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

Enable GPU support for generic Windows job #4104

Merged
merged 11 commits into from
May 8, 2023
Merged

Enable GPU support for generic Windows job #4104

merged 11 commits into from
May 8, 2023

Conversation

pmeier
Copy link
Contributor

@pmeier pmeier commented May 4, 2023

The Linux job has these inputs to control the GPU support:

gpu-arch-type:
description: "GPU arch type to use"
default: "cpu"
type: string
gpu-arch-version:
description: "GPU arch version to use"
default: ""
type: string

They were missing before from the Windows job. Without them, users are forced to use workarounds.

This PR adds them to the Windows job as well. Since we have multiple versions of CUDA already present on the AMI, we only need to set some paths.

@pmeier pmeier requested a review from atalman May 4, 2023 14:00
@vercel
Copy link

vercel bot commented May 4, 2023

The latest updates on your projects. Learn more about Vercel for Git ↗︎

1 Ignored Deployment
Name Status Preview Comments Updated (UTC)
torchci ⬜️ Ignored (Inspect) May 8, 2023 6:36pm

@facebook-github-bot facebook-github-bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label May 4, 2023
Copy link
Contributor Author

@pmeier pmeier left a comment

Choose a reason for hiding this comment

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

Corresponding vision PR: pytorch/vision#7555

@@ -126,12 +135,20 @@ jobs:
echo "/c/Jenkins/Miniconda3/python.exe -m pip install --upgrade pywin32==300"
# Source conda so it's available to the script environment
echo "source /c/Jenkins/Miniconda3/etc/profile.d/conda.sh";
echo "${SCRIPT}";
} > "${RUNNER_TEMP}/exec_script"
echo 'export PATH="/c/Jenkins/Miniconda3/Scripts:${PATH}"'
Copy link
Contributor Author

Choose a reason for hiding this comment

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

timeout: 60
script: |
conda create --yes --quiet -n test python=3.8
conda activate test
python -m pip install --extra-index-url https://download.pytorch.org/whl/nightly/cu116 --pre torch
python -m pip install ---index-url=https://download.pytorch.org/whl/nightly/cu117 --pre torch
Copy link
Contributor Author

Choose a reason for hiding this comment

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

Problem here was that the CUDA 11.6 index no longer has any torch binaries. Since we additionally used --extra-index-url, we pulled the wheel from PyPI: https://github.com/pytorch/test-infra/actions/runs/4883695947/jobs/8715408340?pr=4104#step:9:86

And the Windows wheel on PyPI is not built against CUDA.

This is not related to this PR other than this PR triggered this test.

@pmeier pmeier merged commit 19cf519 into main May 8, 2023
@pmeier pmeier deleted the win-gpu branch May 8, 2023 19:06
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants