NVIDIA / NVIDIA/TransformerEngine
Enable post-RHT amax estimation
Open
enhancement
fp4
- Dominant language
- Python
- Stars
- 3.5k
- Forks
- 831
- Avg merge
- 3d 11h
- Merged PRs (30d)
- 65
Description
Is your feature request related to a problem? Please describe.
RHT+amax kernel prevents fusion.
Describe the solution you'd like
Estimate post rht amax from pre rht amax with a linear function. This eliminates rht+amax kernel. Make this feature optional. Make hyperparameters (amax estimation scale) to be tunable.
Validation
Validate lm loss with dense/moe models. Ensure the convergence is the same or better.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Assessment
This issue has not been assessed yet.