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…
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…