Llama 2: Open Foundation and Fine-Tuned Chat Models
Deep Papers
00:00
How to Build a Chat Model From Scratch
A lot of it was borrowed from previous papers so I wouldn't say that there is, I didn't feel like there's anything totally novel in the architecture itself. You know, that that would be reserved for more foundational changes in the way, you know, humans think about things we have a question in the channel from Manish. And so, just about whether they built this model from scratch or if they use some base model and fine tune with training and human feedback. What I gathered from the paper is that they used the same methodologies that they use for llama one,. But with a new set of pre training data or training data that they gathered with a different pre training process.
Transcript
Play full episode
