Proposal: Contribution of Backend and Device Selection for Inference, and Suggestion to Add LoRA and Model Fine-Tuning Support

[DenisMontes]

Hello everyone,

I would like to make myself available to contribute to the project. However, I was unable to find a roadmap or a "TO DO" list where I could identify areas to work on and submit new pull requests.

In the meantime, I have made some modifications in a local branch, where I implemented functionality that allows users to select both the backend and the device to be used for inference. I believe this could be a useful feature for the community. Therefore, I would like to ask if you consider it appropriate for me to submit these changes, or if you would prefer that I focus on other pending issues.

Additionally, I would like to propose adding support for LoRA and model fine-tuning. I am currently developing a tool designed to create characters using Stable Diffusion and subsequently train LoRA models—or even the base model itself—so that it can consistently generate the same character.

I would appreciate hearing your thoughts and feedback on these proposals.

[wbruna]

Talking only for myself here (and I'm just a minor contributor), but I don't think a PR would not be appropriate - especially if it's fully implemented and working already! I have missed myself a way to force the inference to run on CPU.

So, I'd suggest submitting it anyway, to see how well it's received.