Releases: llv22/pytorch-macOS-cuda
macOS10.13.6-xcode10.1-torch2.2.1
Full Changelog: macOS10.13.6-xcode10.1-torch2.2.0...macOS10.13.6-xcode10.1-torch2.2.1
merge from pytorch/pytorch@v2.2.0...v2.2.1. The other libraries keep as the same to https://github.com/llv22/pytorch-macOS-cuda/releases/tag/macOS10.13.6-xcode10.1-torch2.2.0. so I just upload torch 2.2.1 library.
Full Changelog: macOS10.13.6-xcode10.1-torch2.2.1...macOS10.13.6-xcode10.1-torch2.2.r1
macOS10.13.6-xcode10.1-torch2.2.0
Full Changelog: https://github.com/llv22/pytorch-macOS-cuda/commits/macOS10.13.6-xcode10.1-torch2.2.0
In order to keep up-to-date with the latest release, upgrade for the same environment. Now I setup a mixed CUDA 10.2 with CUDA 10.1 and found the libraires are compatible with each other. I upload all torch libraries for your reference.
torch 2.0.0 GPU on macOS 10.13.6 with Xcode 10.1
In order to keep up-to-date with the latest release, upgrade for the same environment. Now I setup a mixed CUDA 10.2 with CUDA 10.1 and found the libraires are compatible with each other. That means we can use cuda memory control API used for torch 2.2.0 now, although compiling torch 2.2.0 has met with one issue of unsolved issues "default argument from c++ to python object pybind11".
As torch-2.2.0-fixed branch still has issues about default argument, this serves as our latest release library version.