Skip to content

Conversation

@xiaobochen-amd
Copy link
Contributor

First add hipblaslt tune example.

The follow-up plan:

  • use more devices to tune.
  • merge into megatron backend.
  • moe tune tools.
  • tensile tune tools.

@RuibinCheung
Copy link
Contributor

First add hipblaslt tune example.

The follow-up plan:

  • use more devices to tune.
  • merge into megatron backend.
  • moe tune tools.
  • tensile tune tools.

I wonder this tools can provide comparing result between default kernel and tuned kernl (e.g csv or xlsx) ? It may help user to observe the performance of gemm kernel and estimate the gap.

@wenxie-amd wenxie-amd merged commit 7003718 into main Apr 1, 2025
1 check passed
@xiaobochen-amd
Copy link
Contributor Author

First add hipblaslt tune example.
The follow-up plan:

  • use more devices to tune.
  • merge into megatron backend.
  • moe tune tools.
  • tensile tune tools.

I wonder this tools can provide comparing result between default kernel and tuned kernl (e.g csv or xlsx) ? It may help user to observe the performance of gemm kernel and estimate the gap.

Sure, we can develop this feature in future versions.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants