Microsoft announced the open source of Deep Speed ​​Chat to help users train large language models such as ChatGPT, so that everyone can have their own ChatGPT.

It is reported that Deep Speed ​​Chat is developed based on Microsoft's Deep Speed ​​deep learning optimization library, with training, enhanced reasoning and other functions. It also uses RLHF (reinforcement learning with artificial feedback mechanism) technology, which can increase the training speed to more than 15 times that of the existing system, but greatly reduce the cost. For example, a Chat-like GPT model with 13 billion parameters can be trained in just 1.25 hours. (Planet Daily)