README.md

LILAC: Log Parsing using LLMs with Adaptive Parsing Cache

This is the replication package for [FSE24] "LILAC: Log Parsing using LLMs with Adaptive Parsing Cache".

In this paper, we propose LILAC, the first practical LLM-based log parsing framework. LILAC consists of two main components: the ICL-enhanced Parser and the adaptive parsing cache.

The detailed parsing time can be found in detailed parsing time for RQ4.

If you find our paper benefit your research, please kindly cite our related paper:

• Zhihan Jiang, Jinyang Liu, Zhuangbin Chen, Yichen Li, Junjie Huang, Yintong Huo, Pinjia He, Jiazhen Gu, Michael R. Lyu. LILAC: Log Parsing using LLMs with Adaptive Parsing Cache FSE, 2024.

Repository Organization

├── full_dataset/ # Please download and unzip full datasets into this directory
│   └── sampled_examples # Our saved sampled candidates
├── benchmark/
│   ├── evaluation/ # the evaluation code of LILAC
│   └── logparser/ # the implementation code of LILAC
├── result/
│   └── ...... # contains the saved evaluation files
├── sampling/ # the implementation of candidate sampling algorithms
│   ├── logppt_sampling.py # the sampling algorithm of LogPPT
│   └── LILAC_sampling.py # the sampling algorithm of LILAC
├── requirements.txt
├── openai_key.txt # the OpenAI api address and key
└── README.md

Quick Start

Datasets

Please first download the large-scale datasets for log parsing in LogPub from Zenodo and unzip these datasets into the directory of full_dataset.

Installation

1. Install python >= 3.8
2. pip install -r requirements.txt

Execution

• Candidate Sampling (optional)

  We have provided the saved sampled candidate logs for reproducing.

  One can also delete the full_dataset/sampled_examples and execute the LILAC's sampling algorithm as follows:

  cd sampling/
  python LILAC_sampling.py

• Online Log Parsing

  Please first add an OpenAI API key (sk-xxxx) into the second line of openai_key.txt.

  We provide a one-click script to run LILAC for online log parsing.

  ./online_parsing.sh

  One can also go to benchmark/evaluation and execute:

  python LILAC_eval.py --shot [candidates] --example_size [demonstrations] --model [model]

The parsed results and evaluation results will be saved in the result/ directory.

We have provided the saved evaluation metric files of LILAC with different settings in the directory of result/.