Nous Research 发布 Arete-27B：基于 Qwen3.8-27B 微调降低宣传倾向
Nous Research 把 Qwen3.8-27B 微调成了宣传率只有 1% 的版本，权重和数据都开源了，做本地部署的可以看看。
Nous Research 将 Qwen3.8-27B 进行后训练，得到 Arete-27B。团队构建了覆盖言论自由、民主、隐私等七个价值类别的 803 条提示词评测，用于衡量模型采纳中国官方立场的频率。经过约 5300 万 token 的微调后，模型宣传率从 22% 降至 1%，与所测西方模型持平。Hermes Bench 成绩显示微调没有明显性能损失。评测、权重和训练数据均已开源。
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: https://t.co/iAOANQTVZ6
- Decoder12:32原文
