.6 and 11.7. We are excited to announce the release of PyTorch 1.13! This includes stable versions of BetterTransformer. The release tracker should contain all relevant pull requests related to this release as well as links to related issues PyTorch 1.13: beta versions of functorch and improved support for Apple’s new M1 chips are now available Pytorch 1.13 Release Notes torchrun AttributeError caused by file_based_local_timer on Windows #85427.Onnx operator_export_type on the new registry #87735.Onnx ATen Fallback for BUILD_CAFFE2=0 for ONNX-only ops #88504.

Fix cuda/cpu check on NoneType and unit test #88854 and #88970.

Fix logical error in get_default_qat_qconfig #88876.

Double-backward with full_backward_hook causes RuntimeError #88312.

Use the Python frame safely in _pythonCallstack #88993.

_type_line is not safe (command injection) #88868.

Missing masked tensor documentation #89734.

Installation using poetry on Mac M1, failure #88049.

Installation via pip on Amazon Linux 2, regression #88869.

RuntimeError by torch.nn. with bias=False and batch_first=True #88669.

This release is meant to fix the following issues (regressions / silent correctness): PyTorch 1.13.1 Release, small bug fix release

Label matrix 7 software