Hugging Face enables Witty Works Writing Assistant via SetFit and mpnet-base-v2 on Azure

AI Impact Summary

Hugging Face steered Witty Works away from vanilla transformers to a sentence-embedding approach using Sentence Transformers, with SetFit enabling few-shot fine-tuning. The team achieved 0.92 accuracy using 15-20 labeled sentences per word and deployed mpnet-base-v2 on Azure, coupled with logistic regression and KNN for efficient, low-latency inference. This accelerates the ML workflow, reduces annotation cost, and enables a bias-aware writing assistant to reach production faster with scalable performance on Azure.