Hugging Face accelerates Witty Works Writing Assistant with SetFit and mpnet-base-v2 on Azure
AI Impact Summary
Hugging Face guided Witty Works to switch from vanilla transformer embeddings to Sentence Transformers, enabling context-aware classification of non-inclusive words with far less labeled data. The team adopted SetFit for few-shot fine-tuning and used mpnet-base-v2 embeddings with a simple classifier (logistic regression or KNN), achieving 0.92 accuracy with only 15-20 labeled sentences per word. They deployed the model on Azure after a Colab prototype, achieving low latency suitable for real-time suggestions. This collaboration demonstrates how leveraging pre-trained sentence embeddings and efficient fine-tuning can dramatically shorten ML development cycles while controlling labeling costs.
Affected Systems
- Date
- Date not specified
- Change type
- capability
- Severity
- info