Installing older version of Pytorch

I would like to install Pytorch 0.4 and Cuda 9.0 in Jetson TX2. However, none of the JetPacks comes with these dependencies. As far as I have searched, I couldn’t find the corresponding wheel as well. conda and pip do not have torch0,4 for Jetson TX2 as well. How can I install the older version of Torch without building from source?

Hi @linuxfs, JetPack 3.3 used CUDA 9.0. We didn’t build PyTorch wheels dating back to PyTorch v0.4, so you would need to build it from source.

However, in my experience PyTorch has largely been backwards compatible since PyTorch v0.4 and newer. Is there a reason not to use a newer version of PyTorch and JetPack?

The code that I was trying has an explicit dependency for the same hence I was trying to install torch 0.4. Is there any pip or wheel for torch 1.0?

Yes, there is PyTorch v1.0.0 pip wheel for JetPack 4.2/4.3 on this page:

https://forums.developer.nvidia.com/t/pytorch-for-jetson-nano-version-1-5-0-now-available/72048

Here are the links:

PyTorch v1.0.0 - JetPack 4.2 / 4.3