Closed vkuzo closed 3 months ago
@vkuzo has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.
This pull request has been merged in pytorch-labs/float8_experimental@4fb0ada5a138d1c2d572cf73d225c8609e060f79.
Stack from ghstack (oldest at bottom):
300
298
297
296
294
293
291
290
Summary:
Makes
benchmarks/bench_linear_float8.py
support per-tensor scaling configurations.Verified that performance is as we expect
Test Plan:
paste of testing for delayed -> dynamic, changing the tensors one by one: https://gist.github.com/vkuzo/9e8f995e51ef16f483347c0f86bb0ac3
Reviewers:
Subscribers:
Tasks:
Tags:
Differential Revision: D59305789