How to install/update to PyTorch 1.0(stable) with CUDA9.2?

1

Hi~
It’s great to hear PyTorch release a stable 1.0 version today! But when I was going to update it, I noticed on the official website, no 9.2 is written in CUDA’s supporting row, only 8.0, 9.0 10.0 and None. What I’m using currently is PyTorch 0.4.1 with CUDA 9.2 on Windows 10, and I remembered there was a conda installation command of PyTorch with CUDA 9.2 on the official website when it’s version was 0.4.1.
So I am wondering, how to install or update to PyTorch 1.0 with CUDA9.2 ?
Thanks in advance!:smile:

1 Like
2

Hi,

The problem is that each binary is (very) time consuming to do and the process needs to be done from scratch for every CUDA version.
So the decision has been made to support only one version of CUDA for all the 9.X series.

If you stay with CUDA 9.2, I’m afraid you will have to compile from source at the moment :confused:

1 Like
3

Thanks! I think to downgrade my CUDA version to 9.0 is a better choice.

1 Like
4

@albanD can you help me out on how to install PyTorch on aarch64 architecture based armv8 dev board

5

@smth do we support pytorch arm builds? Or is that only for caffe2?

6

@Amrit_Das it should work just fine if you use: https://github.com/pytorch/pytorch#from-source

I tried it on NVIDIA AGX Xavier which has ARM 64 bit processor, using these commands: https://gist.github.com/soumith/45f61db3379db67414814ad9d5f2a427

1 Like