GLMLog: Log Anomaly Detection Method Based on ChatGLM
Hong Zhou, Maosong Chen, Xiaomei Yang, Guang Chen, Yucheng Eason Zhang, Jinhui Yuan, Chao Wang · 2024
The massive scale of logs makes manual analysis of logs impractical. The log anomaly detection scheme based on deep learning also faces many challenges, including the diversification of log types and the rapid increase size of log data. The recent emergence of large language Models provides a new solution for log anomaly detection. Therefore, this paper proposes a novel log anomaly detection scheme based on ChatGLM2-6B, and we call it as GLMLog. We develop GLMLog on ChatGLM2-6B which is pretraininged. Furthermore, ChatGLM2-6B is fine tuned through LoRA fine-tuning using publicly available log data according to the needs of log anomaly detection. In order to more accurately detect the anomaly, GLMLog implements log anomaly detection from content sequences and event sequences. We completed validation testing using the logs provided by LogHub as the dataset. Experimental result shows that after a small number of rounds of fine-tuning, GLMLog can achieve log anomaly detection and its F1-score is basically consistent with Deeplog.