npub1jpp36eeycvlkkdy4d4l4lgswvf8jtdare5vly0w0sk7rc39pzgrqlyu59l (npub1jpp…u59l) general consensus these days is that training a custom model on your own data isn't the right approach - you probably want RAG, Retrieval Augmented Generation
I need to write more about that - here's a few links:
- https://simonwillison.net/2023/Aug/27/wordcamp-llms/#retrieval-augmented-generation
- https://til.simonwillison.net/llms/embed-paragraphs