Leveraging Large Language Models For System Log Analysis - Fault Troubleshooting Radio Units Using Log Data

Sammanfattning

This thesis investigates the application of large language models (LLMs) for systemlog analysis, specifically focusing on fault troubleshooting in radio units usinglog data. The primary objective is to enhance the efficiency and accuracy of systemmonitoring tools through state-of-the-art AI techniques. The research exploresthe utilization of retrieval-augmented generation (RAG) frameworks and parameterefficientfine-tuning (PEFT) methods to process and summarize log data. By employingpre-trained models such as Llama2, Llama3 and Mistral, the study evaluatesdifferent implementations to summarize segments of logs as well as extracting relevantinformation from them.The findings demonstrate that LLMs can significantly automate and improve theanalysis of system logs, providing insights and facilitating easier troubleshooting.Additionally, the study examines the impact of enriching chatbot input data withcontextual information, leading to substantial performance improvements in specializeddomains. Despite the promising results, the research acknowledges limitationsrelated to the quality and structure of log data and the need for source-specificrefinements in context-enrichment methods.The contributions of this thesis are twofold: it presents a viable approach to leveragingLLMs for easier system monitoring and highlights the critical role of contextin enhancing chatbot functionalities. Future research directions include integratingmore advanced models, fine tuning existing models and exploring other state-of-theartmethods to optimize retrieval-augmented generation pipelines.