Torch.cuda.is_available() is False in conda env

Hi, I’m was using PyTorch 1.5.1 version and CUDA 10.2.
But for version compat, I had to make a new conda environment and change PyTorch version to 1.4.0.
In new conda env, torch.cuda.is_available() returns false but in the base env, it returns true.
How should I do?

Maybe you’ve installed the CPU only version in your new environment.
Could you check the installed packages and make sure that cudatoolkit was also installed (assuming you’ve used the conda binaries)?