Build a Llama 2 LangChain conversational agent The largest Llama 2 model, the 70B parameter model, has been designed to fit onto a single a100 GPU, requiring a minimum of 35 gigabytes of GPU memory.