Adding one B200 runner (8xB200 GPUs) to PyTorch CI

[nWEIdia]

Following the steps described in https://github.com/pytorch/test-infra/blob/main/docs/partners_pytorch_ci_runners.md, I'm creating this issue as a support ticket to add B200 runner to pytorch CI.

Current blockers include:

1. After reboot, the runner encountered "Runner connect error: Registration was not found or is not medium trust. ClientType: . Retrying until reconnected."
2. the jobs are encountering "aws command not found" issues.

cc @atalman @malfet @seemethere @ptrblck @tinglvv @eqy

[huydhn]

For the second issue, I think we need to refactor setup-linux job to take into account that B200 is not on AWS. As an example, we need to do the same as https://github.com/pytorch/pytorch/blob/main/.github/workflows/_rocm-test.yml#L91-L102 where, instead of using awscli to login to ECR (to get the Docker image), use aws-actions/configure-aws-credentials and aws-actions/amazon-ecr-login instead.

@nWEIdia Do you want to pick up that tasks or I could prepare a PR for this and rebase your PR pytorch/pytorch#157153 on top after it lands?. I think it's better that I do it instead because I suspect that a change to https://github.com/pytorch-labs/pytorch-gha-infra is needed and that repo is private to Dev Infra. Stay tuned

[nWEIdia]

It would be best if you can do it, thank you @huydhn !

[huydhn]

@nWEIdia I'm trying to test my change in pytorch/pytorch#157341, but it's waiting for the runner to come online https://github.com/pytorch/pytorch/actions/runs/15987382516/job/45100590002?pr=157341. Could you help add the runner on your end? I guess the runner is down because of the first issue you mention, so let's see if we could tackle it too

[nWEIdia]

@nWEIdia I'm trying to test my change in pytorch/pytorch#157341, but it's waiting for the runner to come online https://github.com/pytorch/pytorch/actions/runs/15987382516/job/45100590002?pr=157341. Could you help add the runner on your end? I guess the runner is down because of the first issue you mention, so let's see if we could tackle it too

That is right, I think the runner needs to be re-authenticated using a new token. Hopefully tomorrow we can re-enable it.

[nWEIdia]

@huydhn , the runner is once again up and running and is busy with your test job: https://github.com/pytorch/pytorch/actions/runs/15987382516/job/45100590002?pr=157341
thanks to @malfet for helping connect the runner!

Next: I'm going to look into setting up github App according to https://github.com/pytorch/test-infra/blob/main/docs/partners_pytorch_ci_runners.md

[nWEIdia]

@atalman helped install the github app: https://github.com/apps/nvidia-pytorch-runner-tokens

Following the workflow of

1. getting jwt token: https://docs.github.com/en/apps/creating-github-apps/authenticating-with-a-github-app/generating-a-json-web-token-jwt-for-a-github-app#example-using-python-to-generate-a-jwt
2. curl script querying https://api.github.com/app/installations/${INSTALL_ID}/access_tokens [Note: here Meta or github pytorch pytorch owner needs to provide the INSTALL_ID, the result of this is an access_token
3. use the above returned access token, run curl script again but this time querying https://api.github.com/orgs/${ORGS}/actions/runners/registration-token

We were able to get the tokens needed for future runner registration.

Closing this issue.