Getting started with TensorFlow

Find information for getting started with TensorFlow.

The following TensorFlow related conda packages are included with WML CE:

Table 1. TensorFlow Conda Packages
Name Description Installed with powerai Installed with tensorflow-gpu Installed with powerai-cpu Installed with tensorflow
tensorflow TensorFlow Meta package     ✓ ✓
tensorflow-gpu TensorFlow GPU Meta package ✓ ✓    
tensorflow-base Contains the core TensorFlow Logic ✓ ✓ ✓ ✓
tensorflow-estimator Required TensorFlow Estimator package ✓ ✓ ✓ ✓
tensorboard Visualization Dashboard for TensorFlow ✓ ✓ ✓ ✓
tensorflow-large-model-support TensorFlow Large Model Support ✓      
tensorflow-probability Optional TensorFlow Probability package ✓   ✓  
tf_cnn_benchmarks Optional TensorFlow CNN benchmark package ✓      
ddl-tensorflow Distributed Deep Learning custom operator for TensorFlow ✓      
bazel Fast, scalable, multi-language and extensible build system        
tensorflow-serving-api Serving system for machine learning models ✓   ✓  
tensorflow-serving Serving system for machine learning models        
tensorrt C++ library running pre-trained networks quickly and efficiently ✓ ✓    

You can install packages with conda install <package name>.

The TensorFlow home page has various information, including Tutorials, How to documents, and a Getting Started guide.

Additional tutorials and examples are available from the community, for example:

tensorflow and tensorflow-gpu conda packages

WML CE includes a version of TensorFlow built without GPU support. This inclusion allows for training and inferencing to be done on POWER8 and POWER9 systems that do not have GPUs, or on systems where you want to train and inference without using the GPUs.
  • To install TensorFlow built for CPU support run the following command. Only use the strict-channel-priority setting if you are installing into an x86 environment:
    conda install --strict-channel-priority tensorflow
    This command installs the TensorFlow package, with no packages for GPU support.
    Note: In the prior release, the tensorflow conda package installed the TensorFlow package built for GPU support as that was the only support available in that release.
  • To install TensorFlow built for GPU support run the following command. Only use the strict-channel-priority setting if you are installing into an x86 environment:
    conda install --strict-channel-priority tensorflow-gpu
    This command installs TensorFlow along with the CUDA, cuDNN, and NCCL conda packages used with the GPUs.

TensorFlow Large Model Support (TFLMS)

Large Model Support provides an approach to training large models and batch sizes that cannot fit in GPU memory. It does this by use of a graph editing library that takes the user model's computational graph and automatically adds swap-in and swap-out nodes for transferring tensors from GPU memory to system memory and vice versa during training.

For more information about TensorFlow LMS, see Getting started with TensorFlow large model support.

Distributed Deep Learning (DDL) custom operator for TensorFlow

The DDL custom operator uses IBM Spectrum™ MPI and NCCL to provide high-speed communications for distributed TensorFlow.

The DDL custom operator can be found in the ddl-tensorflow package. For more information about DDL and about the TensorFlow operator, see Integration with deep learning frameworks

TensorFlow with NVIDIA TensorRT (TF-TRT)

NVIDIA TensorRT is a plaform for high-performance deep learning inference. Trained models can be optimized with TensorRT; this is done by replacing TensorRT-compatible subgraphs with a single TRTEngineOp that is used to build a TensorRT engine. TensorRT can also calibrate for lower precision (FP16 and INT8) with a minimal loss of accuracy. After a model is optimized with TensorRT, the TensorFlow workflow is still used for inferencing, including TensorFlow-Serving.

A saved model can be optimized for TensorRT with the following python snippet:

from tensorflow.python.compiler.tensorrt import trt_convert as trt
converter = trt.TrtGraphConverter(
    input_saved_model_dir=input_saved_model_dir)
converter.convert()
converter.save(output_saved_model_dir)

TensorRT is enabled in the tensorflow-gpu and tensorflow-serving packages.

For additional information on TF-TRT, see the official Nvidia docs.

Additional TensorFlow features

The powerai TensorFlow packages include TensorBoard. For more information, see Getting started with TensorBoard.

The TensorFlow package includes support for additional features:

TensorFlow CNN benchmarks

WML CE includes the tf_cnn_benchmarks package that contains a version of the TensorFlow CNN benchmark. This package contains implementations of several popular convolutional models, and is designed to be as fast as possible. The code supports both running on a single machine or running in distributed mode across multiple hosts. This version of tf_cnn_benchmarks is matched to the version of TensorFlow that is included in this release and has additional options to enable DDL and TensorFlow Large Model Support.

TensorFlow 2.0 Technology Preview

WML CE includes a technology preview of TensorFlow 2.0 compiled with GPU support. The focus of TensorFlow 2.0 is simplicity and ease of use. TensorFlow 2.0 leverages Keras as the high-level API for TensorFlow. The TensorFlow 2.0 home page contains examples written for 2.0 and information about migrating 1.X code to 2.0 along with getting started guides for beginners and experts.

The package name is tensorflow2-gpu and it must be installed in a separate conda environment than TensorFlow 1.x.

The following examples are installation commands:

conda create --name tensorflow2-tech-preview python=3.6 tensorflow2-gpu
conda activate tensorflow2-tech-preview

TensorFlow 2.0 has been tested with TensorBoard and TensorFlow Estimator. As the TensorFlow Estimator conda package is dependent on the TensorFlow conda package, it must be installed with the --no-deps flag to avoid TensorFlow 1.X getting installed when estimator is installed. To install TensorFlow Estimator run:

conda install tensorflow-estimator --no-deps

TensorFlow 2.0 is compiled with TensorRT support, however the examples in the tensorrt-samples conda package are not compatible with TensorFlow 2.0. TensorRT has not been tested with TensorFlow 2.0

TensorFlow 2.0 has not been tested with TensorFlow Large Model Support, TensorFlow Serving, TensorFlow Probability or tf_cnn_benchmarks at this time.

Tensorflow Keras-team Keras

TensorFlow includes an implementation of the Keras API (in the tf.keras module) with TensorFlow-specific enhancements. These include support for eager execution for intuitive debugging and fast iteration, support for the TensorFlow SavedModel model exchange format, and integrated support for distributed training. Keras-team Keras should not be used with TensorFlow 2.0, to use keras:

import tensorflow.keras as keras

Automatic mixed precision support

TensorFlow 1.14 includes a feature called Automatic Mixed Precision (AMP) that automatically takes advantage of lower precision hardware such at the Tensor Cores included in NVIDIA's V100 GPUs. AMP can speed up training in certain models. To enable AMP, add the following lines of Python to the model code:

config = tf.ConfigProto()
config.graph_options.rewrite_options.auto_mixed_precision=1