Tomas Ruiz

The command kernels skills add cuda-kernels --claude from the instruction still failed for me. I had to remove cuda-kernels. The command uses a relative working directory, so if you want to install the skill at the user level, you should switch to the home dir:

$ cd ~
$ kernels skills add --claude
Installed 'cuda-kernels' to ~/.claude/skills/cuda-kernels

Ok, so the instructions in the article need to be updated?

Question: Is the baseline in section Isolated RMSNorm benchmark (Pytorch baseline) using torch.compile() or not? Custom kernels should be ideally beating torch.compiled code.

I get an error while trying to install:

$ pip install git+https://github.com/huggingface/kernels.git
...
ERROR: git+https://github.com/huggingface/kernels.git does not appear to be a Python project: neither 'setup.py' nor 'pyproject.toml' found.

Installation with pip install kernels works, but the command has no skills option:

$ kernels skills add cuda-kernels --claude
usage: kernel [-h] {check,download,versions,upload,lock,generate-readme,benchmark} ...