Add evaluation inside training script

[segalinc]

I would be interesting to add an evaluation part in the train.py script to also long in the tracker.
This will generate images based on save_image_epoch config and log them in tensorboard or desired tracker providing the prompts. Similar to the inference part but without the need to convert checkpoint to diffusers and load a pipeline

[DuoLi1999]

Have you implemented it yet? It would be fantastic if we could generate intermediate images during the training process.

[segalinc]

Yes I did. Might be helpful to double check I did it right. Maybe I can create a PR for this or upload my script. I also added logging to comet-ml if that's something that can be useful. My script used precomputed latents tho so it may be a bit different from the base code.

…

On Sun, Feb 4, 2024, 11:57 PM Duo Li ***@***.***> wrote: Have you implemented it yet? It would be fantastic if we could generate intermediate images during the training process. — Reply to this email directly, view it on GitHub <#79 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACHLPEWOS3M57NZ2Q4QO3ATYSCGHNAVCNFSM6AAAAABCJOOLUKVHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4DGNRWGYYDE> . You are receiving this because you authored the thread.Message ID: ***@***.*** com>

[lawrence-cj]

Check this one: https://github.com/PixArt-alpha/PixArt-alpha/blob/master/train_scripts/train_diffusers.py

[segalinc]

this is my version https://gist.github.com/segalinc/dca2ecd8ce3abb1c01ad231f36e28fd4

[lawrence-cj]

Cool. I like it. Have you tested it throughly? And would you mind pull a PR for this changes?

[segalinc]

Sent PR