[CI] Move H100/A100 benchmark & tutorial jobs to OSDC (ARC)#4458
Open
huydhn wants to merge 1 commit into
Open
Conversation
Switch the direct runs-on H100/A100 jobs (run_microbenchmarks, dashboard_perf_test, run_tutorials) onto OSDC/ARC by adding a job-level CUDA container with --gpus all and the ARC runner labels: - run_microbenchmarks: linux.aws.h100 -> mt-l-x86iamx-22-225-h100 - dashboard_perf_test / run_tutorials: linux.aws.a100 -> mt-l-x86iavx512-11-125-a100 Pre-install torch's pure-python deps from the in-cluster pypi-cache so the nightly cu126 install doesn't reach the cache-enforcer-blocked files.pythonhosted.org, and clear PIP_EXTRA_INDEX_URL so the default cpu index can't supply a +cpu torch. The two benchmark-upload jobs get id-token + the gha_workflow_upload-benchmark-results role for OSS dashboard uploads (no EC2 instance profile on ARC).
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/4458
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (1 Unrelated Failure)As of commit d695f80 with merge base 9c010ae ( BROKEN TRUNK - The following job failed but were present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
huydhn
changed the title
huydhn
added
the
module: not user facing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Moves the direct
runs-on:H100/A100 jobs onto OSDC (ARC). These don't uselinux_job, so per the agreed approach they get a job-levelcontainer:added in place rather than being converted to the reusable workflow.run_microbenchmarks.ymllinux.aws.h100→mt-l-x86iamx-22-225-h100dashboard_perf_test.ymllinux.aws.a100→mt-l-x86iavx512-11-125-a100run_tutorials.ymllinux.aws.a100→mt-l-x86iavx512-11-125-a100Each job now runs inside
pytorch/almalinux-builder:cuda13.0withoptions: --gpus all(the container CUDA version is independent of the pip-installed cu126 torch; cuda13.0 matches the ARC nodes' driver). torch's pure-python deps are pre-installed from the in-cluster pypi-cache, andPIP_EXTRA_INDEX_URLis cleared on the torch install so the default cpu index can't supply a+cputorch (same pattern as the H100 PRs).The two upload jobs get
id-token: write+ aconfigure-aws-credentialsstep usingrole/gha_workflow_upload-benchmark-results(ARC pods have no EC2 instance profile).Scope: H100/A100 only — regular GPU (g5/g6/L4) and CPU jobs are intentionally left on EC2. Companion PRs: #4456 (1x/4xH100), #4457 (release_model). Depends on
linux_job_v3-era infra (now onpytorch/test-infra@main).Needs validation
setup-miniconda+ benchmark uploads inside the ARC container (esp. thatrole/gha_workflow_upload-benchmark-resultsis assumable from these pods).actions/checkout/JS actions insidealmalinux-builder(git present; node provided by the ARC container hook).