UserWarning: User provided device_type of 'cuda', but CUDA is not available. Disabling

[perkel666]

I have 3080 and installed it as per git readme.

While running: python tortoise/do_tts.py --kv_cache --half --no_cond_free --preset single_sample --candidates 1 --text "I have a problem Huston, i repeat !" --voice train-grace

This is the full ouput from console:

C:\Users\Perkel\AppData\Local\Programs\Python\Python310\lib\site-packages\torch\amp\autocast_mode.py:202: UserWarning: User provided device_type of 'cuda', but CUDA is not available. Disabling
  warnings.warn('User provided device_type of \'cuda\', but CUDA is not available. Disabling')
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:31<00:00,  3.89s/it]
Computing best candidates using CLVP
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:10<00:00,  1.32s/it]
Transforming autoregressive outputs into audio..
100%|██████████████████████████████████████████████████████████████████████████████████| 10/10 [00:04<00:00,  2.18it/s]
Generating 1 candidates for voice train_grace (seed=None) took 51.85 seconds

---

---

From what I read about this kind of issue it seems to be problem with dependency.

[perkel666]

Oh, never mind.

C:\Users\Perkel\AppData\Local\Programs\Python\Python310

Python 3.10 is not yet compatible with a lot of pytorch.

Use Python 3.9. Leaving this for people to see if they get this problem.

[perkel666]

Turns out problem was different. For some reason original repo of Tortoise had link to PyTorch repo that was supposed to install cuda capable version. Well it didn't for me.

I run this:

pip install torch==1.11.0+cu113 torchvision==0.12.0+cu113 torchaudio==0.11.0+cu113 -f https://download.pytorch.org/whl/torch_stable.html

And with it my problem is gone. Generation time went from 58 seconds on CPU to 9 seconds which is massive improvement.

[noahlessard]

Perkel666's solution to this problem stills works as of 3/16. Thanks perkel