apache/incubator-mxnetArchived

Features

• Deep Learning Frameworks - Provides a high-performance deep learning framework with a multi-language API for building and training neural networks.
• Machine Learning Training - Provides a high-performance framework and programming model for training and optimizing complex neural networks.
• Distributed Deep Learning - Scales the training of deep learning models across multiple compute nodes and GPUs to process massive datasets.
• Distributed Deep Learning Frameworks - Functions as a unified platform for the distributed training and deployment of deep learning models.
• Large-Scale Training Frameworks - Ships infrastructure and orchestration tools for scaling neural network training across massive multi-node compute clusters.
• Model Parallelism - Implements strategies for splitting model parameters across multiple GPUs to enable the training of extremely large models.
• Neural Network Trainers - Implements training loops and a dynamic scheduler to efficiently update model parameters via forward and backward propagation.
• Neural Network Execution Engines - Implements a mutation-aware dependency scheduler that optimizes neural network execution through a dynamic dataflow engine.
• Hybrid Execution Modes - Combines a static computation graph with immediate imperative execution to balance developer flexibility and runtime performance.
• Task Scheduling - Utilizes a mutation-aware dependency scheduler to calculate the optimal execution order of tasks in a dataflow graph.
• Hardware Acceleration Backends - Offers interfaces and backends that allow the integration of custom hardware accelerators to improve model execution performance.
• High-Performance Training Schedulers - Balances developer productivity with efficient hardware execution through a dynamic dependency scheduler.
• Edge AI Model Deployment - Optimizes and deploys machine learning models to run efficiently on local hardware, ARM, and other edge devices.
• Distributed Deployment Utilities - Provides utilities for scaling model inference across multiple hardware devices and nodes using parameter sharding.
• Multi-Language ML APIs - Enables the development of machine learning models using Python, R, Julia, Scala, Go, and JavaScript.
• Backend-Agnostic Engines - Provides a computational engine that decouples neural network operations from specific hardware backends for cross-platform execution.
• Model Bytecode Compilation - Translates high-level model definitions into optimized low-level instructions for deployment on mobile and ARM devices.
• Differential Dataflow Engines - Tracks changes to data buffers within the execution graph to incrementally propagate updates and optimize memory efficiency.
• Cross-Platform Runtimes - Provides a software layer that ensures consistent model execution across diverse hardware architectures including ARM.
• Artificial Intelligence - Distributed deep learning framework with multi-language support.
• Deep Learning Frameworks - Flexible framework supporting both symbolic and imperative programming.
• General Machine Learning - Flexible, distributed deep learning framework for multiple languages.
• Machine Learning and AI - Distributed deep learning framework with multi-language support.
• 机器学习框架 - Portable, distributed deep learning framework supporting multiple programming languages.
• Model Conversion Tools - Scalable deep learning framework with support for ONNX export.
• Segmentation Architectures - Deep learning framework with segmentation examples.
• Machine Learning Frameworks - Flexible, portable deep learning framework for distributed environments.
• Deep Learning Implementations - Deep learning framework for efficient model training and deployment.

Star 历史