Philipp Moritz

Wonderful! Small nit and then it looks good to go if the tests pass :)

Btw, while I was trying out this PR, I got the following error: ```python import torch from vllm import _custom_ops as ops A = torch.randn(8, 4096, dtype=torch.float16, device="cuda") B =...

Ah, I think I know -- I didn't transpose `B` appropriately / it was not in column major order :) Can you add a check in `cutlass_scaled_mm_dq` to make sure...

Thanks! We should also assert that the tensors are contiguous :)

Did you push your changes? Most of my comments still need to be resolved :)

Thanks for the fixes, I have a few more comments! As a mental picture, it should never be possible to crash the python interpreter from python code. Asserts in the...

Otherwise the PR looks good to me now :)

@DonJayamanne Do you know when we can expect the fix to this being available in a released version of this extension? That would be very much appreciated! Thanks for fixing...

Got it, thanks a lot! I think I got confused with the old version that was released on https://github.com/microsoft/vscode-jupyter/releases/tag/2022.09.120 -- it seems on the marketplace a newer version is available....

It looks like this issue was tricky to fix, kudos for getting it solved!