cublaslt ran into an error! On h100

[Ar770]

I'm not sure what is the status of the other issues opened related to this problem, I can confirm that the problem remaines with 0.40.1 on an H100 instance.
See here the complete error:
#538

I was in touch with the Lambdalabs support, maybe it could help.
I added their response here:
#383

[TimDettmers]

Thanks for letting me know. I looked at the details and it turns out Hopper does not support col_turning or col_ampere formats. As such, new logic is needed. I might implement the FP8 version instead for Hopper.

[Ar770]

Thank you!
Do you have ETA to resolve this issue?

[Mythli]

Hi @TimDettmers ,

Thank you for your amazing work.
While I'm not qualified to fix this issue myself I can provide a bounty so you could focus on it.

I can provide an H100 for testing as well.

Please reach out to me at [email protected]

[jieWANGforwork]

I encountered the same error as loading 8bit Llama on H100. Are there any solutions?

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

[hyaticua]

Hi this still seems to be an issue.

[corbt]

Confirming this is still an issue, and would love to get it fixed so we can use H100s for training!

[karroo]

I have the same error on H100!!!

[zhangj1an]

still having the same issue

[Jean-KOUAGOU]

I get the following: Error operation not supported at line 380 in file /mmfs1/gscratch/zlab/timdettmers/git/bitsandbytes/csrc/pythonInterface.c with H100 GPU

[yz26cn]

Still getting the issue on H100 with 8-bit setup. 4-bit works as usual but 8-bit does.