Utility-Diversity Aware Online Batch Selection for LLM Supervised Fine-tuning
Researchers have developed a new framework called UDS (Utility-Diversity Sampling) for more efficient supervised fine-tuning of large language models (LLMs). This method addresses limitations in existing techniques by considering both the utility and diversity of data samples, rather than just utility alone. UDS also avoids reliance on external resources like reference models or validation sets, and it integrates efficiently into the training process without incurring extra time. AI
IMPACT This new sampling method could lead to more efficient and effective LLM fine-tuning, reducing computational costs and improving model performance.