Comparison of deep learning software

|Jason Dai (Intel)

|2016

|{{free|Apache 2.0}}

|{{Yes}}

|Apache Spark

|Scala

|Scala, Python

|

|

|{{No}}

| {{No}}

|

|{{Yes}}

|{{Yes}}

|{{Yes}}

|

|

|{{Yes}}

| Berkeley Vision and Learning Center

| 2013

| {{BSD-lic}}

| {{Yes}}

| Linux, macOS, Windows{{cite web|url=https://github.com/Microsoft/caffe|title=Microsoft/caffe|work=GitHub|date=30 October 2021}}

| C++

| Python, MATLAB, C++

| {{Yes}}

| {{Depends|Under development{{Cite web|url=https://github.com/BVLC/caffe|title=Caffe: a fast open framework for deep learning.|date=July 19, 2019|via=GitHub}}}}

| {{Yes}}

| {{No}}

| {{Yes}}

| {{Yes}}{{Cite web|url=http://caffe.berkeleyvision.org/model_zoo.html|title=Caffe | Model Zoo|website=caffe.berkeleyvision.org}}

| {{Yes}}

| {{Yes}}

| {{No}}

| {{Dunno}}

|{{No}}{{Citation|title=GitHub - BVLC/caffe: Caffe: a fast open framework for deep learning.|date=2019-09-25|url=https://github.com/BVLC/caffe|publisher=Berkeley Vision and Learning Center|access-date=2019-09-25}}

| Preferred Networks

| 2015

| {{BSD-lic}}

| {{Yes}}

| Linux, macOS

| Python

| Python

| {{No}}

| {{No}}

| {{Yes}}

| {{No}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{No}}

| {{Yes}}

|{{No}}{{Citation|url=https://preferred.jp/en/news/pr20191205/|title=Preferred Networks Migrates its Deep Learning Research Platform to PyTorch|date=2019-12-05|access-date=2019-12-27}}

| Skymind engineering team; Deeplearning4j community; originally Adam Gibson

| 2014

| {{Free|Apache 2.0}}

| {{Yes}}

| Linux, macOS, Windows, Android (Cross-platform)

| C++, Java

| Java, Scala, Clojure, Python (Keras), Kotlin

| {{Yes}}

| {{No}}{{cite web|url=https://github.com/deeplearning4j/nd4j/issues/27|title=Support for Open CL · Issue #27 · deeplearning4j/nd4j|work=GitHub}}

| {{Yes}}{{cite web|url=http://nd4j.org/gpu_native_backends.html|title=N-Dimensional Scientific Computing for Java|access-date=2016-02-05|archive-date=2016-10-16|archive-url=https://web.archive.org/web/20161016094035/http://nd4j.org/gpu_native_backends.html|url-status=dead}}{{cite web|url=https://deeplearning4j.org/compare-dl4j-tensorflow-pytorch|title=Comparing Top Deep Learning Frameworks|publisher=Deeplearning4j|access-date=2017-10-31|archive-url=https://web.archive.org/web/20171107011631/https://deeplearning4j.org/compare-dl4j-tensorflow-pytorch|archive-date=2017-11-07|url-status=dead}}

| {{No}}

| {{Yes|Computational Graph}}

| {{Yes}}{{cite web|url=http://deeplearning4j.org/model-zoo|title=Deeplearning4j Models|author1=Chris Nicholson|author2=Adam Gibson|access-date=2016-03-02|archive-url=https://web.archive.org/web/20170211020819/https://deeplearning4j.org/model-zoo|archive-date=2017-02-11|url-status=dead}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{Yes}}{{cite web|url=http://deeplearning4j.org/spark|title=Deeplearning4j on Spark|author=Deeplearning4j|publisher=Deeplearning4j|access-date=2016-09-01|archive-url=https://web.archive.org/web/20170713012632/https://deeplearning4j.org/spark|archive-date=2017-07-13|url-status=dead}}

| {{Yes}}

|Davis King

|2002

| {{Free|Boost Software License}}

| {{Yes}}

|Cross-platform

|C++

|C++, Python

| {{Yes}}

| {{No}}

| {{Yes}}

| {{No}}

| {{Yes}}

| {{Yes}}

| {{No}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

|Mike Innes

|2017

| {{Free|MIT license}}

| {{Yes}}

|Linux, MacOS, Windows (Cross-platform)

|Julia

|Julia

|

| {{No}}

| {{Yes}}

| {{Yes}}{{cite web|url=http://github.com/FluxML/Metalhead.jl|title=Metalhead|date=29 October 2021|publisher=FluxML}}

| {{Yes}}

| {{Yes}}

| {{No}}

| {{Yes}}

| {{Yes}}

| Intel

| 2015

| {{free|Apache 2.0}}

| {{Yes}}

| Linux, macOS, Windows on Intel CPU{{Cite web|url=https://software.intel.com/en-us/intel-daal|title=Intel® Data Analytics Acceleration Library (Intel® DAAL)|date=November 20, 2018|website=software.intel.com}}

| C++, Python, Java

| C++, Python, Java

| {{Yes}}

| {{No}}

| {{No}}

| {{No}}

| {{Yes}}

| {{No}}

|

| {{Yes}}

|

| {{Yes}}

| {{Yes}}

| Intel

| 2017

| {{Proprietary}}

| {{No}}

| Linux, macOS, Windows on Intel CPU{{Cite web|url=https://software.intel.com/en-us/mkl|title=Intel® Math Kernel Library (Intel® MKL)|date=September 11, 2018|website=software.intel.com}}

| C/C++, DPC++, Fortran

| C{{Cite web|url=https://software.intel.com/en-us/mkl-developer-reference-c-deep-neural-network-functions|title=Deep Neural Network Functions|date=May 24, 2019|website=software.intel.com}}

| {{Yes}}{{Cite web|url=https://software.intel.com/en-us/articles/intel-math-kernel-library-intel-mkl-using-intel-mkl-with-threaded-applications|title=Using Intel® MKL with Threaded Applications|date=June 1, 2017|website=software.intel.com}}

| {{No}}

| {{No}}

| {{No}}

| {{Yes}}

| {{No}}

| {{Yes}}{{Cite web|url=https://software.intel.com/en-us/articles/intel-xeon-phi-delivers-competitive-performance-for-deep-learning-and-getting-better-fast|title=Intel® Xeon Phi™ Delivers Competitive Performance For Deep Learning—And Getting Better Fast|date=March 21, 2019|website=software.intel.com}}

| {{Yes}}

|

| {{No}}

| {{Yes}}

| Google

| 2018

| {{free|Apache 2.0}}

| {{Yes}}

| Linux, macOS, Windows

| Python

| Python

|

|

| {{Depends|Only on Linux}}

| {{No}}

| {{Yes}}

| {{No}}

|

|

|

| {{Yes}}

| {{Yes}}

| François Chollet

| 2015

| {{Free|MIT license}}

| {{Yes}}

| Linux, macOS, Windows

| Python

| Python, R

| {{Depends|Only if using Theano as backend}}

| {{Depends|Can use Theano, Tensorflow or PlaidML as backends}}

| {{Yes}}

| {{No}}

| {{Yes}}

| {{Yes}}{{Cite web|url=https://keras.io/applications/|title=Applications - Keras Documentation|website=keras.io}}

| {{Yes}}

| {{Yes}}

| {{No}}{{Cite web|url=https://github.com/keras-team/keras/issues/461|title=Is there RBM in Keras? · Issue #461 · keras-team/keras|website=GitHub}}

| {{Yes}}{{Cite web|url=https://github.com/keras-team/keras/issues/2436|title=Does Keras support using multiple GPUs? · Issue #2436 · keras-team/keras|website=GitHub}}

| {{Yes}}

| MathWorks

| 1992

| {{Proprietary}}

| {{No}}

| Linux, macOS, Windows

| C, C++, Java, MATLAB

| MATLAB

| {{No}}

| {{No}}

| {{Yes|Train with Parallel Computing Toolbox and generate CUDA code with GPU Coder}}{{cite web|title=GPU Coder - MATLAB & Simulink|url=https://www.mathworks.com/products/gpu-coder.html|website=MathWorks|accessdate=13 November 2017}}

| {{No}}

| {{Yes}}{{cite web |title=Automatic Differentiation Background - MATLAB & Simulink |website=MathWorks |date=September 3, 2019 |url=https://www.mathworks.com/help/deeplearning/ug/deep-learning-with-automatic-differentiation-in-matlab.html |access-date=November 19, 2019}}

| {{Yes}}{{cite web|title=Neural Network Toolbox - MATLAB|url=https://www.mathworks.com/products/neural-network.html|website=MathWorks|accessdate=13 November 2017}}{{cite web|title=Deep Learning Models - MATLAB & Simulink|url=https://www.mathworks.com/solutions/deep-learning/models.html|website=MathWorks|accessdate=13 November 2017}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{Yes|With Parallel Computing Toolbox}}{{cite web|title=Parallel Computing Toolbox - MATLAB|url=https://www.mathworks.com/products/parallel-computing.html|website=MathWorks|accessdate=13 November 2017}}

| {{Yes}}

| Microsoft Research

| 2016

| {{Free|MIT license}}{{cite web |url=https://github.com/Microsoft/CNTK/blob/master/LICENSE.md |title=CNTK/LICENSE.md at master · Microsoft/CNTK |work=GitHub}}

| {{Yes}}

| Windows, Linux{{cite web|url=https://github.com/Microsoft/CNTK/wiki/Setup-CNTK-on-your-machine|title=Setup CNTK on your machine|work=GitHub}} (macOS via Docker on roadmap)

| C++

| Python (Keras), C++, Command line,{{cite web|url=https://github.com/Microsoft/CNTK/wiki/CNTK-usage-overview|title=CNTK usage overview|work=GitHub}} BrainScript{{cite web|url=https://github.com/Microsoft/CNTK/wiki/BrainScript-Network-Builder|title=BrainScript Network Builder|work=GitHub}} (.NET on roadmap{{cite web|url=https://github.com/Microsoft/CNTK/issues/960|title=.NET Support · Issue #960 · Microsoft/CNTK|work=GitHub}})

| {{Yes}}{{cite web|url=https://github.com/Microsoft/CNTK/issues/59#issuecomment-178104505|title=How to train a model using multiple machines? · Issue #59 · Microsoft/CNTK|work=GitHub}}

| {{No}}

| {{Yes}}

| {{No}}

| {{Yes}}

| {{Yes}}{{Cite web|url=https://github.com/microsoft/CNTK/issues/140|title=Prebuilt models for image classification · Issue #140 · microsoft/CNTK|website=GitHub}}

| {{Yes}}{{cite web|url=http://www.cntk.ai/|title=CNTK - Computational Network Toolkit|publisher=Microsoft Corporation}}

| {{Yes}}

| {{No}}{{cite web| title= Restricted Boltzmann Machine with CNTK #534| url=https://github.com/Microsoft/CNTK/issues/534| publisher=GitHub, Inc.| date=27 May 2016| access-date=30 October 2023}}

| {{Yes}}{{cite web|url=https://github.com/Microsoft/CNTK/wiki/Multiple-GPUs-and-machines|title=Multiple GPUs and machines|publisher=Microsoft Corporation}}

| {{No}}{{cite web|url=https://github.com/Microsoft/CNTK#disclaimer|title=Disclaimer|date=6 November 2021|publisher=CNTK TEAM}}

| Huawei

| 2020

| {{free|Apache 2.0}}

| {{Yes}}

| Linux, Windows, macOS, EulerOS, openEuler, OpenHarmony, Oniro OS, HarmonyOS, Android

| C++, Rust, Julia, Python, ArkTS, Cangjie, Java (Lite)

|

|

|

|

|

|

|

|

|

|

|

|

|Microsoft

|2018

|{{Free|MIT license}}

|{{Yes}}

|Windows, Linux, macOS

|C#, C++

|C#, F#

|

|

|

|

|

|

|

|

|

|

|{{Yes}}

| Apache Software Foundation

| 2015

| {{Free|Apache 2.0}}

| {{Yes}}

| Linux, macOS, Windows,{{cite web|url=https://github.com/dmlc/mxnet/releases|title=Releases · dmlc/mxnet|work=Github}}{{cite web|url=https://mxnet.readthedocs.io/en/latest/how_to/build.html#building-on-windows|title=Installation Guide — mxnet documentation|work=Readthdocs}} AWS, Android,{{cite web|url=https://mxnet.readthedocs.io/en/latest/how_to/smart_device.html|title=MXNet Smart Device|work=ReadTheDocs|access-date=2016-05-19|archive-url=https://web.archive.org/web/20160921205959/http://mxnet.readthedocs.io/en/latest/how_to/smart_device.html|archive-date=2016-09-21|url-status=dead}} iOS, JavaScript{{cite web|url=https://github.com/dmlc/mxnet.js|title=MXNet.js|work=Github|date=28 October 2021}}

| Small C++ core library

| C++, Python, Julia, MATLAB, JavaScript, Go, R, Scala, Perl, Clojure

| {{Yes}}

| {{No}}

| {{Yes}}

| {{No}}

| {{Yes}}{{Cite web|url=https://mxnet.readthedocs.io/en/latest/|title=— Redirecting to mxnet.io|website=mxnet.readthedocs.io}}

| {{Yes}}{{cite web|url=https://github.com/dmlc/mxnet-model-gallery|title=Model Gallery|work=GitHub|date=29 October 2022}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{Yes}}{{cite web|url=https://mxnet.readthedocs.io/en/latest/how_to/multi_devices.html|title=Run MXNet on Multiple CPU/GPUs with Data Parallel|work=GitHub}}

| {{No}}

| Artelnics

| 2014

| {{Proprietary}}

| {{No}}

| Linux, macOS, Windows

| C++

| Graphical user interface

| {{Yes}}

| {{No}}

| {{Yes}}

| {{No}}

| {{No|Analytical differentiation}}

| {{No}}

| {{No}}

| {{No}}

| {{No}}

| {{Yes}}

| {{Yes}}

| Artelnics

| 2003

| {{LGPL-lic}}

| {{Yes}}

| Cross-platform

| C++

| C++

| {{Yes}}

| {{No}}

| {{Yes}}

| {{No}}

| {{Dunno}}

| {{Dunno}}

| {{No}}

| {{No}}

| {{No}}

| {{Dunno}}

| {{Yes}}

| Vertex.AI, Intel

| 2017

| {{Free|Apache 2.0}}

| {{Yes}}

| Linux, macOS, Windows

| Python, C++, OpenCL

| Python, C++

| {{Dunno}}

| {{Yes|Some OpenCL ICDs are not recognized}}

| {{No}}

| {{No}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

|

| {{Yes}}

| {{Yes}}

| Meta AI

| 2016

| {{BSD-lic}}

| {{Yes}}

| Linux, macOS, Windows, Android{{Cite web|url=https://pytorch.org/mobile/android/|title=PyTorch|date=Dec 17, 2021}}

| Python, C, C++, CUDA

| Python, C++, Julia, R{{Cite web |title=Falbel D, Luraschi J (2023). torch: Tensors and Neural Networks with 'GPU' Acceleration.|url=https://torch.mlverse.org/ |access-date=2023-11-28 |website=torch.mlverse.org |language=en-us}}

| {{Yes}}

| {{Depends|Via separately maintained package}}{{Cite web|url=https://github.com/hughperkins/pytorch-coriander|title=OpenCL build of pytorch: (in-progress, not useable) - hughperkins/pytorch-coriander|date=July 14, 2019|via=GitHub}}{{Cite web|url=https://github.com/artyom-beilis/pytorch_dlprim|title=DLPrimitives/OpenCL out of tree backend for pytorch - artyom-beilis/pytorch_dlprim|date=Jan 21, 2022|via=GitHub}}{{Cite web|url=https://github.com/pytorch/pytorch/issues/488|title=OpenCL Support · Issue #488 · pytorch/pytorch|website=GitHub}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{Yes}}{{Cite web|url=https://github.com/GabrielBianconi/pytorch-rbm/blob/master/rbm.py|title=Restricted Boltzmann Machines (RBMs) in PyTorch|website=GitHub|date=14 November 2022}}

| {{Yes}}

| {{Yes}}

| Apache Software Foundation

| 2015

| {{Free|Apache 2.0}}

| {{Yes}}

| Linux, macOS, Windows

| C++

| Python, C++, Java

| {{No}}

| {{Depends|Supported in V1.0}}

| {{Yes}}

| {{No}}

| {{Dunno}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| Google Brain

| 2015

| {{Free|Apache 2.0}}

| {{Yes}}

| Linux, macOS, Windows,{{Cite web|url=https://www.tensorflow.org/install/pip|title=Install TensorFlow with pip}}{{Cite web|url=https://developers.googleblog.com/2016/11/tensorflow-0-12-adds-support-for-windows.html|title=TensorFlow 0.12 adds support for Windows}} Android

| C++, Python, CUDA

| Python (Keras), C/C++, Java, Go, JavaScript, R,{{cite web |title=tensorflow: R Interface to 'TensorFlow' |author1=Allaire, J.J. |author2=Kalinowski, T. |author3=Falbel, D. |author4=Eddelbuettel, D. |author5=Yuan, T. |author6=Golding, N. |url=https://cran.r-project.org/web/packages/tensorflow/ |publisher=The Comprehensive R Archive Network |date=28 September 2023 |access-date=30 October 2023}} Julia, Swift

| {{No}}

| {{Depends|On roadmap}}{{cite web |url=https://github.com/tensorflow/tensorflow/blob/master/tensorflow/docs_src/about/roadmap.md |title=tensorflow/roadmap.md at master |work=GitHub |date=January 23, 2017 |access-date=May 21, 2017}} but already with SYCL{{cite web|url=https://github.com/tensorflow/tensorflow/issues/22|title=OpenCL support |issue=22 |work=GitHub}} support

| {{Yes}}

| {{Yes}}

| {{Yes}}{{Cite web|url=https://www.tensorflow.org/|title=TensorFlow|website=TensorFlow}}

| {{Yes}}{{Cite web|url=https://github.com/tensorflow/models|title=Models and examples built with TensorFlow|date=July 19, 2019|via=GitHub}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| Université de Montréal

| 2007

| {{BSD-lic}}

| {{Yes}}

| Cross-platform

| Python

| Python (Keras)

| {{Yes}}

| {{Depends|Under development{{cite web|url=http://deeplearning.net/software/theano/tutorial/using_gpu.html|title=Using the GPU: Theano 0.8.2 documentation|access-date=2016-01-21|archive-date=2017-04-01|archive-url=https://web.archive.org/web/20170401163303/http://deeplearning.net/software/theano/tutorial/using_gpu.html|url-status=dead}}}}

| {{Yes}}

| {{No}}

| {{Yes}}{{Cite web|url=http://deeplearning.net/software/theano/library/gradient.html|title=gradient – Symbolic Differentiation — Theano 1.0.0 documentation|website=deeplearning.net}}{{Cite web|url=https://groups.google.com/d/msg/theano-users/mln5g2IuBSU/gespG36Lf_QJ|title=Automatic vs. Symbolic differentiation}}

| {{Depends|Through Lasagne's model zoo{{cite web|url=https://github.com/Lasagne/Recipes/tree/master/modelzoo|title=Recipes/modelzoo at master · Lasagne/Recipes |work=GitHub}}}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{Yes}}{{Cite web|url=http://deeplearning.net/software/theano/tutorial/using_multi_gpu.html|title=Using multiple GPUs — Theano 1.0.0 documentation|website=deeplearning.net}}

| {{No}}

| Ronan Collobert, Koray Kavukcuoglu, Clement Farabet

| 2002

| {{BSD-lic}}

| {{Yes}}

| Linux, macOS, Windows,{{Cite web|url=https://github.com/torch/torch7|title=torch/torch7|date=July 18, 2019|via=GitHub}} Android,{{cite web|url=https://github.com/soumith/torch-android|title=GitHub - soumith/torch-android: Torch-7 for Android|work=GitHub|date=13 October 2021}} iOS

| C, Lua

| Lua, LuaJIT,{{Cite web|url=http://ronan.collobert.com/pub/matos/2011_torch7_nipsw.pdf|title=Torch7: A MATLAB-like Environment for Machine Learning}} C, utility library for C++/OpenCL{{cite web|url=https://github.com/jonathantompson/jtorch|title=GitHub - jonathantompson/jtorch: An OpenCL Torch Utility Library|work=GitHub|date=18 November 2020}}

| {{Yes}}

| {{Depends|Third party implementations{{cite web|url=https://github.com/torch/torch7/wiki/Cheatsheet#opencl|title=Cheatsheet|work=GitHub}}{{cite web|url=https://github.com/hughperkins/distro-cl|title=cltorch|work=GitHub}}}}

| {{Yes}}{{cite web|url=https://github.com/torch/cutorch|title=Torch CUDA backend|work=GitHub}}{{cite web|url=https://github.com/torch/cunn|title=Torch CUDA backend for nn|work=GitHub}}

| {{No}}

| {{Yes|Through Twitter's Autograd{{Cite web|url=https://github.com/twitter/torch-autograd|title=Autograd automatically differentiates native Torch code: twitter/torch-autograd|date=July 9, 2019|via=GitHub}}}}

| {{Yes}}{{cite web|url=https://github.com/torch/torch7/wiki/ModelZoo|title=ModelZoo|work=GitHub}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{No}}

| Wolfram Research

| 2014

| {{Proprietary}}

| {{No}}

| Windows, macOS, Linux, Cloud computing

| C++, Wolfram Language, CUDA

| Wolfram Language

| {{Yes}}

| {{No}}

| {{Yes}}

| {{No}}

| {{Yes}}

| {{Yes}}{{Cite web|url=http://resources.wolframcloud.com/NeuralNetRepository|title=Wolfram Neural Net Repository of Neural Network Models|website=resources.wolframcloud.com}}

| {{Yes}}

| {{Yes}}

| {{Yes}}

| {{Yes}}{{Cite web|url=https://reference.wolfram.com/language/guide/ParallelComputing.html.en|title=Parallel Computing—Wolfram Language Documentation|website=reference.wolfram.com}}

| {{Yes}}

|+

!Format name

!Design goal

!Compatible with other formats

!Self-contained DNN Model

!Pre-processing and Post-processing

!Run-time configuration for tuning & calibration

!DNN model interconnect

!Common platform

|Algorithm training

| {{No}}

| {{No}} / Separate files in most formats

| {{No}}

| {{No}}

| {{No}}

| {{Yes}}

|Algorithm training

| {{Yes}}

| {{No}} / Separate files in most formats

| {{No}}

| {{No}}

| {{No}}

| {{Yes}}