Closed vkuzo closed 3 months ago
Stack from ghstack (oldest at bottom):
Summary:
This PR adds the axiswise scaling granularity to Float8Tensor and ensures that basic ops like transpose and torch._scaled_mm work as expected.
Float8Tensor
torch._scaled_mm
A future PR will add integration with Float8Linear.
Float8Linear
Test Plan:
TODO
Reviewers:
Subscribers:
Tasks:
Tags:
Stack from ghstack (oldest at bottom):
350
349
348
347
346
345
344
Summary:
This PR adds the axiswise scaling granularity to
Float8Tensorand ensures that basic ops like transpose andtorch._scaled_mmwork as expected.A future PR will add integration with
Float8Linear.Test Plan:
TODO
Reviewers:
Subscribers:
Tasks:
Tags: