cannot install on linux mint

[duangdoock]

Hi everyone
I installed and tested successfully iw3 on Windows and would like to do it on Linux Mint.
I did all these following steps in a terminal :

sudo apt-get install git-core libmagickwand-dev libraqm-dev
git clone https://github.com/nagadomi/nunif.git
cd nunif
pip3 install -r requirements-torch.txt
pip3 install -r requirements.txt
sudo apt-get install python3-wxgtk4.0

but I get some errors after typing those following ones :
python3 -m waifu2x.download_models
python3 -m waifu2x.web.webgen

which give me this results :

Traceback (most recent call last):
File "/usr/lib/python3.10/runpy.py", line 187, in _run_module_as_main
mod_name, mod_spec, code = _get_module_details(mod_name, _Error)
File "/usr/lib/python3.10/runpy.py", line 110, in _get_module_details
import(pkg_name)
File "/home/lezamours/nunif/waifu2x/init.py", line 9, in
from .utils import Waifu2x
File "/home/lezamours/nunif/waifu2x/utils.py", line 3, in
import torch
File "/home/lezamours/.local/lib/python3.10/site-packages/torch/init.py", line 290, in
from torch._C import * # noqa: F403
ImportError: libnccl.so.2: cannot open shared object file: No such file or directory
Traceback (most recent call last):
File "/usr/lib/python3.10/runpy.py", line 187, in _run_module_as_main
mod_name, mod_spec, code = _get_module_details(mod_name, _Error)
File "/usr/lib/python3.10/runpy.py", line 110, in _get_module_details
import(pkg_name)
File "/home/lezamours/nunif/waifu2x/init.py", line 9, in
from .utils import Waifu2x
File "/home/lezamours/nunif/waifu2x/utils.py", line 3, in
import torch
File "/home/lezamours/.local/lib/python3.10/site-packages/torch/init.py", line 290, in
from torch._C import * # noqa: F403
ImportError: libnccl.so.2: cannot open shared object file: No such file or directory

Any idea? Thanks in advance for your help!

[nagadomi]

This looks like a pytorch problem, maybe an older pytorch version is installed. (As far as I know, all necessary binaries should be bundled in pytorch wheel package)
Could you try with venv? or use --upgrade option with pip3 install command.

use venv

1. create .venv

python3 -m venv .venv

2. activate .venv (Note that this is required every time you run nunif commands)

source .venv/bin/activate

3. Then install and run it.

pip3 install -r requirements-torch.txt
pip3 install -r requirements.txt
#  select the version that best matches your OS https://extras.wxpython.org/wxPython4/extras/linux/gtk3/
pip3 install -f https://extras.wxpython.org/wxPython4/extras/linux/gtk3/ubuntu-22.04/ wxpython

use --upgrade option

This method potentially destroys the existing pip environment.

pip3 install --upgrade -r requirements-torch.txt
pip3 install --upgrade -r requirements.txt

[duangdoock]

Finally managed to run iw3 but when starting the process of conversion I get this error :

CUDA out of memory. Tried to allocate 138.00 MiB. GPU 0 has a total capacity of 3.94 GiB of which 21.19 MiB is free. Including non-PyTorch memory, this process has 3.62 GiB memory in use. Of the allocated memory 3.44 GiB is allocated by PyTorch, and 120.37 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

[nagadomi]

As the error message says, it is VRAM out of memory.
It seems that VRAM is only 4GB, so try to decrease batch-size, use a smaller model, or turn on Low VRAM.

[nagadomi]

Also if you use an older GPU, FP16 may not work.
#200

[duangdoock]

I followed your suggestions but I still get the same message. It seems to be complicated to make iw3 work on Linux. It may be better for me to get back on Windows. Thanks for your help

[nagadomi]

Once out of memory occurs, the used memory may not be released, so restart the program and try again.
Windows has Shared GPU Memory(Virtual VRAM), so it is less likely to be out of memory error, but if Shared GPU Memory is used it will be very slow.
well, use whichever you prefer.