4 ms·
Correct. You need CUDA, or ROCm, MPS (native to macOS) backends for running deep learning. I found it relatively easy to train some Pytorch model on beefy serve
by thangngoc89 3y ago
Correct. You need CUDA, or ROCm, MPS (native to macOS) backends for running deep learning. I found it relatively easy to train some Pytorch model on beefy server with CUDA and running interference on my Macbook Air.
- astrange 3y agoMPS is a Metal shader library rather than a programming language, which would be MSL (like GLSL/HLSL).
- simlevesque 3y agoThey're referring to this: https://pytorch.org/docs/stable/notes/mps.html https://pytorch.org/docs/stable/notes/mps.html