SwanHubX/SwanLab
⚡️SwanLab - an open-source, modern-design AI training tracking and visualization tool. Supports Cloud / Self-hosted use. Integrated with PyTorch / Transformers / verl / LLaMA Factory / ms-swift / Ultralytics / MMEngine / Keras etc.
What it solves
SwanLab is an AI training analysis and metric observation platform designed for model training teams. It solves the problem of tracking complex training processes by providing tools for visualization, automatic logging, hyperparameter recording, and experiment comparison, which helps researchers identify training issues and accelerate model iteration.
How it works
Users integrate a lightweight SDK into their machine learning pipelines to log scalars, images, audio, text, video, and 3D point clouds. The platform then visualizes this data through a centralized dashboard. It supports both a cloud-based version for remote monitoring and a self-hosted community version for offline environments. It also integrates with over 30 mainstream frameworks like PyTorch, HuggingFace Transformers, and LLaMA Factory.
Who it’s for
It is built for AI researchers and model training teams who need to monitor training progress, compare multiple experiments to find optimal hyperparameters, and collaborate with team members in real-time.
Highlights
- Comprehensive Metric Tracking: Supports a wide variety of data types including scalars, 3D point clouds, biochemical molecules, and custom ECharts.
- Extensive Framework Integration: Seamlessly integrates with 30+ frameworks including PyTorch Lightning, Ultralytics, and XGBoost.
- Broad Hardware Monitoring: Real-time system-level monitoring for a vast array of hardware, including NVIDIA GPUs, Ascend NPUs, AMD ROCm, and various other specialized AI accelerators.
- Collaborative Tools: Features project sharing, team collaboration, and plugin extensions for notifications via Slack, Discord, and Lark.
- Flexible Deployment: Offers both a managed cloud service and self-hosted options via Docker and Kubernetes.