Hacker News new | ask | show | jobs
by valine 665 days ago
If you have the model weights you have roughly the same opportunities as the company that trained the model. The code you need to run inference on the Llama weights is very much open source. The only thing you're missing out on is the training code, which is prohibitively expensive to run for most anyways. Open source training isn't going to give you any unique insights into the "digital brain library" of your model.

Also just to be clear, if you want to set up a RAG with an open weight model and a large dataset there's nothing stopping you. Download Red Pajama and Llama and give it a try.

https://github.com/togethercomputer/RedPajama-Data