-
Notifications
You must be signed in to change notification settings - Fork 296
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
Run only a subset for optim benchmarks, attempt to fix #2392
Conversation
@@ -579,8 +592,6 @@ def run_model(modelName, device, Optim, defaults, maybe_pt2_): | |||
len(params), | |||
params[0].device, | |||
) | |||
if Optim.__name__ == "SGD": | |||
defaults["lr"] = 1e-2 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
no longer necessary as we have a default LR for SGD now
@janeyx99 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
@janeyx99 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
@@ -412,7 +412,7 @@ def get_metrics_by_date( | |||
if not start_date: | |||
raise RuntimeError( | |||
f"No start date in previous JSONS found to compare towards the end date {end_date}. User specified start date: {args.start_date}. " | |||
+ f"Available JSON dates: {available_metrics_jsons.keys()}. No regression info has been generated." |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
available_metrics_jsons doesn't have keys anymore as it's a List[str] and not a dict. It's unclear whether there were ever keys. Just remove the printout for now as it's not needed for the error message.
A first attempt to fix https://github.com/pytorch/benchmark/actions/workflows/userbenchmark-regression-detector.yml by only running a subset of models
Impact of this change:
This workflow is green: https://github.com/pytorch/benchmark/actions/runs/10066723358/job/27828703722
It also takes so much shorter. At one point, the benchmark was 7 hours! Now it's 7 minutes.
Test Plan
Test run: https://github.com/pytorch/benchmark/actions/runs/10065401507
Final test run: https://github.com/pytorch/benchmark/actions/runs/10066723358/job/27828703722