In less than a year, I tried a number of large models at home and abroad, and found that each has its own merits. Recently, I tested the open source version of Tongyi Qianwen and Alibaba Cloud's 720 parameter large model, and the results were unexpected, even surpassing the closed-source large model. This open-source large model performs eye-catching in mathematical problems, such as 0.999 infinite loop and 1 which is greater?, as well as college entrance examination math problems, which are close to the human way of thinking in terms of results and processes. In terms of Chinese performance, semantic understanding is smarter, and it can even solve ethical problems and give wise suggestions. The 72 billion parameter model is expected to become the best choice for enterprise-level applications, and it is currently the best open-source model in China, surpassing the foreign benchmark Llama 2-70B. The emergence of domestic large models, the road to open source is becoming more and more open, which is a good signal for technology sharing, and domestic large models are bound to move towards greater progress!
