PyTorch Distributed

[AliceW]

Is there any way to get the distributed PyTorch running or do I need to build it from source? I installed the latest version from the documentation at Installing PyTorch for Jetson Platform - NVIDIA Docs but now I get

>>> import torch
>>> torch.distributed.is_available()
False
>>>

When I try to install PyTorch from the torch website then CUDA becomes unavailable.

[AliceW]

Also in that documentation the whl file download URL gives a 404.

[dusty_nv]

IIRC the last pre-built PyTorch wheel for JetPack with USE_DISTRIBUTED was PyTorch 1.11:

PyTorch v1.11.0

• JetPack 5.0 (L4T R34.1.0) / JetPack 5.0.1 (L4T R34.1.1) / JetPack 5.0.2 (L4T R35.1.0)
• Python 3.8 - torch-1.11.0-cp38-cp38-linux_aarch64.whl

For newer PyTorch versions that that, you would need to build it from source (instructions are in this source)

[AliceW]

Ah that sucks. I’ll try to build the latest version for CUDA 12.1 with all the distributed stuff enabled then.