Skip to content

Conversation

anmyachev
Copy link
Contributor

@anmyachev anmyachev commented Sep 2, 2025

Current PTI version from DLE 2025.2.1 is 0.13.1

@anmyachev
Copy link
Contributor Author

Hi @jfedorov,

when trying to update to pti 0.13.1 there are more situations in which not all kernels are registered (https://github.com/intel/intel-xpu-backend-for-triton/actions/runs/17471769868/job/49621692636?pr=5015 and https://github.com/intel/intel-xpu-backend-for-triton/actions/runs/17471769866/job/49621692732?pr=5015). Any known workarounds probably?

@jfedorov
Copy link

jfedorov commented Sep 8, 2025

Hi @anmyachev,
thanks for letting know.
Are you saying that with the previous PTI version - there was no such issues?
At the moment - I can not think about workaround as well as what could be causing such behavior.
I believe we have not done any code changes that might introduce such regressions.. But I will looks with attention and will update you.
On a different note - looks like we should add some Triton integration tests into PTI integration testing.

@anmyachev
Copy link
Contributor Author

Are you saying that with the previous PTI version - there was no such issues?

There were, but much less frequently.

I believe we have not done any code changes that might introduce such regressions.. But I will looks with attention and will update you.

could this be due to the DLE version change from 2025.1 to 2025.2? Or does it mainly depend on the driver?

On a different note - looks like we should add some Triton integration tests into PTI integration testing.

If PTI already using PyTorch in CI, then adding this probably won't be that hard.

BTW: thanks for the response!

@anmyachev anmyachev closed this Sep 12, 2025
@anmyachev anmyachev reopened this Sep 12, 2025
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.

2 participants