OpenChat 2

About

OpenChat 2 is a conversational AI model built on the LLaMA architecture, meticulously fine-tuned using around 80,000 cleaned ShareGPT conversations. It is known for generating human-like responses and effectively engaging in multi-turn dialogues, making it ideal for chatbots and virtual assistants. The model demonstrates strong performance, with an 87.1% win-rate on the AlpacaEval benchmark and a 48.1% win-rate on the MT-bench, showcasing its competitive prowess among open-source models 256. Despite these strengths, OpenChat 2 does have limitations in complex reasoning, mathematical tasks, and programming challenges, sometimes leading to inaccurate information outputs known as "hallucinations" 347.

Capabilities

Rankings