Since Google proposed the Transformer architecture in 2017, AI development has gradually stepped into the era of pre-training large models.In June 2018, the number of OpenAI's GPT model parameter has reached 117 million, and the number of model parameter has begun to realize the development of a billion base flyover, which on average doubles every 3-4 months, thus bringing the demand for training arithmetic is also "At the end of 2022, after OpenAI's GPT model emergence capability, the AI industry rapidly entered the AIGC era, which is technically supported by large models.