Repolode
Submit
SwanHubX

SwanHubX/SwanLab

Open-source AI training tracking and visualization platform for logging metrics, hyperparameters and hardware stats, with cloud or self-hosted deployment.

Open-source alternative to:Weights & Biases

View on GitHubWebsite
Stars
4,313
Forks
231
Open issues
47
Last commit
yesterday

What is it?

SwanLab is a platform for tracking and visualizing machine learning training runs. Developers add a few lines of Python to their training code, and SwanLab records scalar metrics, hyperparameters, logs, and system information such as CPU, memory, and accelerator usage. Results appear in a web dashboard where runs can be charted, compared in tables, and shared through persistent links.

It integrates with a wide range of frameworks, including PyTorch, Hugging Face Transformers, PyTorch Lightning, LLaMA Factory, ms-swift, Ultralytics, MMEngine, and Keras. Hardware monitoring covers NVIDIA, AMD, Ascend NPU, Cambricon MLU, Kunlunxin XPU, and several other domestic accelerators. Teams can collaborate within a project and organize experiments into groups.

SwanLab can be used through the hosted cloud service at swanlab.cn or self-hosted with Docker or Kubernetes, and the offline dashboard can be run locally. It is released under the Apache 2.0 license.

Key features

  • 01Logs scalar metrics, hyperparameters, and text from training code with a few lines of Python.
  • 02Charts include line, bar, scatter, heatmap, and custom ECharts, plus media such as images, audio, and video.
  • 03Records hardware and system stats across NVIDIA, AMD, Ascend, and other accelerators.
  • 04Integrates with 30+ training frameworks and libraries.
  • 05Supports team projects, experiment comparison tables, and shareable links.
  • 06Can be self-hosted with Docker or Kubernetes, or run as an offline dashboard.

Use cases

  • Tracking loss and accuracy curves while fine-tuning a large language model.
  • Comparing hyperparameter runs across a team to pick the best configuration.
  • Monitoring GPU or NPU utilization during long training jobs on a cluster.
  • Hosting an internal experiment dashboard on a private network for an air-gapped team.

Topics

#machine-learning#deep-learning#experiment-tracking#visualization#pytorch#llm#mlops#ai-infra#data-science#logging

FAQ

What is SwanLab?
Open-source AI training tracking and visualization platform for logging metrics, hyperparameters and hardware stats, with cloud or self-hosted deployment. SwanLab is a platform for tracking and visualizing machine learning training runs. Developers add a few lines of Python to their training code, and SwanLab records scalar metrics, hyperparameters, logs, and system information such as CPU, memory, and accelerator usage. Results appear in a web dashboard where runs can be charted, compared in tables, and shared through persistent links.
Is SwanLab free to use? What license does it have?
SwanLab is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is SwanLab still maintained?
Actively maintained. The last commit to SwanLab was on Oct 20, 2026.
What can I use SwanLab for?
Tracking loss and accuracy curves while fine-tuning a large language model. Comparing hyperparameter runs across a team to pick the best configuration. Monitoring GPU or NPU utilization during long training jobs on a cluster. Hosting an internal experiment dashboard on a private network for an air-gapped team.
Where can I get SwanLab?
The source code and installation instructions are on GitHub: https://github.com/SwanHubX/SwanLab

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects