why torch.cuda.is_available()= False in main_task()

[felix-ky]

Thank you for open source the code,

I have a problem importing a neural network model in main_task()

and I got torch.cuda.is_available()= False.

I try to allocate GPU for the task by using @ray.remote(num_gpus=1) instead of num_cpus=1

However, if so, the code blocks in ray.get([pg.ready() for pg in pgs]) forever

[LiuRicky]

Thanks for your interest.

For your first question, torch.cuda.is_available()= False. I guess there maybe some error when you install pytorch. You can try reinstall with GPU version torch. Or maybe caused by your cuda installation.

For the second question, code blocks. We do not recommend you change any setting in main_task(). Try solving cuda or torch issue at first.

Besides, I guess you can paste your error information to Deepseek or ChatGPT for solution.