SwanLab

SwanLab is an experiment tracking and visualization platform for AI model training, available as a cloud service and as a self hosted deployment. Researchers add a few lines of Python to log metrics, hyperparameters and media, and SwanLab records them in a dashboard during training and as a record for later review. It integrates with more than 30 mainstream training frameworks, including PyTorch, Transformers, PyTorch Lightning, Ultralytics, Keras and MMEngine. It automatically captures terminal logs, hardware details such as GPU model and memory, and Python and Git information, and monitors CPU and GPU usage. Experiments can be compared in tables and charts, viewed from a phone, used offline, and shared with collaborators through links. Organizations can group members so several people run experiments under one project. It is aimed at AI researchers and training teams who want to track runs and keep a history of what worked. Pricing is published for three tiers. The Community Edition is free at 0 yuan per month for one user seat and is intended for personal, non commercial use, covering training tracking and visualization. The Enterprise Edition is custom priced for teams and adds flexible deployment, Kubernetes support, team permission controls, SSO and dedicated instant messaging support. Dedicated private deployment, built with cloud providers and managed by the SwanLab team, is also custom priced. Self hosting requires Docker and Python and runs on Linux, macOS or Windows.