Closed jwfromm closed 3 months ago
Name | Link |
---|---|
Latest commit | 21194db4c78e7f4ef00044f7e7adf79d286defa3 |
Latest deploy log | https://app.netlify.com/sites/pytorch-fbgemm-docs/deploys/6675a9fc96ba400008fe2e9a |
Deploy Preview | https://deploy-preview-2765--pytorch-fbgemm-docs.netlify.app |
Preview on mobile | Toggle QR Code...Use your smartphone camera to open QR code link. |
To edit notification comments on pull requests, go to your Netlify site configuration.
This pull request was exported from Phabricator. Differential Revision: D58791438
This pull request was exported from Phabricator. Differential Revision: D58791438
This pull request was exported from Phabricator. Differential Revision: D58791438
This pull request has been merged in pytorch/FBGEMM@623f789fa2b37f355729b28f8cea4cf33ab5ac95.
Summary: Redesign our FP8 benchmarking script for more modularity and extend coverage to all ops. We also deprecate the nonprod mm_row_bench that has been used for checking FP8 performance previously.
Most notably, I introduce a class based way to integrate new quantize operators into the benchmark that should make it easier to maintain this script long term.
I think the basic structure here could easily be extended beyond FP8 if we were interested in a universal benchmarking setup.
Reviewed By: jspark1105, jianyuh
Differential Revision: D58791438