[Examples] Update `unconditional_image_generation/train_unconditional.py` to include Weights and Biases Logger

[ash0ts]

Unborked PR of #1209

The current example uses Tensorboard as the only available logger. Adding Weights and Biases allows us to interact with the model metrics, alongside the saved model predictions and logged model checkpoints all in accessible dashboards. This makes viewing my batches of experiments and images a lot more streamlined. The user-facing addition adds an argument for logger which can take one or more of the available options which are currently only [wandb, tensorboard]. An example dashboard trained over few epochs can be seen here. Using both tensorboard and wandb generates an embedded tensorboard instance within wandb.

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[panenyaun]

hello,May I ask you how to combine train _ without. py with LDMPipline，Do you have the source code? Thank you

[patrickvonplaten]

hello,May I ask you how to combine train _ without. py with LDMPipline，Do you have the source code? Thank you

Hey @panenyaun,

Could you please open a new issue for this as it's not really related to this one here?

[panenyaun]

hello,May I ask you how to combine train _ without. py with LDMPipline，Do you have the source code? Thank you

Hey @patrickvonplaten Ok,"how to combine train_unconditional.py with LDMPipline" is my issue,Looking forward to your answers

[ash0ts]

Hey! I made changes based on your suggestions. How does this commit look? https://github.com/huggingface/diffusers/commit/be5d75bf8280993136440665b7b15c7ec0265dc9

@patrickvonplaten @patil-suraj @anton-l

[ash0ts]

Hello @patrickvonplaten! There seems to be a confusion. I linked a new commit that lives on a different branch:

This being the commit: https://github.com/huggingface/diffusers/commit/be5d75bf8280993136440665b7b15c7ec0265dc9

before raising a new PR. I can raise the new PR instead if that's easier for review. In this commit we remove the multi-logger option and show off the HF custom logger functionality to keep the training loop cleaner and more in line with the intended HF "tracker" extensibility

[ash0ts]

Hey, @patrickvonplaten there seems to be a miscommunication on my end. I added a new PR with a minimal implementation of metric tracking. We can later discuss what a more full-fledged example may look like inside of accelerate!

NEW PR: https://github.com/huggingface/diffusers/pull/1719