Researchers from Boston University have developed Platypus, a family of Large Language Models (LLMs) that have topped the HuggingFace’s Open LLM Leaderboard. The team used a procedure called parameter efficient tuning on the Open-Platypus dataset to enhance the models’ performance. The Platypus models, which vary in size, demonstrate exceptional performance in quantitative LLM metrics, achieving high efficiency with minimal fine-tuning data and computational resources. The team also explored the challenge of data contamination in LLM training sets and shared their data filtering process.
Read more at MarkTechPost…