DAINAPP
patreon
New Github project & more.
🕑 Added 2023-05-02 18:33:41 +0000 UTCHi everyone, currently I've just been keeping my head down and focusing on work, and some projects are almost ready. First of all, I've just released a project on GitHub: https://github.com/BurguerJohn/torch-felix
I hope it has some usefulness for developers who train their own AI models. It is based on functions used to train upscalers and (in my case) the Rife-App. It contains various improvements that I've developed while working on the Rife-App and allows the use of more current models than the VGG (typically used for this kind of project).
As I continue developing the Rife-App, I plan to add new improvements to it.
Of course, I'm also training a new Rife-App model with these improvements. There's still a long way to go in the training, not even reaching 50% yet, but I have high expectations for these new models.
I'm also finalizing a GUI for the "Bark" project, which transforms Text -> Audio. It allows creating voices and even a bit of music, although it's still somewhat limited, I think it has a lot of potential.
Slowly, I'm cleaning up the Stable Diffusion code, but I should release a GitHub project for it soon.
The chat projects are more confusing, with changes literally every day, improvements from previous projects, or confusion about using Llama as a base. It's kind of difficult to focus on one project to work on, but that makes sense considering the popularity of ChatGPT. Most likely, I'll end up joining an open-source solution and create some tools to facilitate the use of such a project for you all. I'll focus on that as soon as I finish the Stable Diffusion GUI on GitHub.
Comments
GRisk
For the SD GUI suggestion, it would need a new model for that, SD don't support something like this by default =/ If there is any model that do that, I can try to add it. About Whisper, I did managed to test some improvements back then, but it broke more stuff than it fixed. I can give it another shot now to see if it was improved.
GRisk
Thanks for the links. At least for StoryWriter is pretty much impossible to run on a regular computer. It's a shame, it look awesome.
cool1
Also for the stable diffusion GUI could there please be an option, if possible, for "image to prompt" so you could give an image and it would tell you what prompt would give similar image to the one you loaded? A while ago you said "I'm also working on a small update to Whisper, to improve timing on the subtitles.". Is there any updates on that please?
cool1
For the chat/language model projects, there's this that has been released recently: https://www.mosaicml.com/blog/mpt-7b It's MPT-7B set of language models. It includes MPT-7B Instruct, MPT-7B-Storywriter-65K (which might be allow around 48750 words+ I assume for fiction) and MPT-7B-Chat and it's open source for commercial use. Though it might not be properly optimized yet / might need a lot of VRAM for some things. If these are good language models to use it would be good if there could be a GUI interface for them if possible.
