152334H / tortoise-tts-fast

Fast TorToiSe inference (5x or your money back!)
GNU Affero General Public License v3.0
759 stars 177 forks source link

request: CPU-only version #59

Open eloop001 opened 1 year ago

eloop001 commented 1 year ago

@152334H Would it be easy for you to make an option to run only on CPU. It will be very slow, but it will be good to have it so that I can run the code locally and track the path of how the code works. Also enable a wider audience to get started. I can perhaps give it a go, but my standards/compentencies are way lower than yours.

absane commented 1 year ago

Most scripts have an option to select the device:

image

If not, then you can set an environment variable to tell PyTorch to not use a GPU:

export CUDA_VISIBLE_DEVICES=""

If you can use something like VS Code or PyCharm, you could set breakpoints and sprinkle in print or icecream.ic statements to follow areas of interest. That's been my primary way of deciphering the logic. Working that way, I found no reason why the GPU needed to be skipped, but you may have different goals or objectives than I do.

eloop001 commented 1 year ago

Most scripts have an option to select the device:

image

If not, then you can set an environment variable to tell PyTorch to not use a GPU:

export CUDA_VISIBLE_DEVICES=""

If you can use something like VS Code or PyCharm, you could set breakpoints and sprinkle in print or icecream.ic statements to follow areas of interest. That's been my primary way of deciphering the logic. Working that way, I found no reason why the GPU needed to be skipped, but you may have different goals or objectives than I do.

Thanks @absane There is a fork of the original tortoise-tts repo where someone has done it. I also tried to back-ingeneer before I saw it, and I did indeed get it to run. I was just thinking a lot of peeps would want this functionality, and it could be incorporated in the repo. The person who did do it was hacking a lot.

I will try export CUDA_VISIBLE_DEVICES=""

thanks for your reply.