关于Interface Design的问题

[kelseyfc]

data-copilot的设计很新颖。关于论文中提到的Interface Design这部分的内容，我有一些问题希望得到您的回复。

1. interface design可以理解为是通过prompt让LLM自动生成指定编程语言的代码吗？
2. 仓库中有关于这部分的代码吗？我好像没找到相关的代码和prompt。

[zwq2018]

嗯 interface design是让LLM通过写代码的形式自动生成一个个功能模块 因为接口设计取决于数据和需求，每个人都不太一样，所以没有开源接口设计代码，具体的方法和prompt在论文里已经很详细了。

[bzqweiyi]

嗯 interface design是让LLM通过写代码的形式自动生成一个个功能模块 因为接口设计取决于数据和需求，每个人都不太一样，所以没有开源接口设计代码，具体的方法和prompt在论文里已经很详细了。

for more detail we can define seeds, using chatgpt to generate multi request, while how to using this request automate generage the interface? and cluster the same meaning request? hope to repla, many thx.

[zwq2018]

@bzqweiyi 1-First design some seed requests and keywords according to your requirements, let LLM generate more requests by itself according to the kind of data, for example, combining different keywords

2-Use these seed requests to let LLM generate interfaces one by one, and then check whether the newly generated interfaces can be merged, i.e. merge interfaces with similar functions

3-Iterative processing for all generated requests

Please see our paper for details Besides, you also can refer to self-instruct , which is common method in NLP community