Finding high quality data 📚 is one of the biggest issues in LLM training today. HoleFill🍩 is a simple data curation method that leverages user chatlogs to find online data sources that “fill” the knowledge holes in an LLM. You can try HoleFill🍩 in the interactive colab👇🧵
2
4
30
3K
21
Download Image
Run HoleFill🍩 on your own ChatGPT data to create a Custom GPT✨ that’s expert-level on your interests! And if you’re an LLM chat company, you can run it on all your chatlogs to train an LLM that fills the holes found by all your users 📈🚀 colab.research.google.com/drive/1oVqU9CY…
@khoomeik Interesting approach, well done. Do a one-click integration, then call it "hole in one" ahah ⛳️