3 ms·
Let's Code a Q/A Model From Lex Fridman's Podcasts [video]
- jawerty 3y agoHey HN! In this video I scrape audio from podcasts (for now George Hotz and Lex Fridman as an example) to generate a dataset for a Retriever/Reader system (using a small home grown vector db and LLama 2). I also go through how to load and prompt LLama 2 using the retrieved context and getting an answer. Check out my last video if you want to see how to fine tune LLama 2 on a Google Colab single GPU.