ChatYuan-large-v2 is a functional dialogue language model supporting both Chinese and English languages. The model builds upon v1 with optimizations in fine-tuning data, RLHF, and chain-of-thought reasoning. It features bilingual conversation, code generation, table generation, math operations, and refusal capabilities for harmful content. The v2 version supports up to 4096 tokens and can run on consumer GPUs, PCs, or even mobile devices with INT4 quantization as low as 400M parameters.