intel/intel-extension-for-tensorflow repository overview

PROJECT NOT UNDER ACTIVE MANAGEMENT. This image repo will no longer be maintained by Intel.

⁠Intel® Extension for TensorFlow*

Intel® Extension for TensorFlow*⁠ extends TensorFlow*⁠ with up-to-date feature optimizations for an extra performance boost on Intel hardware.

Intel® Extension for TensorFlow*⁠ is based on the TensorFlow PluggableDevice⁠ interface to bring Intel XPU(GPU, CPU, etc.) devices into TensorFlow*⁠ with flexibility for on-demand performance on the following Intel GPUs:

• Intel® Arc™ A-Series Graphics⁠
• Intel® Data Center GPU Flex Series⁠
• Intel® Data Center GPU Max Series⁠

⁠XPU images

The images below include support for both CPU and GPU optimizations:

The following images include support for Intel® Deep Learning Essentials:

| Tag(s) | TensorFlow | ITEX | Driver | DL Essentials | Dockerfile | | ---------------------- | ----------- | -------------- | ------- | --------------- | | 2.15.0.3-xpu-pip-dl-essentials | v2.15.1⁠ | v2.15.0.3⁠ | 1099⁠| 2025.0.2-6⁠ | v0.4.0-Beta⁠ |

⁠Run the XPU Container

docker run -it --rm \
    --device /dev/dri \
    -v /dev/dri/by-path:/dev/dri/by-path \
    --ipc=host \
    intel/intel-extension-for-tensorflow:xpu

The images below additionally include Jupyter Notebook⁠ server:

⁠Run the XPU Jupyter Container

docker run -it --rm \
    -p 8888:8888 \
    --net=host \
    --device /dev/dri \
    -v /dev/dri/by-path:/dev/dri/by-path \
    --ipc=host \
    intel/intel-extension-for-tensorflow:2.15.0.3-xpu-pip-jupyter

After running the command above, copy the URL (something like http://127.0.0.1:$PORT/?token=***) into your browser to access the notebook server.

The images below are TensorFlow* Serving⁠ with GPU Optimizations:

⁠Run the Serving GPU Container

docker run -it --rm \
    -p 8500:8500 \
    --device /dev/dri \
    -v /dev/dri/by-path:/dev/dri/by-path \
    -v $PWD/workspace:/workspace \
    -w /workspace \
    -e MODEL_NAME=<your-model-name> \
    -e MODEL_DIR=<your-model-dir> \
    intel/intel-extension-for-tensorflow:serving-gpu

For more details, follow the procedure in the Intel® Extension for TensorFlow* Serving⁠ instructions.

⁠CPU only images

The images below are built only with CPU optimizations (GPU acceleration support was deliberately excluded):

The images below additionally include Jupyter Notebook⁠ server:

⁠Run the CPU Jupyter Container

docker run -it --rm \
    -p 8888:8888 \
    --net=host \
    -v $PWD/workspace:/workspace \
    -w /workspace \
    intel/intel-extension-for-tensorflow:2.15.1-pip-jupyter

After running the command above, copy the URL (something like http://127.0.0.1:$PORT/?token=***) into your browser to access the notebook server.

The images below additionally include Horovod⁠:

⁠Setup and Run ITEX Multi-Node Container

Some additional assembly is required to utilize this container with OpenSSH. To perform any kind of DDP (Distributed Data Parallel) execution, containers are assigned the roles of launcher and worker respectively:

SSH Server (Worker)

1. Authorized Keys : /etc/ssh/authorized_keys

SSH Client (Launcher)

1. Private User Key : /root/.ssh/id_rsa

To add these files correctly please follow the steps described below.

1. Setup ID Keys

   You can use the commands provided below to generate the identity keys⁠ for OpenSSH.

   ssh-keygen -q -N "" -t rsa -b 4096 -f ./id_rsa
   touch authorized_keys
   cat id_rsa.pub >> authorized_keys
   

   Copy

2. Configure the permissions and ownership for all of the files you have created so far

   chmod 600 id_rsa config authorized_keys
   chown root:root id_rsa.pub id_rsa config authorized_keys
   

   Copy

3. Create a hostfile for horovod. (Optional)

   Host host1
       HostName <Hostname of host1>
       IdentitiesOnly yes
       IdentityFile ~/.root/id_rsa
       Port <SSH Port>
   Host host2
       HostName <Hostname of host2>
       IdentitiesOnly yes
       IdentityFile ~/.root/id_rsa
       Port <SSH Port>
   ...
   

   Copy

4. Configure Horovod⁠ in your python script

   import horovod.torch as hvd
   
   hvd.init()
   

   Copy

5. Now start the workers and execute DDP on the launcher

   1. Worker run command:

      docker run -it --rm \
          --net=host \
          -v $PWD/authorized_keys:/etc/ssh/authorized_keys \
          -v $PWD/tests:/workspace/tests \
          -w /workspace \
          intel/intel-optimized-tensorflow:2.15.1-pip-multinode \
          bash -c '/usr/sbin/sshd -D'
      

      Copy

   2. Launcher run command:

      docker run -it --rm \
          --net=host \
          -v $PWD/id_rsa:/root/.ssh/id_rsa \
          -v $PWD/tests:/workspace/tests \
          -v $PWD/hostfile:/root/ssh/config \
          -w /workspace \
          intel/intel-optimized-tensorflow:2.15.1-pip-multinode \
          bash -c 'horovodrun --verbose -np 2 -H host1:1,host2:1 /workspace/tests/tf_base_test.py'
      

      Copy

The images below are TensorFlow* Serving⁠ with CPU Optimizations:

⁠Run the Serving CPU Container

docker run -it --rm \
    -p 8500:8500 \
    --device /dev/dri \
    -v /dev/dri/by-path:/dev/dri/by-path \
    -v $PWD/workspace:/workspace \
    -w /workspace \
    -e MODEL_NAME=<your-model-name> \
    -e MODEL_DIR=<your-model-dir> \
    intel/intel-extension-for-tensorflow:serving-cpu

For more details, follow the procedure in the Intel® Extension for TensorFlow* Serving⁠ instructions.

⁠CPU only images with Intel® Distribution for Python*

The images below are built only with CPU optimizations (GPU acceleration support was deliberately excluded) and include Intel® Distribution for Python*⁠:

The images below additionally include Jupyter Notebook⁠ server:

The images below additionally include Horovod⁠:

⁠XPU images with Intel® Distribution for Python*

The images below are built only with CPU and GPU optimizations and include Intel® Distribution for Python*⁠:

The following images include support for Intel® Deep Learning Essentials:

| Tag(s) | TensorFlow | ITEX | Driver | DL Essentials | Dockerfile | | ---------------------- | ----------- | -------------- | ------- | --------------- | | 2.15.0.3-xpu-idp-dl-essentials | v2.15.1⁠ | v2.15.0.3⁠ | 1099⁠| 2025.0.2-6⁠ | v0.4.0-Beta⁠ |

The images below additionally include Jupyter Notebook⁠ server:

⁠Build from Source

To build the images from source, clone the AI Containers⁠ repository, follow the main README.md file to setup your environment, and run the following command:

cd pytorch
docker compose build tf-base
docker compose run tf-base

You can find the list of services below for each container in the group:

⁠License

View the License⁠ for the Intel® Extension for TensorFlow*⁠.

The images below also contain other software which may be under other licenses (such as TensorFlow*, Jupyter*, Bash, etc. from the base).

It is the image user's responsibility to ensure that any use of The images below comply with any relevant licenses for all software contained within.

* Other names and brands may be claimed as the property of others.