Introduction

Welcome to the Anemoi Training user guide. Anemoi Training is a framework designed for training machine learning models for weather forecasting. This guide will walk you through the key components and processes involved in using Anemoi Training effectively.

Anemoi Training is built to be flexible and customizable, allowing you to adjust various aspects of the model and training process without modifying the underlying code. The framework supports different model architectures, including Graph Neural Networks (GNNs), Graph Transformers, and Transformers with Flash Attention.

User Journey

As you progress through this guide, you’ll learn how to:

1. Configure Your Training: Understand how to set up and customize your training pipeline using YAML-based configuration files.

2. Configure Your Data Handling: Learn about data routing, normalization strategies, and how to set up your dataset for training.

3. Choose and Configure Models: Explore the different model architectures available and how to select the best one for your needs.

4. Execute and Monitor Experiments: Get hands-on with running your training jobs and tracking your experiments using tools like MLflow.

5. Optimize Performance: Discover techniques for fine-tuning your model, including learning rate scheduling and loss function scaling.

6. Debug and Troubleshoot: Learn strategies for identifying and resolving issues that may arise during the training process.

7. Scale Your Training: Understand how to leverage distributed training to accelerate your model development.

Whether you’re new to machine learning for weather forecasting or an experienced practitioner, this guide will provide you with the knowledge and tools to make the most of Anemoi Training.

Let’s begin by exploring how to configure your training pipeline in the next section.