News for hackers. Stuff that matters.

manny_c

Re:OpenAI shows 'Strawberry' to feds, races to launch it on Friday August 30, 2024 @05:07AM (#41398953)
Ollama 3.1 405B model already uses synthetic data for SFT. They basically trained a bunch of models to generate, correct, and reject synthetic data for various tasks. They cover it in section 4.3 in the "The Llama 3 Herd of Models" paper.