search

FalkonML / falkon

Large-scale, multi-GPU capable, kernel solver
https://falkonml.github.io/falkon/
MIT License
181 stars 22 forks source link

Getting error #32

Closed MrHuff closed 3 years ago

MrHuff commented 3 years ago

Hi!

Thank you for writing this library. It's very cool to finally be able to scale kernel regression to a billion points!

Running into the following:

RuntimeError: [KeOps] This KeOps shared object has been compiled without cuda support: 1) to perform computations on CPU, simply set tagHostDevice to 0 2) to perform computations on GPU, please recompile the formula with a working version of cuda.

Thank you for the help!

Best regards, Robert

Giodiro commented 3 years ago

Hi!

Are you trying to use it on a GPU? You should check that CUDA is installed properly, maybe you could try sending the output of nvcc --version

Also try to delete the KeOps cache (if you are on ubuntu or similar you can run rm -rf ~/.cache/pykeops*) and run falkon again to force recompilation.

MrHuff commented 3 years ago

Hi,

NVCC version is 11.1.

Tried clearing the pykeops cache, but it gave me the same error.

Maybe it pykeops doesn't support NVCC 11.1?

Thank you!

Best regards, Robert

MrHuff commented 3 years ago

Hi,

After some debugging, it seems like pykeops wasn't able to detect my CUDA installation.

-- Looking for a CUDA compiler -- Looking for a CUDA compiler - NOTFOUND -- No CUDA support -- Configuring done -- Generating done

Closing this!

MrHuff commented 3 years ago

closed, found bug on my own

Giodiro commented 3 years ago

Great, happy you managed to fix it :)