iterative/cml
Cml
CML is a pipeline automation tool for training and evaluating machine learning models, functioning as a CI/CD system for machine learning. It serves as a cloud compute orchestrator and Git-based workflow manager that automates model training cycles through branch management, automated commits, and integrated reporting.
The project distinguishes itself by provisioning ephemeral cloud instances or Kubernetes nodes to provide specialized hardware for compute-heavy tasks. It also manages remote compute runners, allowing the connection of self-hosted GPU clusters or on-premise machines to execute containerized machine learning workflows.
The system covers a broad range of capabilities including ML experiment tracking, where performance metrics and visualizations are posted directly into version control pull requests. It handles ML pipeline automation from the initial data import and versioning to the generation of formatted workflow reports and external visualization links.
The tool provides additional utility for infrastructure management through SSH-based remote debugging and the ability to resume interrupted jobs.
