AssertionError: CUDA is not available

[ChaoYue0307]

I have cuda version 9.0.176 on my server, but still got the error when running train.py in the title, how to handle that?
Thanks

[yihong-chen]

Did you test your environment settings with some pytorch baselines, for example, the MNIST classification ? It seems that the CUDA isn't installed well.

[ChaoYue0307]

Thanks and I reinstalled pytorch and cuda, the problem is solved.
But new problem comes with
RuntimeError: cuda runtime error (10) : invalid device ordinal at torch/csrc/cuda/Module.cpp:32
do you have any idea about that?

Thanks

[yihong-chen]

It seems that you should check your device_id. It shouldn't be larger than the number of available GPUs.

[ChaoYue0307]

Really thanks a lot, indeed the problem is the device_id, your default is 7 but I have only 4 gpus on my server.