|
| 1 | +--- |
| 2 | +title: Deep Learning & AI frameworks |
| 3 | +titleSuffix: Azure Data Science Virtual Machine |
| 4 | +description: Available deep learning frameworks and tools on Azure Data Science Virtual Machine. |
| 5 | +keywords: data science tools, data science virtual machine, tools for data science, linux data science |
| 6 | +services: machine-learning |
| 7 | +ms.service: machine-learning |
| 8 | +ms.subservice: data-science-vm |
| 9 | + |
| 10 | +author: gvashishtha |
| 11 | +ms.author: gopalv |
| 12 | +ms.topic: conceptual |
| 13 | +ms.date: 10/1/2019 |
| 14 | +--- |
| 15 | + |
| 16 | +# Deep learning and AI frameworks for the Azure Data Science VM |
| 17 | +Deep learning frameworks on the DSVM are listed below. |
| 18 | + |
| 19 | +## [Caffe](https://github.com/BVLC/caffe) |
| 20 | + |
| 21 | +| | | |
| 22 | +| ------------- | ------------- | |
| 23 | +| Version(s) supported | | |
| 24 | +| Supported DSVM editions | Linux (Ubuntu) | |
| 25 | +| How is it configured / installed on the DSVM? | Caffe is installed in `/opt/caffe`. Samples are in `/opt/caffe/examples`.| |
| 26 | +| How to run it | use X2Go to sign in to your VM, and then start a new terminal and enter the following:<br/>`cd /opt/caffe/examples`<br/>`source activate root`<br/>`jupyter notebook`<br/><br/>A new browser window opens with sample notebooks. Binaries are installed in /opt/caffe/build/install/bin.<br/><br/>Installed version of Caffe requires Python 2.7 and won't work with Python 3.5, which is activated by default. To switch to Python 2.7, run `source activate root` to switch to Anaconda environment.| |
| 27 | + |
| 28 | +## [Caffe2](https://github.com/caffe2/caffe2) |
| 29 | + |
| 30 | +| | | |
| 31 | +| ------------- | ------------- | |
| 32 | +| Version(s) supported | | |
| 33 | +| Supported DSVM editions | Linux (Ubuntu) | |
| 34 | +| How is it configured / installed on the DSVM? | Caffe2 is installed in the [Python 2.7 (root) conda environment. | |
| 35 | +| How to run it | Terminal: Start Python, and import Caffe2. <br/> * JupyterHub: [Connect to JupyterHub](dsvm-ubuntu-intro.md#how-to-access-the-ubuntu-data-science-virtual-machine), and then go to the Caffe2 directory to find sample notebooks. Some notebooks require the Caffe2 root to be set in the Python code; enter /opt/caffe2. | |
| 36 | + |
| 37 | +## [Chainer](https://chainer.org/) |
| 38 | + |
| 39 | +| | | |
| 40 | +| ------------- | ------------- | |
| 41 | +| Version(s) supported | 5.2 | |
| 42 | +| Supported DSVM editions | Linux (Ubuntu) | |
| 43 | +| How is it configured / installed on the DSVM? | Chainer is installed in Python 3.5. | |
| 44 | +| How to run it | Terminal: Activate the Python 3.5 environment, run `python`, and then `import chainer`. <br/> * JupyterHub: [Connect to JupyterHub](dsvm-ubuntu-intro.md#how-to-access-the-ubuntu-data-science-virtual-machine), and then go to the Chainer directory to find sample notebooks.| |
| 45 | + |
| 46 | +## [CUDA, cuDNN, NVIDIA Driver](https://developer.nvidia.com/cuda-toolkit) |
| 47 | + |
| 48 | +| | | |
| 49 | +| ------------- | ------------- | |
| 50 | +| Version(s) supported | 10.0.130| |
| 51 | +| Supported DSVM editions | Windows and Linux | |
| 52 | +| How is it configured / installed on the DSVM? |_nvidia-smi_ is available on the system path. | |
| 53 | +| How to run it | Open a command prompt (on Windows) or a terminal (on Linux), and then run _nvidia-smi_. | |
| 54 | + |
| 55 | + |
| 56 | +## [Horovod](https://github.com/uber/horovod) |
| 57 | + |
| 58 | +| | | |
| 59 | +| ------------- | ------------- | |
| 60 | +| Version(s) supported | 0.16.1| |
| 61 | +| Supported DSVM editions | Linux (Ubuntu) | |
| 62 | +| How is it configured / installed on the DSVM? | Horovod is installed in Python 3.5 | |
| 63 | +| How to run it | Activate the correct environment at the terminal, and then run Python. | |
| 64 | + |
| 65 | +## [Keras](https://keras.io/) |
| 66 | + |
| 67 | +| | | |
| 68 | +| ------------- | ------------- | |
| 69 | +| Version(s) supported | 2.2.4 | |
| 70 | +| Supported DSVM editions | Windows and Linux | |
| 71 | +| How is it configured / installed on the DSVM? | Keras is installed in Python 3.6 on Windows and in Python 3.5 in Linux | |
| 72 | +| How to run it | Activate the correct environment at the terminal, and then run Python. | |
| 73 | + |
| 74 | +## [Microsoft Cognitive Toolkit (CNTK)](https://docs.microsoft.com/cognitive-toolkit/) |
| 75 | + |
| 76 | +| | | |
| 77 | +| ------------- | ------------- | |
| 78 | +| Version(s) supported | 2.5.1 | |
| 79 | +| Supported DSVM editions | Windows and Linux | |
| 80 | +| How is it configured / installed on the DSVM? | CNTK is installed in Python 3.6 on [Windows 2016](dsvm-languages.md#python-windows-server-2016-edition) and in Python 3.5 on [Linux](./dsvm-languages.md#python-linux-edition)) | |
| 81 | +| How to run it | Terminal: Activate the correct environment and run Python. <br/>Jupyter: Connect to [Jupyter](provision-vm.md) or [JupyterHub](dsvm-ubuntu-intro.md#how-to-access-the-ubuntu-data-science-virtual-machine), and then open the CNTK directory for samples. | |
| 82 | + |
| 83 | +## [MXNet](https://mxnet.apache.org/) |
| 84 | +| | | |
| 85 | +| ------------- | ------------- | |
| 86 | +| Version(s) supported | 1.3.0 | |
| 87 | +| Supported DSVM editions | Windows and Linux | |
| 88 | +| How is it configured / installed on the DSVM? | MXNet is installed in `C:\dsvm\tools\mxnet` on Windows and `/dsvm/tools/mxnet` on Ubuntu. Python bindings are installed in Python 3.6 on [Windows 2016](dsvm-languages.md#python-windows-server-2016-edition) and in Python 3.5 on [Linux](./dsvm-languages.md#python-linux-edition)) R bindings are also included in the Ubuntu DSVM. | |
| 89 | +| How to run it | Terminal: Activate the correct conda environment, then run `import mxnet`. <br/>Jupyter: Connect to [Jupyter](provision-vm.md#access-the-dsvm) or [JupyterHub](dsvm-ubuntu-intro.md#how-to-access-the-ubuntu-data-science-virtual-machine), and then open the `mxnet` directory for samples. | |
| 90 | + |
| 91 | +## [MXNet Model Server](https://github.com/awslabs/mxnet-model-server#quick-start) |
| 92 | + |
| 93 | +| | | |
| 94 | +| ------------- | ------------- | |
| 95 | +| Version(s) supported | 1.0.1 | |
| 96 | +| Supported DSVM editions | Windows and Linux | |
| 97 | +| How is it configured / installed on the DSVM? | MXNet Model Server is installed in Python 3.6 on [Windows 2016](dsvm-languages.md#python-windows-server-2016-edition) and in Python 3.5 on [Linux](./dsvm-languages.md#python-linux-edition)) | |
| 98 | +| How to run it | Terminal: Run `sudo systemctl stop jupyterhub` to stop the JupyterHub service first, because both listen on the same port. Then activate the correct conda environment and run `mxnet-model-server --start --models squeezenet=https://s3.amazonaws.com/model-server/model_archive_1.0/squeezenet_v1.1.mar` | |
| 99 | + |
| 100 | +## [NVidia System Management Interface (nvidia-smi)](https://developer.nvidia.com/nvidia-system-management-interface) |
| 101 | + |
| 102 | +| | | |
| 103 | +| ------------- | ------------- | |
| 104 | +| Version(s) supported | | |
| 105 | +| Supported DSVM editions | Windows and Linux | |
| 106 | +| What is it for? | NVIDIA tool for querying GPU activity | |
| 107 | +| How is it configured / installed on the DSVM? | `nvidia-smi` is on the system path. | |
| 108 | +| How to run it | On a virtual machine **with GPU's**, open a command prompt (on Windows) or a terminal (on Linux), and then run `nvidia-smi`. | |
| 109 | + |
| 110 | +## [PyTorch](https://pytorch.org/) |
| 111 | + |
| 112 | +| | | |
| 113 | +| ------------- | ------------- | |
| 114 | +| Version(s) supported | 1.2.0 (Windows 2016, Windows 2019, Ubuntu 16.04), 1.4.0 (Ubuntu 18.04) | |
| 115 | +| Supported DSVM editions | Linu, Windows | |
| 116 | +| How is it configured / installed on the DSVM? | Installed in [Python 3.5](dsvm-languages.md#python-linux-edition). Sample Jupyter notebooks are included, and samples are in /dsvm/samples/pytorch. | |
| 117 | +| How to run it | Terminal: Activate the correct environment, and then run Python.<br/>* [JupyterHub](dsvm-ubuntu-intro.md#how-to-access-the-ubuntu-data-science-virtual-machine): Connect, and then open the PyTorch directory for samples. | |
| 118 | + |
| 119 | +## [TensorFlow](https://www.tensorflow.org/) |
| 120 | + |
| 121 | +| | | |
| 122 | +| ------------- | ------------- | |
| 123 | +| Version(s) supported | 1.13 | |
| 124 | +| Supported DSVM editions | Windows, Linux | |
| 125 | +| How is it configured / installed on the DSVM? | Installed in Python 3.5 on [Linux](dsvm-languages.md#python-linux-edition) and Python 3.6 on [Windows 2016](dsvm-languages.md#python-windows-server-2016-edition) | |
| 126 | +| How to run it | Terminal: Activate the correct environment, and then run Python. <br/> * Jupyter: Connect to [Jupyter](provision-vm.md) or [JupyterHub](dsvm-ubuntu-intro.md#how-to-access-the-ubuntu-data-science-virtual-machine), and then open the TensorFlow directory for samples. | |
| 127 | + |
| 128 | +## [TensorFlow Serving](https://www.tensorflow.org/serving/) |
| 129 | + |
| 130 | +| | | |
| 131 | +| ------------- | ------------- | |
| 132 | +| Version(s) supported | 1.12 | |
| 133 | +| Supported DSVM editions | Linux | |
| 134 | +| How is it configured / installed on the DSVM? | tensorflow_model_server is available at the terminal. | |
| 135 | +| How to run it | Samples are available [online](https://www.tensorflow.org/serving/). | |
| 136 | + |
| 137 | + |
| 138 | +## [Theano](https://github.com/Theano/Theano) |
| 139 | + |
| 140 | +| | | |
| 141 | +| ------------- | ------------- | |
| 142 | +| Version(s) supported | 1.0.3 | |
| 143 | +| Supported DSVM editions | Linux | |
| 144 | +| How is it configured / installed on the DSVM? |Theano is installed in Python 2.7 (_root_), and in Python 3.5 (_py35_) environment. | |
| 145 | +| How to run it | Terminal: Activate the Python version you want (root or py35), run Python, and then import Theano.<br/>* Jupyter: Select the Python 2.7 or 3.5 kernel, and then import Theano. <br/>To work around a recent math kernel library (MKL) bug, you need to first set the MKL threading layer as follows:<br/><br/>`export MKL_THREADING_LAYER=GNU` | |
0 commit comments