Skip to content

Conversation

naruto110
Copy link

To adapt to the RTX 5090, we have modified the setup.py file.

To adapt to the RTX 5090, we have modified the **setup.py** file.
@tridao
Copy link
Collaborator

tridao commented Aug 27, 2025

rtx 5090 is sm120 and it's already included. Why would removing sm100 help?

@johnnynunez
Copy link
Contributor

johnnynunez commented Aug 28, 2025

rtx 5090 is sm120 and it's already included. Why would removing sm100 help?

could you add thor and spark?
11.0 and 12.1?

also mamba must be adapted to cuda 13

@naruto110
Copy link
Author

rtx 5090 is sm120 and it's already included. Why would removing sm100 help?

Yes, the setup file has indeed already included it. However, after compiling and installing, although the package can be imported, the RTX 5090 cannot be properly initialized to train a model. I suspect this is because the default library compiled for sm100 is being used. Once I removed sm100, the 5090 was able to start training normally.

@johnnynunez
Copy link
Contributor

rtx 5090 is sm120 and it's already included. Why would removing sm100 help?

Yes, the setup file has indeed already included it. However, after compiling and installing, although the package can be imported, the RTX 5090 cannot be properly initialized to train a model. I suspect this is because the default library compiled for sm100 is being used. Once I removed sm100, the 5090 was able to start training normally.

i don't have problems with actual setup:

git clone --depth=1 --recursive https://github.com/state-spaces/mamba  /opt/mamba

# Navigate to the directory containing mamba's setup.py
cd /opt/mamba

MAX_JOBS="$(nproc)" \
MAMBA_FORCE_BUILD="TRUE" \
MAMBA_SKIP_CUDA_BUILD="FALSE" \
python3 setup.py bdist_wheel --dist-dir=/opt/mamba/wheels
pip3 install /opt/mamba/wheels/mamba*.whl

@naruto110
Copy link
Author

i don't have problems with actual setup:

Thanks for your reply. I have a RuntimeError if I don't delete sm100 from the setup file.

"RuntimeError: CUDA error: no kernel image is available for execution on the device CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1 Compile with TORCH_USE_CUDA_DSA to enable device-side assertions."

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.

3 participants