Introduction
LMMs Engine is a flexible and extensible framework designed for training large multimodal models. It provides:
Modular Architecture: Built on factory patterns, making it easy to extend with custom components
Flexible Training: Support for various model architectures, datasets, and training strategies
Configuration-Driven: Lightweight framework with configuration-based component building
Video Support: Native support for video processing with multiple backend options
Scalable: Designed to handle large-scale training with performance optimization
Getting Started
To begin using LMMs Engine, proceed to the Quick Start guide or explore the User Guide for detailed information about specific features.
Key Concepts
Factory Pattern: Used to encapsulate component creation for models, trainers, datasets, and processors
Builder Pattern: Lazy initialization of components, loading them only when needed
MVC Architecture: Clean separation of concerns for future extensibility
For more details about the architecture, see the Design Principle.