Habana Base Deep Learning AMI released and Framework cleanup

* Fix PYTHON path for pip installs TF framework
* --user for TF install directions
* Remove coming soon for Habana Base AMI
* Fix dockerfiles to workaround TF 2.6.0 release install bug

README.md

@@ -17,9 +17,9 @@
  - [Check Habana Package Installation for no Docker](#check-habana-package-installation-for-no-docker)
  - [Install SW Stack](#install-sw-stack)
  - [Check TF/Horovod Habana packages](#check-tfhorovod-habana-packages)
- - [Install TF/Horovod Habana python packages](#install-tfhorovod-habana-python-packages)
+ - [Install TF/Horovod Habana packages](#install-tfhorovod-habana-packages)
  - [Check PT Habana packages](#check-pt-habana-packages)
- - [Install PT Habana python packages](#install-pt-habana-python-packages)
+ - [Install PT Habana packages](#install-pt-habana-packages)
  - Docker
  - [Do you want to use prebuilt docker or build docker yourself?](#do-you-want-to-use-prebuilt-docker-or-build-docker-yourself)
  - [How to Build Docker Images from Habana Dockerfiles](#how-to-build-docker-images-from-habana-dockerfiles)
@@ -196,12 +196,6 @@ Setup complete, please proceed to [Setup Complete](#Setup-Complete)
 
 ## Habana Deep Learning AMI from AWS Marketplace
 
-<center>
-
-**--- Coming Soon ---**
-
-</center>
-
 When using the Habana Deep Learning AMI from AWS Marketplace, you can either directly use containers or install a framework and proceed from there to run directly on the AMI. 
 <br />
 
@@ -999,7 +993,7 @@ ${PYTHON} -m pip list | grep habana
 <center>
 
 ### Are the required python packages installed on your system?
-[Yes](#Setup-Complete) • [No](#Install-TFHorovod-Habana-python-packages)
+[Yes](#Setup-Complete) • [No](#install-tfhorovod-habana-packages)
 
 </center>
 
@@ -1009,7 +1003,7 @@ ${PYTHON} -m pip list | grep habana
 
 <br />
 
-## Install TF/Horovod Habana python packages
+## Install TF/Horovod Habana packages
 This section describes how to obtain and install the TensorFlow software package. The package consists of two main components: 
 
 Base **habana-tensorflow** Python package - Libraries and modules needed to execute TensorFlow on a **single Gaudi** device. 
@@ -1021,7 +1015,9 @@ Scale-out **habana-horovod** Python package - Libraries and modules needed to ex
 
 <br />
 
-The following example scripts include instructions from the steps [Base Installation (Single Node)](#Base-Installation-Single-Node) and [Scale-out Installation](#Scale-out-Installation) that can be used for your reference. The scripts install TF 2.5.1.
+The following example scripts include instructions from the steps [Base Installation (Single Node)](#Base-Installation-Single-Node) and [Scale-out Installation](#Scale-out-Installation) that can be used for your reference. The scripts install TF 2.5.1. 
+The scripts are using Python3 from ``/usr/bin/`` with version according to the [Support Matrix](#SynapseAi-Support-Matrix). 
+Make sure, that Python3 is installed there, and if not, update the bash scripts with appropriate ``PYTHON=<path>``.
 
 Ubuntu 18.04 example script [u18_tensorflow_installation.sh](https://github.com/HabanaAI/Setup_and_Install/blob/r1.0.1/installation_scripts/u18_tensorflow_installation.sh).
 
@@ -1149,16 +1145,30 @@ This will search for and list all packages with the word Habana.
 ### Base Installation (Single Node)
 The habana-tensorflow package contains all the binaries and scripts to run topologies on a single-node. 
 
-1. Before installing habana-tensorflow, install supported TensorFlow version. See [Support Matrix](#SynapseAi-Support-Matrix). If no TensorFlow package is available, PIP will automatically fetch it.
+1. All the steps listed below are using to ``${PYTHON}`` environment variable, which must be set to appropriate version of Python, according to [Support Matrix](#SynapseAi-Support-Matrix).  
 ```
-${PYTHON} -m pip install tensorflow-cpu==<supported_tf_version>
+export PYTHON=/usr/bin/python<VER> # i.e. for U18 it's PYTHON=/usr/bin/python3.7
 ```
+2. Before installing habana-tensorflow, install supported TensorFlow version. See [Support Matrix](#SynapseAi-Support-Matrix). If no TensorFlow package is available, PIP will automatically fetch it. 
+**NOTE:**
+After TensorFlow release version 2.7.0, TensorFlow 2.6.0 has a broken dependency to TensorFlow Estimator, Keras and Tensorboard.
+To overcome this dependency, user needs to explicitly install the proper version of those packages before installing TensorFlow.
 
-2. habana-tensorflow is available in the Habana Vault. To allow PIP to search for the habana-tensorflow package, –extra-index-url needs to be specified:
 ```
-${PYTHON} -m pip install habana-tensorflow==1.0.1.81 --extra-index-url https://vault.habana.ai/artifactory/api/pypi/gaudi-python/simple
+# Only when installing tensorflow-cpu==2.6.0
+${PYTHON} -m pip install --user tensorflow-estimator==2.6.0
+${PYTHON} -m pip install --user tensorboard==2.6.0
+${PYTHON} -m pip install --user keras==2.6.0
+```
+Then install tensorflow-cpu:
 ```
-3. Run the below command to make sure the habana-tensorflow package is properly installed:
+${PYTHON} -m pip install --user tensorflow-cpu==<supported_tf_version>
+```
+3. habana-tensorflow is available in the Habana Vault. To allow PIP to search for the habana-tensorflow package, –extra-index-url needs to be specified:
+```
+${PYTHON} -m pip install --user habana-tensorflow==1.0.1.81 --extra-index-url https://vault.habana.ai/artifactory/api/pypi/gaudi-python/simple
+```
+4. Run the below command to make sure the habana-tensorflow package is properly installed:
 ```
 ${PYTHON} -c "import habana_frameworks.tensorflow as htf; print(htf.__version__)"
 ```
@@ -1212,11 +1222,11 @@ export PATH=$MPI_ROOT/bin:$PATH
 ```
 Install mpi4py binding
 ```
-python3 -m pip install mpi4py==3.0.3
+${PYTHON} -m pip install --user mpi4py==3.0.3
 ```
 3. habana-horovod is also stored in the Habana Vault. To allow PIP to search for the habana-horovod package, –extra-index-url needs to be specified:
 ```
-${PYTHON} -m pip install habana-horovod==1.0.1.81 --extra-index-url https://vault.habana.ai/artifactory/api/pypi/gaudi-python/simple
+${PYTHON} -m pip install --user habana-horovod==1.0.1.81 --extra-index-url https://vault.habana.ai/artifactory/api/pypi/gaudi-python/simple
 ```
 
 #### See also:
@@ -1274,7 +1284,7 @@ Python dependencies are gatehered in [model_requirements.txt](https://github.com
 
 Download the file and invoke:
 ```
-python3 -m pip install -r model_requirements.txt
+${PYTHON} -m pip install --user -r model_requirements.txt
 ```
 
 <br />
@@ -1371,7 +1381,7 @@ Check for habana-torch and habana-torch-hcl
 <center>
 
 ### Are the required python packages installed on your system?
-[Yes](#Setup-Complete) • [No](#Install-PT-Habana-python-packages)
+[Yes](#Setup-Complete) • [No](#install-pt-habana-packages)
 
 </center>
 
@@ -1381,7 +1391,7 @@ Check for habana-torch and habana-torch-hcl
 
 <br />
 
-## Install PT Habana python packages
+## Install PT Habana packages
 ### Install Habana Pytorch
 <details>
 <summary>Ubuntu distributions</summary>
@@ -2224,7 +2234,7 @@ It will look similar to this:
 * <details>
  <summary>TF 2.6.0</summary>
 
- ### Pull docker
+ ### Pull docker 
  ```
  docker pull vault.habana.ai/gaudi-docker/1.0.1/ubuntu18.04/habanalabs/tensorflow-installer-tf-cpu-2.6.0:1.0.1-81
  ```
@@ -2275,8 +2285,8 @@ It will look similar to this:
 
 * <details>
  <summary>TF 2.6.0</summary>
-
- ### Pull docker
+ 
+ ### Pull docker 
  ```
  docker pull vault.habana.ai/gaudi-docker/1.0.1/amzn2/habanalabs/tensorflow-installer-tf-cpu-2.6.0:1.0.1-81
  ```

dockerfiles/Dockerfile_amzn2_tensorflow_installer

@@ -56,6 +56,9 @@ RUN wget https://bootstrap.pypa.io/get-pip.py && \
  python3 get-pip.py pip==21.0.1 && \
  rm -rf get-pip.py && \
  pip3 install -r requirements-training-release.txt && \
+ pip3 install tensorflow-estimator==2.6.0 \
+ pip3 install tensorboard==2.6.0 \
+ pip3 install keras==2.6.0 \
  pip3 install tensorflow-cpu==${TF_VERSION} \
  tensorflow-model-optimization==0.5.0 && \
  # pycocotools has to be installed in separated process otherwise it fails with 'numpy.ufunc size changed'

dockerfiles/Dockerfile_ubuntu_tensorflow_installer

@@ -45,6 +45,9 @@ COPY requirements-training-release.txt requirements-training-release.txt
 RUN python3 -m pip install pip==21.0.1 && \
  pip3 install -r requirements-training-release.txt && \
  pip3 uninstall --yes habana tensorflow && \
+ pip3 install tensorflow-estimator==2.6.0 \
+ pip3 install tensorboard==2.6.0 \
+ pip3 install keras==2.6.0 \
  # tensorflow-cpu and -model have to be installed in separated processes otherwise old version of tf will be imported
  pip3 install tensorflow-cpu==${TF_VERSION} && \
  pip3 install tensorflow-model-optimization==0.5.0 && \

installation_scripts/al2_tensorflow_installation.sh

@@ -29,6 +29,7 @@ export MPI_ROOT=/usr/local/openmpi
 export LD_LIBRARY_PATH=$MPI_ROOT/lib:$LD_LIBRARY_PATH
 export OPAL_PREFIX=$MPI_ROOT
 export PATH=$MPI_ROOT/bin:$PATH
+export PYTHON=/usr/bin/python3.7
 echo "export MPI_ROOT=${MPI_ROOT}" | sudo tee -a /etc/profile.d/habanalabs.sh
 echo "export OPAL_PREFIX=${MPI_ROOT}" | sudo tee -a /etc/profile.d/habanalabs.sh
 echo 'export LD_LIBRARY_PATH=${MPI_ROOT}/lib:${LD_LIBRARY_PATH}' | sudo tee -a /etc/profile.d/habanalabs.sh
@@ -46,14 +47,13 @@ wget --no-verbose https://download.open-mpi.org/release/open-mpi/v4.0/openmpi-"$
  /sbin/ldconfig
 
 export MPICC=${MPI_ROOT}/bin/mpicc
-python3 -m pip install mpi4py==3.0.3
+${PYTHON} -m pip install --user mpi4py==3.0.3
 
 #install base tensorflow package
-python3 -m pip install tensorflow-cpu==2.5.1
-#instal Habana tensorflow bridge & Horovod
-python3 -m pip install habana-tensorflow==1.0.1.81 --extra-index-url https://vault.habana.ai/artifactory/api/pypi/gaudi-python/simple
-python3 -m pip install habana-horovod==1.0.1.81 --extra-index-url https://vault.habana.ai/artifactory/api/pypi/gaudi-python/simple
+${PYTHON} -m pip install --user tensorflow-cpu==2.5.1
+#install Habana tensorflow bridge & Horovod
+${PYTHON} -m pip install --user habana-tensorflow==1.0.1.81 --extra-index-url https://vault.habana.ai/artifactory/api/pypi/gaudi-python/simple
+${PYTHON} -m pip install --user habana-horovod==1.0.1.81 --extra-index-url https://vault.habana.ai/artifactory/api/pypi/gaudi-python/simple
 
 source /etc/profile.d/habanalabs.sh
-python3 -c 'import tensorflow as tf;import habana_frameworks.tensorflow as htf;htf.library_loader.load_habana_module();x = tf.constant(2);y = x + x;assert y.numpy() == 4, "Sanity check failed: Wrong Add output";assert "HPU" in y.device, "Sanity check failed: Operation not executed on Habana";print("Sanity check passed")'
-
+${PYTHON} -c 'import tensorflow as tf;import habana_frameworks.tensorflow as htf;htf.load_habana_module();x = tf.constant(2);y = x + x;assert y.numpy() == 4, "Sanity check failed: Wrong Add output";assert "HPU" in y.device, "Sanity check failed: Operation not executed on Habana";print("Sanity check passed")'

installation_scripts/u18_tensorflow_installation.sh

@@ -26,6 +26,7 @@ export MPI_ROOT=/usr/local/openmpi
 export LD_LIBRARY_PATH=$MPI_ROOT/lib:$LD_LIBRARY_PATH
 export OPAL_PREFIX=$MPI_ROOT
 export PATH=$MPI_ROOT/bin:$PATH
+export PYTHON=/usr/bin/python3.7
 echo "export MPI_ROOT=${MPI_ROOT}" | sudo tee -a /etc/profile.d/habanalabs.sh
 echo "export OPAL_PREFIX=${MPI_ROOT}" | sudo tee -a /etc/profile.d/habanalabs.sh
 echo 'export LD_LIBRARY_PATH=${MPI_ROOT}/lib:${LD_LIBRARY_PATH}' | sudo tee -a /etc/profile.d/habanalabs.sh
@@ -46,14 +47,13 @@ wget --no-verbose https://download.open-mpi.org/release/open-mpi/v4.0/openmpi-"$
  rm -rf openmpi-"${OPENMPI_VER}"* && \
  sudo /sbin/ldconfig
 
-python3 -m pip install mpi4py==3.0.3
+${PYTHON} -m pip install --user mpi4py==3.0.3
 
 #install base tensorflow package
-python3 -m pip install tensorflow-cpu==2.5.1
-#instal Habana tensorflow bridge & Horovod
-python3 -m pip install habana-tensorflow==1.0.1.81 --extra-index-url https://vault.habana.ai/artifactory/api/pypi/gaudi-python/simple
-python3 -m pip install habana-horovod==1.0.1.81 --extra-index-url https://vault.habana.ai/artifactory/api/pypi/gaudi-python/simple
+${PYTHON} -m pip install --user tensorflow-cpu==2.5.1
+#install Habana tensorflow bridge & Horovod
+${PYTHON} -m pip install --user habana-tensorflow==1.0.1.81 --extra-index-url https://vault.habana.ai/artifactory/api/pypi/gaudi-python/simple
+${PYTHON} -m pip install --user habana-horovod==1.0.1.81 --extra-index-url https://vault.habana.ai/artifactory/api/pypi/gaudi-python/simple
 
 source /etc/profile.d/habanalabs.sh
-python3 -c 'import tensorflow as tf;import habana_frameworks.tensorflow as htf;htf.library_loader.load_habana_module();x = tf.constant(2);y = x + x;assert y.numpy() == 4, "Sanity check failed: Wrong Add output";assert "HPU" in y.device, "Sanity check failed: Operation not executed on Habana";print("Sanity check passed")'
-
+${PYTHON} -c 'import tensorflow as tf;import habana_frameworks.tensorflow as htf;htf.load_habana_module();x = tf.constant(2);y = x + x;assert y.numpy() == 4, "Sanity check failed: Wrong Add output";assert "HPU" in y.device, "Sanity check failed: Operation not executed on Habana";print("Sanity check passed")'