Supervised Finetuning

Supervised finetuning (SFT) adapts pre-trained large language models (LLMs) and other foundation models to specific tasks by training them on labeled data. Current research emphasizes improving SFT's efficiency, addressing issues like catastrophic forgetting (loss of pre-trained knowledge) and the high cost of data annotation, often exploring techniques like experimental design and parameter-efficient fine-tuning methods. This work is crucial for enhancing the performance and applicability of powerful models across diverse domains, from medical image analysis and speech recognition to improving the alignment of LLMs with human preferences and educational goals.