The chujiezheng/Llama3-70B-Chinese-Chat-ExPO model is a 70 billion parameter Llama 3-based language model, extrapolated using the ExPO method (alpha = 0.3) from shenzhi-wang/Llama3-70B-Chinese-Chat and Meta-Llama-3-70B-Instruct. This experimental model aims to achieve superior alignment with human preference by applying extrapolation to SFT and DPO/RLHF checkpoints. It is specifically adapted for Chinese chat, though its Chinese capabilities are still undergoing comprehensive evaluation.
No reviews yet. Be the first to review!