Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Increase minimum PyTorch requirements to 1.2.0 #89

Conversation

Jean-Francois-Lafleche
Copy link
Contributor

@Jean-Francois-Lafleche Jean-Francois-Lafleche commented Dec 11, 2019

Description

  • Increase minimum PyTorch requirements to 1.2.0
  • Add minimum requirement for CUDA to 10.0.130
  • Add minimum display driver version to 410.48 (minimum version for CUDA 10.0.130)
  • Add minimum PyTorch version assert in setup.py

Related Issues

#65

Testing PyTorch versions (before changes)

PyTorch 0.4.1, CUDA 9

  • Docker Container: pytorch/pytorch:0.4.1-cuda9-cudnn7-devel
    • python setup.py develop (fatal error: torch/extension.h: No such file or directory)

PyTorch 1.0, CUDA 10.0

  • Docker Container: pytorch/pytorch:1.0-cuda10.0-cudnn7-devel
    • python setup.py develop (error: 'TORCH_CHECK' was not declared in this scope)

PyTorch 1.1.0, CUDA 10.0

  • Docker Container: pytorch/pytorch:1.1.0-cuda10.0-cudnn7.5-devel
    • python setup.py develop (error: 'TORCH_CHECK' was not declared in this scope)

PyTorch 1.2, CUDA 9.2

  • Docker Container: nvidia/cuda:9.2-base-ubuntu16.04

PyTorch 1.2, CUDA 10.0

  • Docker Container: pytorch/pytorch:1.2-cuda10.0-cudnn7-devel
    • python setup.py develop (successful build)

Tests (after changes)

  • Docker Container: pytorch/pytorch:1.2-cuda10.0-cudnn7-devel
    • python setup.py develop (successful build)
  • Docker Container: pytorch/pytorch:1.0.1-cuda10.0-cudnn7-devel
    • python setup.py develop (Warning logged)

@Caenorst
Copy link
Collaborator

Why are we testing pytorch/pytorch:1.0.1-cuda10.0-cudnn7-devel if Pytorch 1.2 is the new minimum requirements ?

@Jean-Francois-Lafleche
Copy link
Contributor Author

@Caenorst Testing pytorch/pytorch:1.0.1-cuda10.0-cudnn7-devel is to confirm that setup will fail with an assert error message if pytorch < 1.2 is installed.

@Caenorst
Copy link
Collaborator

Do we actually want to have an assertion ? Why not just having a warning message ? I agree that we don't wanna officially support old version of Pytorch / CUDA / python but should we explicitly make the installation crash while it could potentially works ?

@Jean-Francois-Lafleche
Copy link
Contributor Author

I agree. A warning is better than crashing the installation. Fixed.

Copy link
Collaborator

@Caenorst Caenorst left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

just a little comment to remove and we are good to go. Thanks !

setup.py Outdated Show resolved Hide resolved
@Jean-Francois-Lafleche Jean-Francois-Lafleche merged commit 5a56696 into NVIDIAGameWorks:master Dec 16, 2019
@Jean-Francois-Lafleche Jean-Francois-Lafleche deleted the pytorch-version branch December 16, 2019 16:15
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

2 participants