Efficient Financial Language Understanding via Distillation with Synthetic Data

In three linesDistillation framework with synthetic data for financial sentiment analysis. Knowledge transfer from large instruction-tuned teacher to compact student models. Clustering-based seed selection generates synthetic examples via few-shot prompting. Compact model outperforms teacher on complex/noisy text with minimal supervision.

Summary generated by Claude — human-verified