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
Add opacus microbenchmarks code #368
Comments
moaradwan
pushed a commit
to moaradwan/opacus
that referenced
this issue
Aug 25, 2022
Summary: ## Types of changes - [ ] Bug fix (non-breaking change which fixes an issue) - [X] New feature (non-breaking change which adds functionality) - [ ] Breaking change (fix or feature that would cause existing functionality to change) - [ ] Docs change / refactoring / dependency upgrade ## Motivation and Context / Related issue There's a task pytorch#368 for committing benchmark code. In this change I add these benchmarks into CI integration tests. To choose thresholds I ran the benchmarks locally on all the layers with (batch size: 16, num_runs: 100, num_repeats: 20, forward_only: False), and generated the following report: | | memory* | memory* | memory* | memory* | memory* | runtime | runtime | runtime | runtime | runtime | |--------------|---------|--------|------------|--------|-------------|------------------------|----------------------|--------------------|------------------------|--------------------| | value | control | dp | dp/control | gsm | gsm/control | control | dp | dp/control | gsm | gsm/control | | base_layer | | | | | | | | | | | | conv | 0.0 | | | 0.0 | | 2.021756922606001 | | | 3.2889059911645036 | 1.6267563891534373 | | embedding | 0.0 | | | 0.0 | | 0.002484286398502263 | | | 0.013664713416999803 | 5.5004581698946 | | groupnorm | 0.0 | | | 0.0 | | 0.0001871487290072764 | | | 0.00043170701800136156 | 2.306759016165034 | | gru | 0.0 | 0.0 | | 0.0 | | 0.045029744959007065 | 0.057370035271503174 | 1.2740475284443677 | 0.2402042072270033 | 5.334345274344187 | | instancenorm | 0.0 | | | 0.0 | | 0.004493124293996517 | | | 0.006058429501005777 | 1.3483779002287433 | | layernorm | 0.0 | | | 0.0 | | 0.00011227587499979562 | | | 0.0002241125804985131 | 1.9960884784814286 | | linear | 0.0 | | | 0.0 | | 0.001010556231000001 | | | 0.003052972127999998 | 3.021080900148341 | | lstm | 0.0 | 0.0 | | 0.0 | | 0.052634652085002925 | 0.06508583683050075 | 1.2365586975931682 | 0.2982182763324963 | 5.665816425477371 | | mha | 0.0 | 0.0 | | 0.0 | | 0.018872260358001765 | 0.01870937360499738 | 0.9913689854890476 | 0.02688384014700477 | 1.424516175435558 | | rnn | 0.0 | 0.0 | | 0.0 | | 0.01576623683249454 | 0.02184348723049516 | 1.3854597937711604 | 0.10178373254250346 | 6.455803856296582 | (*) This report wasn't generated on a machine with CUDA so the memory wasn't measured. Will update later when it runs in CI on a GPU machine. Using the report and section 3 in the [paper](https://arxiv.org/pdf/2109.12298.pdf), I parameterised the runtime and memory thresholds for different layers. ## How Has This Been Tested (if it applies) I ran the jobs locally and generated reports. ## Checklist - [X] The documentation is up-to-date with the changes I made. - [X] I have read the **CONTRIBUTING** document and completed the CLA (see **CONTRIBUTING**). - [ ] All tests passed, and additional code has been covered with new tests. Pull Request resolved: pytorch#479 Differential Revision: D38999201 Pulled By: moaradwan fbshipit-source-id: 3d02931970e39ea331674c9f0676db9e22c5edaa
facebook-github-bot
pushed a commit
that referenced
this issue
Aug 31, 2022
Summary: ## Types of changes - [ ] Bug fix (non-breaking change which fixes an issue) - [X] New feature (non-breaking change which adds functionality) - [ ] Breaking change (fix or feature that would cause existing functionality to change) - [ ] Docs change / refactoring / dependency upgrade Issue: #368 ## Motivation and Context / Related issue There's a task #368 for committing benchmark code. In this change I add these benchmarks into CI integration tests. To choose thresholds I ran the benchmarks locally on all the layers with (batch size: 16, num_runs: 100, num_repeats: 20, forward_only: False), ~and generated the following report:~ (please check the [comment below](#481 (comment))) Using the [report](#481 (comment)) and section 3 in the [paper](https://arxiv.org/pdf/2109.12298.pdf), I parameterised the runtime and memory thresholds for different layers. ## How Has This Been Tested (if it applies) - I ran the jobs locally and generated reports. - Local CircleCI config validation `circleci config process .circleci/config.yml` - Local CircleCI job run: `circleci local execute --job JOB_NAME` ## Checklist - [X] The documentation is up-to-date with the changes I made. - [X] I have read the **CONTRIBUTING** document and completed the CLA (see **CONTRIBUTING**). - [x] All tests passed, and additional code has been covered with new tests. Pull Request resolved: #481 Reviewed By: ffuuugor Differential Revision: D39026827 Pulled By: moaradwan fbshipit-source-id: 5c6a7b1b5faaca9e5e95bd172b479a02f59b2b69
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Merge the code used for opacus microbenchmarks into opacus repo
The text was updated successfully, but these errors were encountered: