- Definition
- Initial AI/ML workflow implementation for O-RAN environment. Need to interact with another project to accomplish a whole life cycle management of the AI model.
- Project Scope
- AI 모델 라이프 사이클 관리 ( 모델 학습 파이프라인 관리 / 학습완료된 모델 버전 관리 및 배포시스템 연동 / AI 서비스 (Application) 관리 ) , Dashboard
- AI 모델 학습 환경 ( 데이터 추출 / Feature 관리 및 연동 / AI 모델 저장 기능 / 모델 학습 플랫폼 지원 or 모델 학습/학습 파이프라인 구동 환경 )
- AI 모델 추론 환경 ( 모델 서빙 플랫폼 지원 or 모델 추론 서비스 구동 환경 )
- g rel scope
AIML Framework (AIMLFW)
Mission: Stand-alone installation (separated from existing platform deployment) and initial AIML workflow modules
Original primary goals:
• Stand-alone installation with Kubeflow as a Training host backend and Kserve as a Inference host backend
• Manual Deployment of ML rApp and ML xApp
• Training Job Management: Create/Edit/Delete usecases and Training pipelines and monitoring current training jobs
• Data Extraction for model training from data lake
• Model feature database for Training pipeline
• Trained model storage
• Sample ML pipeline and ML xApp : QoE Prediction model using LSTM with data from ricapp/qp
G release source code, container images and deployment instructions
TODO
Repo. hierarchy
Repository (Hierarchy) | Description | |||
...
awmf
...
AI Workflow Management Function
...
aiml-fw |
...
O-RAN AIML Framework Project
...
awmf |
tm | Training Manager : pipeline and model management | |
...
athp |
...
...
dataextraction | Data broker, Interaction with Data lake (SMO) | ||
...
|
...
SDK for Training Host
...
featurestore | Feature store |
...
modelstore | Model Storage |
...
|
...
Training Platform Service (ex. Kubeflow)
...
kubeflowadatper | Adapter for Kubeflow |
...
|
...
AI Inference Host Platform
...
ips |
...
Inference Platform Service (ex. Kserve)
...
kserveadapter | Adapter for Kserve |
...
dep | Deployment scripts aiml workflow → it/dep linking | ||
...
portal |
...
GUI Modules in O-RAN project
...
aiml-dashboard | GUI for AIML Workflow | |||
...
|
...
...
qp-aimlfw | Sample pipeline and ML xApp for QoE prediction | ||
...
