Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Please add LLaVA #45

Open
Xyndra opened this issue Nov 15, 2023 · 11 comments
Open

Please add LLaVA #45

Xyndra opened this issue Nov 15, 2023 · 11 comments

Comments

@Xyndra
Copy link

Xyndra commented Nov 15, 2023

Please I want LLaVA.
https://llava-vl.github.io/

@eletroswing
Copy link

I loved the idea of ​​having this opensource and completely free

@AltayYuzeir
Copy link

Maybe even combination between LLaVA 1.5 and DeepSeek Coder.

@Anon2578
Copy link

Yes this is a very cool idea I’ll be here for it once it’s up and ready

@eletroswing
Copy link

@AltayYuzeir how llava 1.5 and deepseek coder can be implemented to work together or what it would look like in theory?

@AltayYuzeir
Copy link

Hmmm, perhaps I was inaccurate in my statement. I have been thinking about it, and I cannot comprehend direct communication on the neural network level between the two LLMs. I came up with two concepts:

  1. If LLaVA is asked to produce a detailed description of the drawing, also including spacing between elements and alignments, the description can be passed on DeepSeek Coder
  2. Maybe similar to LLaVA-Med, LLaVA can be fine-tuned on UI layouts, HTML and CSS

@eletroswing
Copy link

@AltayYuzeir I'm going to create a dataset of 'images to tailwind' using deepseek coder, and then we can train this custom model, what do you think?

@AltayYuzeir
Copy link

I am not the best AI or UI programmer, but I believe such resource will be beneficial in any case for the future. As far as I understand it, Tailwind is a high-level framework on top of CSS, featuring custom classes absent in native CSS. Tailwind is not explicitly supported on DeepSeek Coder, as per info from their GitHub page. I was thinking it might be perhaps more flexible to use DeepSeek to generate more general webpages built on top of raw HTML and CSS. Please, correct me if I am wrong.

@eletroswing
Copy link

@AltayYuzeir You are right! Similar to what vercel is doing with v0.dev. Anyway, I'm going to continue creating the dataset using tailwind used in public projects!

@AltayYuzeir
Copy link

I guess one could make a dataset for fine tuning based on UIs made with Dash (Python) and Shiny (R)?
I wonder about the capabilities of DeepSeek Coder for Dash and Shiny specifically, since they are libraries for their respective languages, but are also frameworks, which function quite differently than most packages.

@orkutmuratyilmaz
Copy link

@eletroswing and @AltayYuzeir, supporting Ollama can solve all these requirements at once. Please check #52

@bocklucas
Copy link

#99

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

6 participants