-
Notifications
You must be signed in to change notification settings - Fork 287
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
Log PT2 chromium events to scuba #2424
base: main
Are you sure you want to change the base?
Conversation
This pull request was exported from Phabricator. Differential Revision: D61392607 |
Summary: X-link: pytorch/pytorch#133859 This diff implements a bunch of views for internal scuba viewing. TODOS that I might punt to another diff: - Saving cache stats via counter is definitely sus here, but there's not really a good way to track "fx graph cache hit for this compile phase" right now. Will think about this more. - We should definitely log frame id, compile id, etc - We should definitely be logging configs. That way, we can A/B test based on whether a config is turned on. - idk what I'm doing with compile_uuid yet, but it's useful when you want to look at samples for a single run. I think if we had mast job info this field is not needed, but it's nice to be able to drill down to a single run and get its chrome trace view or icicle view, so idk Reviewed By: ezyang Differential Revision: D61392607
1249d76
to
04c5382
Compare
This pull request was exported from Phabricator. Differential Revision: D61392607 |
Summary: X-link: pytorch/benchmark#2424 Pull Request resolved: #133859 This diff implements a bunch of views for internal scuba viewing. TODOS that I might punt to another diff: - Saving cache stats via counter is definitely sus here, but there's not really a good way to track "fx graph cache hit for this compile phase" right now. Will think about this more. - We should definitely log frame id, compile id, etc - We should definitely be logging configs. That way, we can A/B test based on whether a config is turned on. - idk what I'm doing with compile_uuid yet, but it's useful when you want to look at samples for a single run. I think if we had mast job info this field is not needed, but it's nice to be able to drill down to a single run and get its chrome trace view or icicle view, so idk Test Plan: All of the above views are run with nanogpt benchmark: ``` buck run mode/opt caffe2/benchmarks/dynamo:torchbench -- --training --backend=inductor --only nanogpt --performance ``` Reviewed By: ezyang Differential Revision: D61392607
This pull request was exported from Phabricator. Differential Revision: D61392607 |
Summary: Pull Request resolved: #2424 X-link: pytorch/pytorch#133859 This diff implements a bunch of views for internal scuba viewing. TODOS that I might punt to another diff: - Saving cache stats via counter is definitely sus here, but there's not really a good way to track "fx graph cache hit for this compile phase" right now. Will think about this more. - We should definitely log frame id, compile id, etc - We should definitely be logging configs. That way, we can A/B test based on whether a config is turned on. - idk what I'm doing with compile_uuid yet, but it's useful when you want to look at samples for a single run. I think if we had mast job info this field is not needed, but it's nice to be able to drill down to a single run and get its chrome trace view or icicle view, so idk Reviewed By: ezyang Differential Revision: D61392607
04c5382
to
abba6fd
Compare
Summary: X-link: pytorch/benchmark#2424 Pull Request resolved: #133859 This diff implements a bunch of views for internal scuba viewing. TODOS that I might punt to another diff: - Saving cache stats via counter is definitely sus here, but there's not really a good way to track "fx graph cache hit for this compile phase" right now. Will think about this more. - We should definitely log frame id, compile id, etc - We should definitely be logging configs. That way, we can A/B test based on whether a config is turned on. - idk what I'm doing with compile_uuid yet, but it's useful when you want to look at samples for a single run. I think if we had mast job info this field is not needed, but it's nice to be able to drill down to a single run and get its chrome trace view or icicle view, so idk Test Plan: All of the above views are run with nanogpt benchmark: ``` buck run mode/opt caffe2/benchmarks/dynamo:torchbench -- --training --backend=inductor --only nanogpt --performance ``` Reviewed By: ezyang Differential Revision: D61392607
Summary: X-link: pytorch/benchmark#2424 This diff implements a bunch of views for internal scuba viewing. TODOS that I might punt to another diff: - Saving cache stats via counter is definitely sus here, but there's not really a good way to track "fx graph cache hit for this compile phase" right now. Will think about this more. - We should definitely log frame id, compile id, etc - We should definitely be logging configs. That way, we can A/B test based on whether a config is turned on. - idk what I'm doing with compile_uuid yet, but it's useful when you want to look at samples for a single run. I think if we had mast job info this field is not needed, but it's nice to be able to drill down to a single run and get its chrome trace view or icicle view, so idk Test Plan: All of the above views are run with nanogpt benchmark: ``` buck run mode/opt caffe2/benchmarks/dynamo:torchbench -- --training --backend=inductor --only nanogpt --performance ``` Reviewed By: ezyang Differential Revision: D61392607
Summary: X-link: pytorch/pytorch#133859 This diff implements a bunch of views for internal scuba viewing. TODOS that I might punt to another diff: - Saving cache stats via counter is definitely sus here, but there's not really a good way to track "fx graph cache hit for this compile phase" right now. Will think about this more. - We should definitely log frame id, compile id, etc - We should definitely be logging configs. That way, we can A/B test based on whether a config is turned on. - idk what I'm doing with compile_uuid yet, but it's useful when you want to look at samples for a single run. I think if we had mast job info this field is not needed, but it's nice to be able to drill down to a single run and get its chrome trace view or icicle view, so idk Reviewed By: ezyang Differential Revision: D61392607
abba6fd
to
139c7f1
Compare
This pull request was exported from Phabricator. Differential Revision: D61392607 |
Summary: X-link: pytorch/benchmark#2424 Pull Request resolved: #133859 This diff implements a bunch of views for internal scuba viewing. TODOS that I might punt to another diff: - Saving cache stats via counter is definitely sus here, but there's not really a good way to track "fx graph cache hit for this compile phase" right now. Will think about this more. - We should definitely log frame id, compile id, etc - We should definitely be logging configs. That way, we can A/B test based on whether a config is turned on. - idk what I'm doing with compile_uuid yet, but it's useful when you want to look at samples for a single run. I think if we had mast job info this field is not needed, but it's nice to be able to drill down to a single run and get its chrome trace view or icicle view, so idk Test Plan: All of the above views are run with nanogpt benchmark: ``` buck run mode/opt caffe2/benchmarks/dynamo:torchbench -- --training --backend=inductor --only nanogpt --performance ``` Reviewed By: ezyang Differential Revision: D61392607
This pull request was exported from Phabricator. Differential Revision: D61392607 |
Summary: Pull Request resolved: #2424 X-link: pytorch/pytorch#133859 This diff implements a bunch of views for internal scuba viewing. TODOS that I might punt to another diff: - Saving cache stats via counter is definitely sus here, but there's not really a good way to track "fx graph cache hit for this compile phase" right now. Will think about this more. - We should definitely log frame id, compile id, etc - We should definitely be logging configs. That way, we can A/B test based on whether a config is turned on. - idk what I'm doing with compile_uuid yet, but it's useful when you want to look at samples for a single run. I think if we had mast job info this field is not needed, but it's nice to be able to drill down to a single run and get its chrome trace view or icicle view, so idk Reviewed By: ezyang Differential Revision: D61392607
139c7f1
to
31e19f1
Compare
Summary: X-link: pytorch/pytorch#133859 This diff implements a bunch of views for internal scuba viewing. TODOS that I might punt to another diff: - Saving cache stats via counter is definitely sus here, but there's not really a good way to track "fx graph cache hit for this compile phase" right now. Will think about this more. - We should definitely log frame id, compile id, etc - We should definitely be logging configs. That way, we can A/B test based on whether a config is turned on. - idk what I'm doing with compile_uuid yet, but it's useful when you want to look at samples for a single run. I think if we had mast job info this field is not needed, but it's nice to be able to drill down to a single run and get its chrome trace view or icicle view, so idk Reviewed By: ezyang Differential Revision: D61392607
31e19f1
to
1cd7776
Compare
This pull request was exported from Phabricator. Differential Revision: D61392607 |
Summary: X-link: pytorch/pytorch#133859 This diff implements a bunch of views for internal scuba viewing. TODOS that I might punt to another diff: - Saving cache stats via counter is definitely sus here, but there's not really a good way to track "fx graph cache hit for this compile phase" right now. Will think about this more. - We should definitely log frame id, compile id, etc - We should definitely be logging configs. That way, we can A/B test based on whether a config is turned on. - idk what I'm doing with compile_uuid yet, but it's useful when you want to look at samples for a single run. I think if we had mast job info this field is not needed, but it's nice to be able to drill down to a single run and get its chrome trace view or icicle view, so idk Reviewed By: ezyang Differential Revision: D61392607
1cd7776
to
14b2d39
Compare
This pull request was exported from Phabricator. Differential Revision: D61392607 |
Summary: X-link: pytorch/pytorch#133859 This diff implements a bunch of views for internal scuba viewing. TODOS that I might punt to another diff: - Saving cache stats via counter is definitely sus here, but there's not really a good way to track "fx graph cache hit for this compile phase" right now. Will think about this more. - We should definitely log frame id, compile id, etc - We should definitely be logging configs. That way, we can A/B test based on whether a config is turned on. - idk what I'm doing with compile_uuid yet, but it's useful when you want to look at samples for a single run. I think if we had mast job info this field is not needed, but it's nice to be able to drill down to a single run and get its chrome trace view or icicle view, so idk Reviewed By: ezyang Differential Revision: D61392607
14b2d39
to
ec3ec05
Compare
This pull request was exported from Phabricator. Differential Revision: D61392607 |
1 similar comment
This pull request was exported from Phabricator. Differential Revision: D61392607 |
Summary: Pull Request resolved: #2424 X-link: pytorch/pytorch#133859 This diff implements a bunch of views for internal scuba viewing. TODOS that I might punt to another diff: - Saving cache stats via counter is definitely sus here, but there's not really a good way to track "fx graph cache hit for this compile phase" right now. Will think about this more. - We should definitely log frame id, compile id, etc - We should definitely be logging configs. That way, we can A/B test based on whether a config is turned on. - idk what I'm doing with compile_uuid yet, but it's useful when you want to look at samples for a single run. I think if we had mast job info this field is not needed, but it's nice to be able to drill down to a single run and get its chrome trace view or icicle view, so idk Reviewed By: ezyang Differential Revision: D61392607
ec3ec05
to
a4b4503
Compare
Summary: Pull Request resolved: #2424 X-link: pytorch/pytorch#133859 This diff implements a bunch of views for internal scuba viewing. TODOS that I might punt to another diff: - Saving cache stats via counter is definitely sus here, but there's not really a good way to track "fx graph cache hit for this compile phase" right now. Will think about this more. - We should definitely log frame id, compile id, etc - We should definitely be logging configs. That way, we can A/B test based on whether a config is turned on. - idk what I'm doing with compile_uuid yet, but it's useful when you want to look at samples for a single run. I think if we had mast job info this field is not needed, but it's nice to be able to drill down to a single run and get its chrome trace view or icicle view, so idk Reviewed By: ezyang Differential Revision: D61392607
This pull request was exported from Phabricator. Differential Revision: D61392607 |
a4b4503
to
83dd4f4
Compare
Summary:
X-link: pytorch/pytorch#133859
This diff implements a bunch of views for internal scuba viewing.
TODOS that I might punt to another diff:
Differential Revision: D61392607