MILES-CREDIT Documentation#
Welcome to the documentation for MILES-CREDIT, the NSF NCAR Community Research Earth Digital Intelligent Twin project. CREDIT is a machine learning-based research platform for understanding the best practices for training and operating global and regional AI autoregressive models, built as part of the NSF NCAR Machine Integration and Learning for Earth Systems (MILES) group.
CREDIT enables users to train, run, and evaluate AI-based numerical weather and climate models. This documentation will guide you through installation, configuration, training, inference, evaluation, and extending the system with custom datasets and models.
New here? Start with the Quickstart — it gets you from zero to a running training job in under 10 minutes.
What you’ll find here:
How to install CREDIT from source
How to set up and train a model
How to run inference and evaluate results
How to contribute datasets, models, and enhancements
Config file reference for reproducible HPC runs
Tutorial videos for visual guidance
If you encounter issues or have suggestions, please open an issue on our GitHub repository. Contributions are welcome!
Getting Started
Configuration File
- What’s in the Configuration File?
- CREDIT Configuration Guide
- Overview
- General Setup
- Data Configuration
- Physics and Normalization Files
- Data Preprocessing and Temporal Configuration
- Training Configuration
- CREDIT supports single-GPU, multi-GPU, and distributed training.
- Model Configuration
- Post-Block (
post_conf) - Stochastic Kinetic Energy Backscatter (SKEBS)
- Conservation Schemes
- Tracer Fixer: Ensuring Non-Negative Water Content
- Global Mass Fixer
- Global Water Fixer
- Global Energy Fixer
- Best Practices
- Loss Configuration
- Prediction (Inference) Configuration
- PBS Job Submission (HPC)
- Troubleshooting
- Best Practices
- Additional Resources
Training and Inference
Contributing
Adding New Models and Datasets