Platypus2 13B
About
Platypus2-13B is an advanced large language model built on the LLaMA 2 transformer architecture, fine-tuned for instructions. Developed by Cole Hunter and Ariel Lee, it's primarily aimed at English language tasks and is trained with a STEM and logic-centric dataset, garage-bAInd/Open-Platypus. The model excels in text and code generation, as well as conversational AI, utilizing LoRA (Low-Rank Adaptation) for efficient training on a single A100 80GB GPU. Despite its potency, it faces challenges like potential biases and inaccuracies, particularly outside English contexts, necessitating careful usage and safety evaluations. Quantized versions by TheBloke offer various optimizations for specific hardware, balancing size, speed, and accuracy, while performance varies across benchmarks like ARC and TruthfulQA.