audio-diffusion-pytorch-trainer
audio-diffusion-pytorch-trainer copied to clipboard
archinetai
Error executing job with overrides: ['exp=base_test']
Hello! I tried to use " python train.py exp=base_test " in Eshell, but some errors were found as shown follows:
Yeah, I'm having the same issue. I can see that datamodule isn't in config.yaml, so I guess it's just broken...
Not a huge deal to write a new training script.
Yeah, I'm having the same issue. I can see that
datamoduleisn't inconfig.yaml, so I guess it's just broken... Not a huge deal to write a new training script.
Thanks, so you suppose that I need to write a new training script with datamodule?
I am facing this issue right now too! Can you please provide some pointers on what needs to be changed? The config yaml file or train.py?
I updated the base_test.yaml it should work now. Make sure to have the latest versions installed (pip install audio-diffusion-pytorch audio-encoders-pytorch -U)
Okay, running now. Thanks!
I'm curious; since I'm not familiar with Hydra or OmegaConf (yet) I'm not really clear what I'm training on when running base_test.yaml. Is this training on my files at DIR_DATA in my .env, or it is something from YouTube (just looking at the contents of base_test.datamodule).
Any clarification appreciated.
It downloads a single song from YouTube as a test, you can see the url in the base_test.yaml config.
Okay, I assumed it must download something, but does it only use that song during this test training? Mostly I'm confused about the step of adding my data path to .env if it's just going to train on one YouTube song.
What would I change to use my own data (i.e., the path I specified in .env)?
I see that base_small.yaml references WAVDataset... is that what I'd use to point to DIR_DATA?
UPDATE: I've been looking at train.py, but I see that I can probably get a bit further with the notebook...
I updated the
base_test.yamlit should work now. Make sure to have the latest versions installed (pip install audio-diffusion-pytorch audio-encoders-pytorch -U)
Hello. I have tried to run this script again, but our computer server () is in china where the net was forbidden to browse or download information from Youtube. ( From the pic you can see the details that shell shows me “DownloadError” ) so what do I can to change the data path to another url to download the song? (i.e., the path I specified in .env or base_test.yaml)?
I updated the
base_test.yamlit should work now. Make sure to have the latest versions installed (pip install audio-diffusion-pytorch audio-encoders-pytorch -U)
Thank you for the attempted fix! Unfortunately the current commit (d5f6870624ae38e25704cb3df3f148507cf8cf51) still experiences this error. I've put the full error below in text form so that it can be more easily searched by other people with the same issue:
Expand for full error
(venv7) quoththeraver@ip-26-0-139-220:~/workspace/audio-diffusion-pytorch-trainer$ python train.py exp=base_test [2023-01-12 20:35:59,628][main.utils][INFO] - Disabling python warnings!During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/admin/home-quoththeraver/workspace/audio-diffusion-pytorch-trainer/venv7/lib/python3.8/site-packages/hydra/_internal/utils.py", line 650, in _locate
obj = import_module(mod)
File "/usr/lib/python3.8/importlib/init.py", line 127, in import_module
return _bootstrap._gcd_import(name[level:], package, level)
File "
The above exception was the direct cause of the following exception:
Traceback (most recent call last): File "/admin/home-quoththeraver/workspace/audio-diffusion-pytorch-trainer/venv7/lib/python3.8/site-packages/hydra/_internal/instantiate/_instantiate2.py", line 134, in _resolve_target target = _locate(target) File "/admin/home-quoththeraver/workspace/audio-diffusion-pytorch-trainer/venv7/lib/python3.8/site-packages/hydra/_internal/utils.py", line 658, in _locate raise ImportError( ImportError: Error loading 'main.module_base.Datamodule': SystemError('initialization of _internal failed without raising an exception')
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "train.py", line 110, in
Specifically, this happens when following the install and run instructions in the readme (with a fresh virtual environment). Interestingly, I was able to successfully run the experiment on my local machine which already had plenty of packages installed, so there is clearly something wrong with the requirements. When I install the packages below (from my local machine's full pip freeze) I am able to run the experiment on the remote machine.
I'd submit a PR with this updated requirements.txt, but obviously that would be pretty messy since not all of these packages are required. In case the issue was simply versioning, I did make a version of the repo's requirments.txt with explicit versions based on what I had in my local environment (see below) but unfortunately that also failed which means that there are some missing packages.
requirements with versions (fails).txt
Because I am now currently unblocked I think I'll end my debugging here, but I wanted to share this info with you guys so that future people are unblocked and maybe someone will go through each package to see where the problem is :P
When this error manifested to me, it was related to the fact that datamodule in the hydra config was a child of exp and not a root key. As such it had to be accessed as config.exp.datamodule and not config.datamodule, and the same for all keys under exp. The problem was that # @package _global_ was missing at the beggining of the yaml configuration! So if you delete that by mistake this error happens!
