Nous Research fine-tuned Qwen3.8-27B to launch Arete-27B, lowering the promotional response rate from 22% to 0.9%.
Nous Research conducted approximately 53 million tokens of supervised fine-tuning on Alibaba's Qwen3.8-27B to release the open-weight model Arete-27B. This reduced the proportion of promotional responses from 22% to 0.9%, while improving factual accuracy scores from 2.32 to 2.98.
Return to original text
A short introduction is retained here; the full text from the source website is not reposted. For specific wording and subsequent updates, please refer to the original article.
Open AIHOT original