No CUDA runtime is found, using CUDA_HOME='/usr/local/cuda'

fastai
1

I have installed Cuda on my machine and torch.cuda.is_available() returns True,
but when I execute a notebook by multiFit project

multifit/notebooks/ MLDoc-JA-multifit_fp16.ipynb

I recieve
No CUDA runtime is found, using CUDA_HOME=’/usr/local/cuda’

It’s not an error and it continues, but I wonder why it doesn’t recognize the GPU?

2

Setting

'CUDA_VISIBLE_DEVICES=0,1'

at the beginning of file can solve the problem…

3

I think this only means that if you do echo $CUDA_HOME it doesnt print anything, so you can set it if you want with CUDA_HOME=/usr/local/cuda (which was already used by the command)