Repository of SCELM, SoftwareChange Project from AIOps Lab, Nankai University.
Maintained by Weihua Kuang, Guided by Yongqian Sun and Shenglin Zhang.
feel free to email [email protected] if you have any function problems.
⭐️ 2025/03/25: Our Paper A Multimodal Intelligent Change Assessment Framework for Microservice Systems Based on Large Language Models (SCELM) accepted by FSE 2025 Industry (45/165)!
2025/03/10: Test & Deploy SCELM on ByteDance enterprise data.
2025/01/23: Paper submission to FSE 2025.
2025/01/10: Refactor code.
... : more detailed depolyment tutorials comming soon!
Using Large Language Model to improve efficiency of software change maintainance.
Our method achieved SOTA performance on the business scenario:
- Yun Zhanghu Technology
- preliminary tests on ByteDance
- Ollama, deploy models on local server
- version 0.5.1
- recommended model: gemma2-9b, qwen2.5-7b,llama3.1-7b
- Redis, vector database
- version 7.2.5
download the model files from the web and place them under the /model folder
- m3e-base, vector embeeding model.
- all-MiniLM-L6-v2, semantic similarity matching model.
ollama serve
(optional) Test ollama model chat
models supported according to: Ollama Models
ollama pull <model_name>
<input chat content>
redis-server
more detailed command according to Redis tutorials,
(recommended) Use Redis Insight to visually manage data, mind your connection port.
-
Part1_AreaText-Generate
Generate areatext according to change information.
-
Part2_LLM-Generate
Change analysis by LLM using Retriever-Augmented Generation.
according to codes in /utils
Any function problems connect Weihua Kuang by [email protected]
Any baseline experiments connect Chao Shen by [email protected]
- RAG functional code, experiment design & Paper script by Weihua Kuang,
- Baseline experiment by Chao Shen,
- Co-guided & Mainly paper writing by other writers.
- Tinghua Zheng, his intern experience in Yun Zhanghu Technology leads to this project.
- Guangba Yu helps for one baseline experiment about his work ChangeRCA.
- Hailin Zhang helps run our SCELM on preliminary data of Bytedance.