Deployment and Comparison of Large Language Models Based on Virtual Cluster

[Objective] Currently, large language model (LLM) is one of research highlights in the field of natural language processing. This paper selected some open-source LLMs for deployment and comparison from the perspective of consumer-grade GPU and support for Chinese and English. [Coverage] This paper uses keywords search and citation secondary search to collect papers and information from international computer journals, conferences and open source code warehouse. [Methods] From the perspective of supporting both Chinese and English, we selected LLaMA, MOSS, ChatGLM-6B, ColossalChat, and Chinese-LLaMA-Alpaca for deployment at the same virtual task, on the virtual cluster with consumer-grade GPU. Furthermore, we made horizontal comparisons on semantic understanding, logical reasoning, code programming, ancient poetry, and legal questions, and then, discuss the advantages and disadvantages of these models. [Results] Limited parameters scale, most of them are not very friendly to support Chinese, have weak Chinese understanding abilities, and have varying abilities in logical reasoning. At present, researchers have paid less attention to issues such as Chinese support and resource consumption. They generally focus on increasing the scale of model parameters and using higher graphics card resources for model training and inference. [Conclusions] Although the development of LLMs is rapid, many models do not fully support Chinese. Understanding Chinese ability needs to be further improved, and more efforts need to be made in logical reasoning. It is believed that in the future, there will be more large language models that consume lower resources and support stronger Chinese.