The researchers introduce “Platypus,” a state-of-the-art refined Large Language Model (LLM) that has achieved top rank in the HuggingFace Open LLM Leaderboard. The strength of Platypus stems from the unique “Open-Platypus” dataset, a curated subset of other public datasets. The dataset emphasizes improving LLMs’ proficiency in STEM and logic. Furthermore, Platypus showcases efficiency by allowing a 13B model to be trained using just 25k questions in a short span of 5 hours on a single A100 GPU. This achievement highlights not just the efficiency of the model but the quality and potential of the Open-Platypus dataset.
Publication date: 14 Aug 2023
Project Page: https://platypus-llm.github.io
Paper: https://arxiv.org/abs/2308.07317v1
