SwanLab
An open-source, modern AI training tracking and visualization platform that integrates with more than 50 frameworks such as PyTorch, Transformers and Ultralytics, usable as a cloud service or self-hosted, including on Kubernetes.
SwanLab is an AI training analysis platform aimed at model-training teams, integrating with more than 50 mainstream frameworks including PyTorch, TensorFlow, Transformers, verl, LLaMA Factory, ms-swift, Ultralytics and MMEngine so it combines easily with existing experiment code. It logs and visualizes training metrics, with line charts, experiment grouping, pinned experiments and projects, baseline comparison, hardware monitoring for GPUs including AMD ROCm, and HTML charts that can embed generated content.
It can be used as a hosted online service or self-hosted, including a Kubernetes deployment with Prometheus and Grafana, and provides an API for programmatic access to experiment data alongside a command-line interface and an agent skill aimed at automated research workflows. A rewritten SDK improves metric-logging performance for large-scale training runs.