Open unixman0x6576676C313531303933 opened 1 month ago
https://github.com/gitmylo/audio-webui/issues/231#issuecomment-2226785978 might be helpful for now. Until an actual fix is made.
Thank you ! It worked, eventually...) Had to also resolve dependencies for torch and cuda.. Somehow after executing run.bat (even in a fresh venv and install) I always ended up with the wrong version of cuda for PyTorch, maybe it's a pip thing with it trying to resolve the dependencies.. but I don't really know. Also had problems with fairseq and audiocraft requiring conflicting dependencies. Namely fairseq 0.12.4 requires hydra-core<1.1,>=1.0.7, but after the install process I always winded up with hydra-core 1.3.2 and omegaconf 2.3.0. The workaround with fairseq was to first uninstall all that was conflicting and then:
pip install hydra-core==1.1
pip install audiocraft==1.3.0
pip install fairseq==0.12.2 --no-deps
pip show hydra-core audiocraft fairseq
and then finally to satisfy all other fairseq dependencies using
pip install bitarray cffi cython numpy regex sacrebleu torch torchaudio tqdm
In the end I ended up with hydra-core 1.1 and omegaconf 2.1.2,
pip install audiocraft==1.3.0
ERROR: Could not find a version that satisfies the requirement torch==2.1.0 (from audiocraft) (from versions: 2.2.0, 2.2.1, 2.2.2, 2.3.0, 2.3.1, 2.4.0) ERROR: No matching distribution found for torch==2.1.0
Digging into that apparently audiocraft can't get their act together and provide a working version either. Trying to install their stable version gives a different error that is supposedly down to requirement for python 3.9, which doesn't actually solve the issue either. I have most 3.x versions installed, no dice. The errors setuptools spits out speak volumes about the code quality on their end as well.
At some point it becomes a question of freezing dependencies or outright shipping them as stale versions instead of trying to rely on others to provide stable releases to depend on. :/
@gitmylo
Temporarily:
+import subprocess
+import sys
+def install(package):
+ subprocess.check_call([sys.executable, "-m", "pip", "install", package])
for e in em.states.values():
e.activate()
+ install('huggingface-hub==0.23.2')
print('Preparing')
Can also add:
install('--no-cache-dir torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --extra-index-url https://download.pytorch.org/whl/cu118')
In the cases where that doesn't install properly for some reason I guess.
Also, it hates spaces or special chars in folder names, so don't do "New Folder" or "$git" as folder names... yeah I should know better on that.
Hello, sadly the installer does not work for me( Really want to try the project, but always get the same error... be it a fresh Docker install or a regular one
huggingface-hub>=0.23.2,<1.0 is required for a normal functioning of this module, but found huggingface-hub==0.23.0. Try:
pip install transformers -U
orpip install -e '.[dev]'
Which makes me wonder in circles, because when I install the correct version with `pip install transformers -U entering the virtual environment as the installer suggests via activate_only.bat , the next execution of run.bat installs huggingface-hub==0.23.0 and I am back at square one. I tried locking the required version of huggingface-hub, but i do not have any experience with python and this all is very new to me.
Any help will be very appreciated