Nerogar
[Feat]: Lumina 2.0 finetune/lora support?
Describe your use-case.
Lumina 2.0 came out very recently, already has comfy support, i would like to be able to finetune it and create loras for it for my projects as it has Apache 2 license.
What would you like to see as a solution?
Add support for Lumina 2.0 finetuning and lora training.
Have you considered alternatives? List them here.
There's no alternatives to OneTrainer :P
As Comfy will no longer support "non-standard" Lora formats anymore, this likely needs to also wait for a proper Lora standard to be agreed upon by the training tools and generation tools, or else you will have a Lora with no place to use it besides the sampling tool of OneTrainer.
So at least finetuning then? They have ready-to-go scripts for it, altho i guess they won't be up to par with what we have in OneTrainer.
ai-toolkit merged support for that in their main branch today:
https://github.com/ostris/ai-toolkit/commit/0b8a32def7c483aac22e97056eddb61a88fe55a4