hiyouga/LlamaFactory

Features

• Large Language Model Fine-Tuning Frameworks - Consolidates diverse optimization techniques into a single environment for end-to-end model adaptation and training.
• Model Inference Servers - Exposes trained models via standardized network protocols to facilitate scalable and reliable prediction services.
• Language Model Fine-Tuning - Simplifies complex model refinement by offering a unified interface for both full-parameter and efficient training methods.
• Experiment Tracking - Captures real-time performance metrics and training progress to assist in evaluating model quality during development.
• Model Fine-Tuning and Adaptation - Enables the customization of pre-trained models to specific domains or tasks through a streamlined training pipeline.
• Multi-Backend Model Construction - Standardizes data loading and optimization logic across various hardware backends and model architectures.
• Low-Code Machine Learning Tools - Offers a visual interface that allows users to manage training workflows without writing extensive custom code.
• Parameter Efficient Fine-Tuning - Reduces computational overhead by updating only a subset of model parameters during the adaptation process.
• Integrated Development Platforms - Bundles configuration, training, and monitoring tools into a single environment for the entire model lifecycle.
• Machine Learning Training - Utilizes structured configuration files to decouple training parameters from execution logic, ensuring reproducible experiments.
• Model Inference APIs - Deploys refined models as local API endpoints for immediate integration into external software applications.
• Local Model Inference Servers - Hosts models locally to serve low-latency predictions through standard network APIs.
• Experiment Tracking Systems - Streams training loop metrics to external systems for real-time visualization and comparative analysis.

Star history