MLOPS/SERVING
Model Serving Tools Comparison & Architecture
개발허재
2023. 1. 3. 11:07
mlops 대표 오픈 소프트웨어
https://github.com/kserve/kserve , https://github.com/SeldonIO/seldon-core , https://github.com/bentoml/BentoML , https://github.com/mlflow/mlflow, https://github.com/apache/airflow
Kubernetes를 기반으로 구축된 두 가지 인기 있는 모델 서빙 프로젝트
kserver, seldon core
why mlops model serving on k8s?
handle high loads(높은 처리량) , application 오토스케일링(kubectl autoscale deployment mlflow-serving --cpu-percent=80 --min=1 --max=12) , serving them as an API endpoint(docker image containerizing & with proxy server)
Multi-framework serving with KServe or Seldon Core
KServe VS Seldon Core VS BentoML
kubelflow VS mlflow
https://kgw7401.tistory.com/48
*mlflow도 k8s 위에서 띄울수 있다. 하지만. https://github.com/pdemeulenaer/mlflow-on-kubernetes/blob/main/k8s/mlflow_deployment.yaml 와 같이 official이 아님.
kserve 각 컴포넌트를 하나하나 잘 설명해놓은 글 : https://devocean.sk.com/blog/techBoardDetail.do?ID=164093
kserve architecture
