I didn't get a response from Teratail, so I'm posting it here.
I would like to move the AI image generation package called BigSleep.
I installed the latest version of Anaconda 3.9 on Windows 10 (with RTX 2060 Super) and tried to install it from the command prompt as follows, but it failed.
$conda create-npainter
$ conda activate paint
$ conda install python==3.6.10
$ conda install cudnn=7.6.5=cuda10.0_0
$ python-mpip install --upgrade pip setuptools
$ pip install big-sleep
$ dream "apyramid made of ice"
Traceback (most recent call last):
File "C:\Users\hoge\anaconda3\envs\painter\lib\runpy.py", line 193, in_run_module_as_main
"__main__", mod_spec)
File "C:\Users\hoge\anaconda3\envs\painter\lib\runpy.py", line85, in_run_code
exec(code, run_globals)
File "C:\Users\hoge\anaconda3\envs\painter\Scripts\dream.exe\_main__.py", line 4, in <module>
File "C:\Users\hoge\anaconda3\envs\painter\lib\site-packages\big_sleep\_init__.py", line 1, in <module>
from big_sleep.big_sleep import BigSleep, Image
File "C:\Users\hoge\anaconda3\envs\painter\lib\site-packages\big_sleep\big_sleep.py", line 26, in <module>
assert torch.cuda.is_available(), 'CUDA must be available in order to use Big Sleep'
AssertionError: CUDA must be available in order to use Big Sleep
Reference Site:
TensorFlow to build Anaconda environment in Ubuntu ~CUDA, cuDNN installation ~
Install Big Sleep (CLIP+BigGAN) for AI image generation
When I searched the error message, there was information that told me to install pytorch, so I tried, but it didn't work.
$conda install pytorch torchvision torchaudio cudatoolkit=10.0-cpytorch-conda-forge
$ dream "apyramid made of ice"
Traceback (most recent call last):
File "C:\Users\hoge\anaconda3\envs\painter\lib\runpy.py", line 193, in_run_module_as_main
"__main__", mod_spec)
File "C:\Users\hoge\anaconda3\envs\painter\lib\runpy.py", line85, in_run_code
exec(code, run_globals)
File "C:\Users\hoge\anaconda3\envs\painter\Scripts\dream.exe\_main__.py", line 4, in <module>
File "C:\Users\hoge\anaconda3\envs\painter\lib\site-packages\big_sleep\_init__.py", line 1, in <module>
from big_sleep.big_sleep import BigSleep, Image
File "C:\Users\hoge\anaconda3\envs\painter\lib\site-packages\big_sleep\big_sleep.py", line 26, in <module>
assert torch.cuda.is_available(), 'CUDA must be available in order to use Big Sleep'
AssertionError: CUDA must be available in order to use Big Sleep
Other than that, I reinstalled Anaconda and tried again, and installed the driver from the NVIDIA site instead of via conda, but it didn't work.
python machine-learning windows-10 gpu cuda
Can't you install -cpytorch from PyPI with -ctorch?
I solved myself.
I started everything from scratch and it worked.
© 2024 OneMinuteCode. All rights reserved.