Srikar Kashyap Pulipaka

Synthetic data augmentation and per-language threshold tuning can significantly boost the performance of LLMs on multilingual tasks, outperforming alternative architectures that showed promise on the development set.