Low-cost dataset search for fine-tuning Llama on web tasks
The goal is to fine-tune a Llama model so it can turn a user request into a complete static web page or frontend component. The output needs to use HTML, CSS, and plain , work on its own, adapt to different screen sizes, and appear as a complete code block. The main problem is finding the right dataset.
Many datasets are either LeetCode-style algorithm problems or broken code snippets that do not come with matching user instructions. The needed dataset should follow the , with a clear instruction and a complete self-contained web page as the answer. The work is being done with limited student resources, including and free cloud GPUs from Kaggle or Colab.
Paid data labeling and heavy use of expensive proprietary APIs such as GPT-4 are not practical. A useful option would be a hidden dataset or an pipeline that filters existing codebases and creates matching instructions using free or APIs.
Key points
- The target is a Llama model specialized in complete frontend and static web page generation.
- The wanted data pairs a user instruction with a full HTML, CSS, and result.
- Common coding datasets are a poor fit because they focus on or incomplete snippets.
- The budget constraint rules out heavy use of paid labeling or expensive proprietary APIs.
- An filtering and reverse-prompting pipeline could help reduce setup cost.