TensorFlow Release 19.04

The NVIDIA container image of TensorFlow, release 19.04, is available on NGC.

Contents of the TensorFlow container

This container image contains the complete source of the version of NVIDIA TensorFlow in /opt/tensorflow. It is pre-built and installed as a system Python module.

To achieve optimum TensorFlow performance, for image based training, the container includes a sample script that demonstrates efficient training of convolutional neural networks (CNNs). The sample script may need to be modified to fit your application.

Driver Requirements

Release 19.04 is based on CUDA 10.1, which requires NVIDIA Driver release 418.xx.x+. However, if you are running on Tesla (Tesla V100, Tesla P4, Tesla P40, or Tesla P100), you may use NVIDIA driver release 384.111+ or 410. The CUDA driver's compatibility package only supports particular drivers. For a complete list of supported drivers, see the CUDA Application Compatibility topic. For more information, see CUDA Compatibility and Upgrades.

GPU Requirements

Release 19.04 supports CUDA compute capability 6.0 and higher. This corresponds to GPUs in the Pascal, Volta, and Turing families. Specifically, for a list of GPUs that this compute capability corresponds to, see CUDA GPUs. For additional support details, see Deep Learning Frameworks Support Matrix.

Key Features and Enhancements

This TensorFlow release includes the following key features and enhancements.

Accelerating Inference In TensorFlow With TensorRT (TF-TRT)

For step-by-step instructions on how to use TF-TRT, see Accelerating Inference In TensorFlow With TensorRT User Guide. To view the key features, software enhancements and improvements, and known issues, see the Release Notes.
Note: The TF-TRT examples were relocated from /workspace/nvidia-examples/tftrt/scripts/ to workspace/nvidia-examples/inference.

Automatic Mixed Precision (AMP)

Automatic mixed precision converts certain float32 operations to operate in float16 which can run much faster on Tensor Cores. Automatic mixed precision is built on two components:
  • a loss scaling optimizer
  • graph rewriter
For models already using a tf.Optimizer() for both compute_gradients() and apply_gradients() operations, automatic mixed precision can be enabled by defining the following environment variable before calling the usual float32 training script:
export TF_ENABLE_AUTO_MIXED_PRECISION=1
Models implementing their own optimizers can use the graph rewriter on its own (while implementing loss scaling manually) with the following environment variable:
export TF_ENABLE_AUTO_MIXED_PRECISION_GRAPH_REWRITE=1

For more information about how to access and enable Automatic mixed precision for TensorFlow, see Automatic Mixed Precision Training In TensorFlow from the TensorFlow User Guide, along with Training With Mixed Precision.

Tensor Core Examples

These examples focus on achieving the best performance and convergence from NVIDIA Volta Tensor Cores by using the latest deep learning example networks for training.

Each example model trains with mixed precision Tensor Cores on Volta, therefore you can get results much faster than training without tensor cores. This model is tested against each NGC monthly container release to ensure consistent accuracy and performance over time. This container includes the following tensor core examples.
  • An implementation of the SSD320 v1.2 model. The SSD320 v1.2 model is based on the SSD: Single Shot MultiBox Detector paper, which describes SSD as “a method for detecting objects in images using a single deep neural network”. Our implementation is based on the existing model from the TensorFlow models repository.

  • An implementation of the Neural Collaborative Filtering (NCF) model. The NCF model is a neural network that provides collaborative filtering based on implicit feedback, specifically, it provides product recommendations based on user and item interactions. The training data for this model should contain a sequence of user ID, item ID pairs indicating that the specified user has interacted with, for example, was given a rating to or clicked on, the specified item.

  • An implementation of the Bert model. BERT, or Bidirectional Encoder Representations from Transformers, is a new method of pre-training language representations which obtains state-of-the-art results on a wide array of Natural Language Processing (NLP) tasks. This model is based on BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding paper. NVIDIA's BERT is an optimized version of Google's official implementation, leveraging mixed precision arithmetic and Tensor Cores on V100 GPUS for faster training times while maintaining target accuracy.

  • An implementation of the U-Net Industrial Defect Segmentation model. This U-Net model is adapted from the original version of the U-Net model which is a convolutional auto-encoder for 2D image segmentation. U-Net was first introduced by Olaf Ronneberger, Philip Fischer, and Thomas Brox in the paper: U-Net: Convolutional Networks for Biomedical Image Segmentation. This work proposes a modified version of U-Net, called TinyUNet which performs efficiently and with very high accuracy on the industrial anomaly dataset DAGM2007.

  • An implementation of the GNMT v2 model. The GNMT v2 model is similar to the one discussed in the Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation paper. The most important difference between the two models is in the attention mechanism. In our model, the output from the first LSTM layer of the decoder goes into the attention module, then the re-weighted context is concatenated with inputs to all subsequent LSTM layers in the decoder at the current timestep.

  • An implementation of the ResNet-50 v1.5 model. The ResNet-50 v1.5 model is a modified version of the original ResNet-50 v1 model. The difference between v1 and v1.5 is in the bottleneck blocks which requires downsampling, for example, v1 has stride = 2 in the first 1x1 convolution, whereas v1.5 has stride = 2 in the 3x3 convolution. The following features were implemented in this model; data-parallel multi-GPU training with Horovod, tensor cores (mixed precision) training, and static loss scaling for Tensor Cores (mixed precision) training.

Known Issues

  • There is a known performance regression with TensorFlow 1.13.1 for some networks when run with small batch sizes. As a workaround, increase the batch size.
  • The AMP preview implementation is not compatible with Distributed Strategies. We recommend using Horovod for parallel training with AMP.
  • A known issue in TensorFlow results in the error Cannot take the length of Shape with unknown rank when training variable sized images with the Keras model.fit API. Details are provided here and a fix will be available in a future release.