Microsoft/nniArchived
Nni
NNI is an AutoML toolkit designed to automate machine learning lifecycles. It functions as a hyperparameter optimization framework, a neural architecture search tool, and a model compression suite.
The project provides a distributed training orchestrator to manage machine learning workloads across local machines, remote servers, and cloud platforms. It enables the discovery of efficient model structures through reinforcement learning and one-shot optimization methods, while utilizing Bayesian and evolutionary algorithms to automate hyperparameter tuning.
Additional capabilities include tools for reducing model size and complexity via pruning and quantization to increase inference speed. The system also handles the coordination of training across multiple compute resources to scale experiments.
Features
- Automated Machine Learning Tools - Provides an automated machine learning toolkit to streamline model selection, tuning, and pipeline construction.
- Distributed Training Managers - Configures and scales machine learning training jobs across multiple compute nodes.
- Distributed Training Orchestration - Manages parallelization and synchronization of training workloads across computing clusters.
- Hyperparameter Tuning - Automates the search for optimal model settings using Bayesian optimization and evolutionary algorithms.
- Model Compression Suites - Ships a comprehensive toolkit for reducing model size and complexity through techniques such as pruning and quantization.
- Hyperparameter Optimization - Provides automated methods for searching and selecting the best configuration parameters for machine learning models using Bayesian and evolutionary algorithms.
- Neural Architecture Search - Implements automated methods for discovering efficient neural network architectures using reinforcement learning and one-shot optimization.
- Distributed Task Orchestration - Manages the scheduling and execution of machine learning trials across a cluster of compute resources.
- Distributed Training Coordination - Coordinates and synchronizes machine learning training tasks across distributed clusters and cloud platforms.
- Model Pruning - Identifies and removes redundant neural network weights to reduce model size and computational requirements.
- One-Shot Methods - Implements one-shot architecture search to evaluate potential sub-architectures without retraining every candidate.
- Docker Container Deployments - Packages the execution environment into Docker images to ensure consistent behavior across infrastructures.
- Containerized Training Environments - Provides pre-configured container images to ensure consistent dependencies and drivers for model training.
- Optimization Trials - Implements architectures that distribute hyperparameter optimization trials across multiple processes or machines.
- Automated Machine Learning - Toolkit for neural architecture search and hyperparameter tuning.
- AutoML - Toolkit for automating the machine learning lifecycle.
- Machine Learning Libraries - AutoML toolkit for model lifecycle management.
