Edit model card

Llama3-70B-Orion-Chinese-SE

概述

本模型是从failspy/Llama-3-70B-Instruct-abliterated-v3经过DPO微调而来

相较于Orion-zhen/Llama3-70B-Orion-Chinese, 本模型精简了DPO数据集, 力求对原模型影响最小的情况下获得较好的中文性能和emoji体验

本模型的emoji触发概率高于Orion-zhen/Llama3-70B-Orion-Chinese, 在其他方面稍逊于后者

Downloads last month
0
Safetensors
Model size
70.6B params
Tensor type
BF16
·

Datasets used to train Orion-zhen/Llama3-70B-Orion-Chinese-SE

Collection including Orion-zhen/Llama3-70B-Orion-Chinese-SE