MLflow: metadata and artifact paths

MLflow: metadata and artifact paths An architecture diagram generated by Archify. MLflow SDK client · training / evaluation · Architecture component MLflow SDK client training / evaluation Tracking server · metadata + artifact API · Architecture component Tracking server metadata + artifact API SQL backend · experiments / runs / models · Architecture component SQL backend experiments / runs / models Artifact store · files / model weights · Architecture component Artifact store files / model weights Browser UI · HTTP API requests · Architecture component Browser UI HTTP API requests metadata / proxy SQL proxy files HTTP direct mode: client credentials Legend Frontend Backend Stores External

Metadata

  • • New default: SQLite; choose an explicit production DB
  • • Local SDK can use SQL without a separate HTTP server

Artifact permissions

  • • Proxy: server permissions / direct: client permissions
  • • Changing server flags does not rewrite existing artifact URIs