dmlc/mxnetArchived

Features

• Deep Learning - Serves as a scalable deep learning framework for training and deploying neural networks using hybrid programming.
• Distributed Training - Distributes deep learning tasks across multiple GPUs and machines using automatic parallelism for large datasets.
• Distributed Training - Scales the training of large neural networks across multiple GPUs and machines to handle massive datasets.
• Hardware Acceleration Abstractions - Provides a unified hardware acceleration abstraction layer that allows custom accelerators to be plugged into the core engine.
• Hybrid Programming Paradigms - Balances rapid development speed and high execution efficiency by combining symbolic and imperative programming styles.
• Machine Learning Engines - Functions as a distributed machine learning engine that scales training workloads via automatic parallelism.
• Hardware-Agnostic Inference Layers - Implements a hardware-agnostic backend that decouples model execution logic from specific hardware accelerators.
• Neural Network Deployment - Enables trained models to be executed across both mobile and desktop platforms using flexible language bindings.
• Neural Network Trainers - Provides an optimized training environment using a hybrid of symbolic and imperative programming to execute neural network training loops.
• Distributed Parallelism - Implements automatic distributed parallelism to scale model training and inference across multiple GPUs and machines.
• Hybrid Execution Modes - Combines a static computation graph for performance optimization with an imperative interface for flexible model development.
• Dataflow Dependency Scheduling - Schedules operations by tracking dependencies in a dynamic graph to execute nodes immediately as data dependencies are met.
• Deep Learning Frameworks - Provides a scalable environment to build and optimize deep learning models through distributed training workloads.
• Hardware Acceleration Backends - Offers an extensible interface for integrating custom accelerator libraries and proprietary hardware without modifying core code.
• Mobile - Optimizes and deploys deep learning models for ARM-based portable devices using hardware accelerators.
• Multi-Language Model Development - Enables the building and execution of neural networks using various languages including Python, R, Scala, and Go.
• Mobile ARM Cross-Compilation - Provides the ability to cross-compile native code for ARM architectures to enable neural network execution on mobile devices.
• Cross-Platform Runtimes - Provides a cross-platform runtime for deploying trained models to mobile and desktop devices across various languages.
• Mobile Model Deployment - Supports the deployment of neural networks on portable smart devices using ARM-native cross-compilation.
• Hardware Acceleration - Provides an extensible interface to integrate proprietary hardware accelerators and specialized libraries.
• Cross-Language Bindings Layers - Uses a C++ core with wrapper layers to expose deep learning functionality across multiple high-level programming languages.
• Machine Learning - Flexible deep learning and GPU-accelerated computing.

Star history