Large Language Model-Based Optical Network Log Analysis Using LLaMA2 with Instruction Tuning DOI
Yue Pang, Min Zhang, Yanli Liu

и другие.

Journal of Optical Communications and Networking, Год журнала: 2024, Номер 16(11), С. 1116 - 1116

Опубликована: Окт. 2, 2024

The optical network encompasses numerous devices and links, generating a significant volume of logs. Analyzing these logs is for optimization, failure diagnosis, health monitoring. However, the large-scale diverse formats present several challenges, including high cost difficulty manual processing, insufficient semantic understanding in existing analysis methods, strict requirements data security privacy. Generative artificial intelligence (GAI) with powerful language generation capabilities has potential to address challenges. Large models (LLMs) as concrete realization GAI are well-suited analyzing DCI logs, replacing human experts enhancing accuracy. Additionally, LLMs enable intelligent interactions administrators, automating tasks improving operational efficiency. Moreover, fine-tuning open-source protects privacy enhances log Therefore, we introduce propose method instruction tuning using LLaMA2 parsing, anomaly detection classification, analysis, report generation. Real extracted from field-deployed was used design construct datasets. We utilized dataset demonstrated evaluated effectiveness proposed scheme. results indicate that this scheme improves performance tasks, especially 14% improvement exact match rate 13% F1-score 23% usability compared best baselines.

Язык: Английский

When Large Language Models Meet Optical Networks: Paving the Way for Automation DOI Open Access
Danshi Wang,

Yidi Wang,

Xiaotian Jiang

и другие.

Electronics, Год журнала: 2024, Номер 13(13), С. 2529 - 2529

Опубликована: Июнь 27, 2024

Since the advent of GPT, large language models (LLMs) have brought about revolutionary advancements in all walks life. As a superior natural processing (NLP) technology, LLMs consistently achieved state-of-the-art performance numerous areas. However, are considered to be general-purpose for NLP tasks, which may encounter challenges when applied complex tasks specialized fields such as optical networks. In this study, we propose framework LLM-empowered networks, facilitating intelligent control physical layer and efficient interaction with application through an LLM-driven agent (AI-Agent) deployed layer. The AI-Agent can leverage external tools extract domain knowledge from comprehensive resource library specifically established This is user input well-crafted prompts, enabling generation instructions result representations autonomous operation maintenance To improve LLM’s capability professional stimulate its potential on details performing prompt engineering, establishing library, implementing illustrated study. Moreover, proposed verified two typical tasks: network alarm analysis optimization. good response accuracies semantic similarities 2400 test situations exhibit great LLM

Язык: Английский

Процитировано

4
Large Language Model-Based Optical Network Log Analysis Using LLaMA2 with Instruction Tuning DOI
Yue Pang, Min Zhang, Yanli Liu

и другие.

Journal of Optical Communications and Networking, Год журнала: 2024, Номер 16(11), С. 1116 - 1116

Опубликована: Окт. 2, 2024

The optical network encompasses numerous devices and links, generating a significant volume of logs. Analyzing these logs is for optimization, failure diagnosis, health monitoring. However, the large-scale diverse formats present several challenges, including high cost difficulty manual processing, insufficient semantic understanding in existing analysis methods, strict requirements data security privacy. Generative artificial intelligence (GAI) with powerful language generation capabilities has potential to address challenges. Large models (LLMs) as concrete realization GAI are well-suited analyzing DCI logs, replacing human experts enhancing accuracy. Additionally, LLMs enable intelligent interactions administrators, automating tasks improving operational efficiency. Moreover, fine-tuning open-source protects privacy enhances log Therefore, we introduce propose method instruction tuning using LLaMA2 parsing, anomaly detection classification, analysis, report generation. Real extracted from field-deployed was used design construct datasets. We utilized dataset demonstrated evaluated effectiveness proposed scheme. results indicate that this scheme improves performance tasks, especially 14% improvement exact match rate 13% F1-score 23% usability compared best baselines.

Язык: Английский

Процитировано

1