Hello, I’m a full-stack developer working on projects for clients. Many of these clients have their own design systems, which may include custom components or customized frameworks like NextUI. My idea was to use fine-tuning to help the model understand and adapt to these design systems. This involves describing the components within the system, providing code examples for various component variations (e.g., from Storybook or Markdown documentation), and showing practical examples from the codebase. The ultimate goal is to have a model that can respond appropriately within the context of a client’s project and even generate new UIs using the correct components.
My initial attempts at fine-tuning seemed promising, but I encountered some challenges when prompting the model. While it recognized individual components well, it struggled to grasp the larger context. For instance, when I fine-tuned it with components like Buttons, Inputs, Variants, and Colors and then tried to generate a LoginForm, the model defaulted to standard React instead of utilizing the design system. However, I did find that with more specific prompts, the model started to use the design components effectively.
I have a few questions:
- Is fine-tuning the right approach for this task, or is it better suited for something else?
- Has anyone else tackled a similar challenge and is open to sharing insights?
- Additionally, are there any valuable articles, blogs, videos, papers, or resources that delve into this topic further?
During my research, I mainly came across information on adjusting the model’s tone but not many real-world examples of teaching it new concepts.
I hope this helps, and if you have any more questions or need further clarification, please feel free to ask.