maxmustermann/DeepLearningExamples
1
0
Fork 0
Code Issues Pull requests Projects Releases Wiki Activity
DeepLearningExamples/TensorFlow2/Recommendation/WideAndDeep/README.md

41 KiB
Raw Blame History

Wide & Deep Recommender Model Training For TensorFlow 2

This repository provides a script and recipe to train the Wide and Deep Recommender model to achieve state-of-the-art accuracy. The content of the repository is tested and maintained by NVIDIA.

Model overview

Recommendation systems drive engagement on many of the most popular online platforms. As the volume of data available to power these systems grows exponentially, Data Scientists are increasingly turning from more traditional machine learning methods to highly expressive deep learning models to improve the quality of their recommendations.

Google's Wide & Deep Learning for Recommender Systems has emerged as a popular model for Click Through Rate (CTR) prediction tasks thanks to its power of generalization (deep part) and memorization (wide part). The differences between this Wide & Deep Recommender Model and the model from the paper is the size of the deep part of the model. Originally, in Google's paper, the fully connected part was three layers of 1024, 512, and 256 neurons. Our model consists of 5 layers each of 1024 neurons.

This model is trained with mixed precision using Tensor Cores on NVIDIA Volta and NVIDIA Ampere GPU architectures. Therefore, researchers can get results 4.5 times faster than training without Tensor Cores, while experiencing the benefits of mixed precision training. This model is tested against each NGC monthly container release to ensure consistent accuracy and performance over time.

Model architecture

Wide & Deep refers to a class of networks that use the output of two parts working in parallel - wide model and deep model - to make a binary prediction of CTR. The wide model is a linear model of features together with their transforms. The deep model is a series of 5 hidden MLP layers of 1024 neurons. The model can handle both numerical continuous features as well as categorical features represented as dense embeddings. The architecture of the model is presented in Figure 1.


Figure 1. The architecture of the Wide & Deep model.

Applications and dataset

As a reference dataset, we used a subset of the features engineered by the 19th place finisher in the Kaggle Outbrain Click Prediction Challenge. This competition challenged competitors to predict the likelihood with which a particular ad on a website's display would be clicked on. Competitors were given information about the user, display, document, and ad in order to train their models. More information can be found here.

Default configuration

The Outbrain Dataset is preprocessed in order to get features input to the model. To give context to the acceleration numbers described below, some important properties of our features and model are as follows.

Features:

  • Request Level:

    • 5 scalar numeric features dtype=float32
    • 8 categorical features dtype=int32
    • 8 trainable embeddings of (dimension, cardinality of categorical variable): (128,300000), (19,4), (128,100000), (64,4000), (64,1000), (64,2500), (64,300), (64,2000)
    • 8 trainable embeddings for wide part of size 1 (serving as an embedding from the categorical to scalar space for input to the wide portion of the model)

  • Item Level:

    • 8 scalar numeric features dtype=float32
    • 5 categorical features dtype=int32
    • 5 trainable embeddings of (dimension, cardinality of categorical variable): (128,250000), (64,2500), (64,4000), (64,1000), (128,5000)
    • 5 trainable embeddings for wide part of size 1 (working as trainable one-hot embeddings)

Features describe both the user (Request Level features) and Item (Item Level Features).

  • Model:
    • Input dimension is 26 (13 categorical and 13 numerical features)
    • Total embedding dimension is 1043
    • 5 hidden layers each with size 1024
    • Total number of model parameter is ~90M
    • Output dimension is 1 (y is the probability of click given Request-level and Item-level features)
    • Loss function: Binary Crossentropy

For more information about feature preprocessing, go to Dataset preprocessing.

Model accuracy metric

Model accuracy is defined with the MAP@12 metric. This metric follows the way of assessing model accuracy in the original Kaggle Outbrain Click Prediction Challenge. In this repository, the leaked clicked ads are not taken into account since in industrial setup Data Scientists do not have access to leaked information when training the model. For more information about data leak in Kaggle Outbrain Click Prediction challenge, visit this blogpost by the 19th place finisher in that competition.

Training and evaluation script also reports Loss (BCE) values.

Feature support matrix

The following features are supported by this model:

Feature Wide & Deep
Horovod Multi-GPU (NCCL) Yes
Accelerated Linear Algebra (XLA) Yes
Automatic mixed precision (AMP) Yes

Features

Horovod is a distributed training framework for TensorFlow, Keras, PyTorch and MXNet. The goal of Horovod is to make distributed deep learning fast and easy to use. For more information about how to get started with Horovod, see: Horovod: Official repository.

Multi-GPU training with Horovod Our model uses Horovod to implement efficient multi-GPU training with NCCL. For details, see example sources in this repository or see: TensorFlow tutorial.

XLA is a domain-specific compiler for linear algebra that can accelerate TensorFlow models with potentially no source code changes. Enabling XLA results in improvements to speed and memory usage: most internal benchmarks run ~1.1-1.5x faster after XLA is enabled. For more information on XLA visit official XLA page.

Mixed precision training

Mixed precision is the combined use of different numerical precisions in a computational method. Mixed precision training offers significant computational speedup by performing operations in half-precision format while storing minimal information in single-precision to retain as much information as possible in critical parts of the network. Since the introduction of Tensor Cores in Volta, and following with both the Turing and Ampere architectures, significant training speedups are experienced by switching to mixed precision -- up to 3x overall speedup on the most arithmetically intense model architectures. Using mixed precision training previously required two steps:

  1. Porting the model to use the FP16 data type where appropriate.
  2. Adding loss scaling to preserve small gradient values.

The ability to train deep learning networks with lower precision was introduced in the Pascal architecture and first supported in CUDA 8 in the NVIDIA Deep Learning SDK.

For more information:

For information on the influence of mixed precision training on model accuracy in train and inference, go to Training accuracy results.

Enabling mixed precision

To enable Wide & Deep training to use mixed precision, add the additional flag --amp to the training script. Refer to the Quick Start Guide for more information.

Enabling TF32

TensorFloat-32 (TF32) is the new math mode in NVIDIA A100 GPUs for handling tensor operations. TF32 running on Tensor Cores in A100 GPUs can provide up to 10x speedups compared to single-precision floating-point math (FP32) on Volta GPUs.

TF32 Tensor Cores can speed up networks using FP32, typically with no loss of accuracy. It is more robust than FP16 for models which require high dynamic range for weights or activations.

For more information, refer to the TensorFloat-32 in the A100 GPU Accelerates AI Training, HPC up to 20x blog post.

TF32 is supported in the NVIDIA Ampere GPU architecture and is enabled by default.

Glossary

Request level features Features that describe the person and context to which we wish to make recommendations.

Item level features Features that describe those objects which we are considering recommending.

Setup

The following section lists the requirements that you need to meet in order to start training the Wide & Deep model.

Requirements

This repository contains Dockerfile which extends the TensorFlow2 NGC container and encapsulates some dependencies. Aside from these dependencies, ensure you have the following components:

Supported GPUs:

For more information about how to get started with NGC containers, see the following sections from the NVIDIA GPU Cloud Documentation and the Deep Learning Documentation:

For those unable to use the TensorFlow2 NGC container, to set up the required environment or create their own container, see the versioned NVIDIA Container Support Matrix.

Quick Start Guide

To train your model using the default parameters of the Wide & Deep model on the Outbrain dataset in TF32 or FP32 precision, perform the following steps. For the specifics concerning training and inference with custom settings, see the Advanced section.

  1. Clone the repository.
git clone https://github.com/NVIDIA/DeepLearningExamples
  1. Go to WideAndDeep TensorFlow2 directory within the DeepLearningExamples repository:
cd DeepLearningExamples/TensorFlow2/Recommendation/WideAndDeep
  1. Download the Outbrain dataset.

The Outbrain dataset can be downloaded from Kaggle (requires Kaggle account). Unzip the downloaded archive (for example, to /raid/outbrain/orig) and set the HOST_OUTBRAIN_PATH variable to the parent directory:

HOST_OUTBRAIN_PATH=/raid/outbrain
  1. Build the Wide & Deep Container.
cd DeepLearningExamples/TensorFlow2/Recommendation/WideAndDeep
docker build . -t wd2
  1. Preprocess the Outbrain dataset.

5.1. Start an interactive session in the Wide&Deep Container. Run preprocessing against the original Outbrain dataset to parquets. You can run preprocessing using NVTabular preprocessing (GPU).

docker run --runtime=nvidia --gpus=all --rm -it --ipc=host -v ${HOST_OUTBRAIN_PATH}:/outbrain wd2 bash

5.2. Start NVTabular GPU preprocessing. For more information, go to the Dataset preprocessing section.

bash scripts/preproc.sh

The result of preprocessing script is NVTabular dataset stored in parquets. Files are generated in ${HOST_OUTBRAIN_PATH}/data.

  1. Train the model

6.1. Start an interactive session in the Wide & Deep Container

docker run --runtime=nvidia --gpus=all --rm -it --ipc=host -v ${HOST_OUTBRAIN_PATH}:/outbrain wd2 bash

6.2. Run training (${GPU} is a arbitrary number of gpu to be used)

horovodrun -np ${GPU} sh hvd_wrapper.sh python main.py

Training with Mixed Precision training with XLA:

horovodrun -np ${GPU} sh hvd_wrapper.sh python main.py --xla --amp

For complete usage, run:

python main.py -h
  1. Run validation or evaluation. If you want to run validation or evaluation, you can either:
  • use the checkpoint obtained from the training commands above, or
  • download the pretrained checkpoint from NGC.

In order to download the checkpoint from NGC, visit ngc.nvidia.com website and browse the available models. Download the checkpoint files and unzip them to some path, for example, to $HOST_OUTBRAIN_PATH/checkpoints/ (which is the default path for storing the checkpoints during training). The checkpoint requires around 700MB disk space.

  1. Start validation/evaluation. In order to validate the checkpoint on the evaluation set, run the main.py script with the --evaluate and --use_checkpoint flags.
horovodrun -np ${GPU} sh hvd_wrapper.sh python main.py --evaluate --use_checkpoint

Now that you have your model trained and evaluated, you can choose to compare your training results with our Training accuracy results. You can also choose to benchmark yours performance to Training and evaluation performance benchmark. Following the steps in these sections will ensure that you achieve the same accuracy and performance results as stated in the Results section.

Advanced

The following sections provide greater details of the dataset, running training, and the training results.

Scripts and sample code

These are the important scripts in this repository:

  • main.py - Python script for training the Wide & Deep recommender model.
  • scripts/preproc.sh - Bash script for Outbrain dataset preparation for training, preprocessing and saving into NVTabular format.
  • data/outbrain/dataloader.py - Python file containing NVTabular data loaders for train and evaluation set.
  • data/outbrain/features.py - Python file describing the request and item level features as well as embedding dimensions and hash buckets’ sizes.
  • trainer/model/widedeep.py - Python file with model definition.
  • trainer/run.py - Python file with training and evaluation setup.

Parameters

These are model parameters in the main.py script:

Scope parameter Comment Default Value
location of datasets --train_data_pattern TRAIN_DATA_PATTERN Pattern of training file names /outbrain/data/train/*.parquet
location of datasets --eval_data_pattern EVAL_DATA_PATTERN Pattern of eval file names /outbrain/data/valid/*.parquet
location of datasets --use_checkpoint Use checkpoint stored in model_dir path False
location of datasets --model_dir MODEL_DIR Destination where model checkpoint will be saved /outbrain/checkpoints
location of datasets --results_dir RESULTS_DIR Directory to store training results /results
location of datasets --log_filename LOG_FILENAME Name of the file to store dlloger output log.json
training parameters --global_batch_size GLOBAL_BATCH_SIZE Total size of training batch 131072
training parameters --eval_batch_size EVAL_BATCH_SIZE Total size of evaluation batch 131072
training parameters --num_epochs NUM_EPOCHS Number of training epochs 20
training parameters --cpu Run computations on the CPU Currently not supported
training parameters --amp Enable automatic mixed precision conversion False
training parameters --xla Enable XLA conversion False
training parameters --linear_learning_rate LINEAR_LEARNING_RATE Learning rate for linear model 0.02
training parameters --deep_learning_rate DEEP_LEARNING_RATE Learning rate for deep model 0.00012
training parameters --deep_warmup_epochs DEEP_WARMUP_EPOCHS Number of learning rate warmup epochs for deep model 6
model construction --deep_hidden_units DEEP_HIDDEN_UNITS [DEEP_HIDDEN_UNITS ...] Hidden units per layer for deep model, separated by spaces [1024, 1024, 1024, 1024, 1024]
model construction --deep_dropout DEEP_DROPOUT Dropout regularization for deep model 0.1
run mode parameters --evaluate Only perform an evaluation on the validation dataset, don't train False
run mode parameters --benchmark Run training or evaluation benchmark to collect performance metrics False
run mode parameters --benchmark_warmup_steps BENCHMARK_WARMUP_STEPS Number of warmup steps before start of the benchmark 500
run mode parameters --benchmark_steps BENCHMARK_STEPS Number of steps for performance benchmark 1000
run mode parameters --affinity{socket,single,single_unique,
socket_unique_interleaved,
socket_unique_continuous,disabled}		 Type of CPU affinity socket_unique_interleaved

Command-line options

To see the full list of available options and their descriptions, use the -h or --help command-line option:

python main.py -h

Getting the data

The Outbrain dataset can be downloaded from Kaggle (requires Kaggle account).

Dataset guidelines

The dataset contains a sample of users’ page views and clicks, as observed on multiple publisher sites. Viewed pages and clicked recommendations have additional semantic attributes of the documents. The dataset contains sets of content recommendations served to a specific user in a specific context. Each context (i.e. a set of recommended ads) is given a display_id. In each such recommendation set, the user has clicked on exactly one of the ads.

The original data is stored in several separate files:

  • page_views.csv - log of users visiting documents (2B rows, ~100GB uncompressed)
  • clicks_train.csv - data showing which ad was clicked in each recommendation set (87M rows)
  • clicks_test.csv - used only for the submission in the original Kaggle contest
  • events.csv - metadata about the context of each recommendation set (23M rows)
  • promoted_content.csv - metadata about the ads
  • document_meta.csv, document_topics.csv, document_entities.csv, document_categories.csv - metadata about the documents

During the preprocessing stage, the data is transformed into 87M rows tabular data of 26 features. The dataset is split into training and evaluation parts that have approx 60M and approx 27M rows, respectively. Splitting into train and eval is done in this way so that random 80% of daily events for the first 10 days of the dataset form a training set and remaining part (20% of events daily for the first 10 days and all events in the last two days) form an evaluation set. Eventually the dataset is saved in NVTabular parquet format.

Dataset preprocessing

Dataset preprocessing aims in creating in total 26 features: 13 categorical and 13 numerical. These features are obtained from the original Outbrain dataset in NVTabular preprocessing.

NVTabular GPU preprocessing

The NVTabular dataset is preprocessed using the script provided in data/outbrain/nvtabular. The workflow consists of:

  • separating out the validation set for cross-validation
  • filling missing data with themode, median, or imputed values most frequent value
  • joining click data, ad metadata, and document category, topic and entity tables to create an enriched table.joining the tables for the ad clicks data
  • computing 7 click-through rates (CTR) for ads grouped by 7 features different contexts
  • computing attribute cosine similarity between the landing page and ad to be featured on the page features of the clicked ads and the viewed ads
  • math transformations of the numeric features (logarithmic, normalization)
  • categorifying data using hash-bucketing
  • storing the result in a Parquet format

Most of the code describing operations in this workflow are in data/outbrain/nvtabular/utils/workflow.py and leverage NVTabular v0.6.1. As stated in its repository, NVTabular, a component of NVIDIA Merlin Open Beta, is a feature engineering and preprocessing library for tabular data that is designed to quickly and easily manipulate terabyte scale datasets and train deep learning based recommender systems. It provides a high-level abstraction to simplify code and accelerates computation on the GPU using the RAPIDS Dask-cuDF library. The NVTabular Outbrain workflow has been successfully tested on DGX-1 V100 and DGX A100 for single and multigpu preprocessing.

For more information about NVTabular, refer to the NVTabular documentation.

Training process

The training can be started by running the main.py script. By default, the script is in train mode. Other training related configs are also present in the trainer/utils/arguments.py and can be seen using the command python main.py -h. Training happens with NVTabular data loader on a NVTabular training dataset files that match --train_data_pattern. Training is run for --num_epochs epochs with a global batch size of --global_batch_size in strong scaling mode (i.e. the effective batch size per GPU equals global_batch_size/gpu_count).

The model: tf.keras.experimental.WideDeepModel consists of a wide part and deep part with a sigmoid activation in the output layer (see Figure 1) for reference and trainer/model/widedeep.py for model definition).

During training (default configuration): Two separate optimizers are used to optimize the wide and the deep part of the network:

  • FTLR (Follow the Regularized Leader) optimizer is used to optimize the wide part of the network.
  • RMSProp optimizer is used to optimize the deep part of the network.

Checkpoint of the model:

  • Can be loaded at the beginning of training when --use_checkpoint is set.
  • is saved into --model_dir after each training epoch. Only the last checkpoint is kept.
  • Contains information about number of training epochs.

The model is evaluated on an evaluation dataset after every training epoch training log is displayed in the console and stored in --log_filename.

Every 100 batches with training metrics: bce loss

Every epoch after training, evaluation metrics are logged: bce loss and MAP@12 value

Evaluation process

The evaluation can be started by running the main.py --evaluation script. Evaluation is done on NVTabular parquet dataset stored in --eval_data_pattern. Other evaluation related configs are also present in the trainer/utils/arguments.py and can be seen using the command python main.py -h.

During evaluation (--evaluation flag):

  • Model is restored from checkpoint in --model_dir if --use_checkpoint is set.
  • Evaluation log is displayed in console and stored in --log_filename.
  • Every 100 batches evaluation metrics are logged: bce loss.

After the whole evaluation, the total evaluation metrics are logged: bce loss and MAP@12 value.

Performance

Benchmarking

The following section shows how to run benchmarks measuring the model performance in training mode.

Training and evaluation performance benchmark

Benchmark script is prepared to measure performance of the model during training (default configuration) and evaluation (--evaluation). Benchmark runs training or evaluation for --benchmark_steps batches, however measurement of performance starts after --benchmark_warmup_steps. Benchmark can be run for single and 8 GPUs and with a combination of XLA (--xla), AMP (--amp), batch sizes (--global_batch_size , --eval_batch_size) and affinity (--affinity).

In order to run benchmark follow these steps:

Run Wide & Deep Container (${HOST_OUTBRAIN_PATH} is the path with Outbrain dataset):

docker run --runtime=nvidia --gpus=all --rm -it --ipc=host -v ${HOST_OUTBRAIN_PATH}:/outbrain wd2 bash

Run the benchmark script:

horovodrun -np ${GPU} sh hvd_wrapper.sh python main.py --benchmark

Results

The following sections provide details on how we achieved our performance and accuracy in training.

Training accuracy results

Training accuracy: NVIDIA DGX A100 (8x A100 80GB)

Our results were obtained by running the main.py training script in the TensorFlow2 NGC container on NVIDIA DGX A100 with (8x A100 80GB) GPUs.

GPUs Batch size / GPU XLA Accuracy - TF32 (MAP@12) Accuracy - mixed precision (MAP@12) Time to train - TF32 (minutes) Time to train - mixed precision (minutes) Time to train speedup (TF32 to mixed precision)
1 131072 Yes 0.65656 0.65654 13.40 9.48 1.41
1 131072 No 0.65662 0.65656 17.75 13.38 1.33
8 16384 Yes 0.65672 0.65665 4.82 4.50 1.07
8 16384 No 0.65671 0.65655 5.71 5.72 1.00

To achieve the same results, follow the steps in the Quick Start Guide.

Training accuracy: NVIDIA DGX-1 (8x V100 32GB)

Our results were obtained by running the main.py training script in the TensorFlow2 NGC container on NVIDIA DGX-1 with (8x V100 32GB) GPUs.

GPUs Batch size / GPU XLA Accuracy - FP32 (MAP@12) Accuracy - mixed precision (MAP@12) Time to train - FP32 (minutes) Time to train - mixed precision (minutes) Time to train speedup (FP32 to mixed precision)
1 131072 Yes 0.65658 0.65664 62.89 18.65 3.37
1 131072 No 0.65662 0.65658 71.53 25.18 2.84
8 16384 Yes 0.65668 0.65655 12.21 8.89 1.37
8 16384 No 0.65665 0.65654 14.38 7.17 2.01

To achieve the same results, follow the steps in the Quick Start Guide.

Training accuracy plots

Models trained with FP32, TF32 and Automatic Mixed Precision (AMP), with and without XLA enabled achieve similar accuracy.

The plot represents MAP@12 in a function of steps (step is single batch) during training for default precision (FP32 for Volta architecture (DGX-1) and TF32 for Ampere GPU architecture (DGX-A100)) and AMP for XLA and without it for NVTabular dataset. All other parameters of training are default.


Figure 2. Learning curves for different configurations on single gpu.

Training stability test

Training of the model is stable for multiple configurations achieving the standard deviation of 10e-4. The model achieves similar MAP@12 scores for A100 and V100, training precisions, XLA usage and single/multi GPU. The Wide and Deep model was trained for 9140 training steps (20 epochs, 457 batches in each epoch, every batch containing 131072), starting from 20 different initial random seeds for each setup. The training was performed in the 21.09 Merlin Tensorflow Training NGC container on NVIDIA DGX A100 80GB and DGX-1 32GB machines with and without mixed precision enabled, with and without XLA enabled for NVTabular generated dataset. The provided charts and numbers consider single and 8 GPU training. After training, the models were evaluated on the validation set. The following plots compare distributions of MAP@12 on the evaluation set. In columns there is single vs 8 GPU training, in rows DGX A100 and DGX-1 V100.


Figure 3. Training stability plot, distribution of MAP@12 across different configurations. 'All configurations' refer to the distribution of MAP@12 for cartesian product of architecture, training precision, XLA usage, single/multi GPU.

Training stability was also compared in terms of point statistics for MAP@12 distribution for multiple configurations. Refer to the expandable table below.

Full tabular data for training stability tests
GPUs Precicision XLA Mean Std Min Max
DGX A100 1 TF32 Yes 0.65656 0.00016 0.6563 0.6569
DGX A100 1 TF32 No 0.65662 0.00013 0.6563 0.6568
DGX A100 1 AMP Yes 0.65654 0.00010 0.6563 0.6567
DGX A100 1 AMP No 0.65656 0.00011 0.6564 0.6568
DGX A100 8 TF32 Yes 0.65672 0.00012 0.6565 0.6570
DGX A100 8 TF32 No 0.65671 0.00013 0.6565 0.6569
DGX A100 8 AMP Yes 0.65665 0.00014 0.6564 0.6569
DGX A100 8 AMP No 0.65655 0.00012 0.6564 0.6568
DGX-1 V100 1 FP32 Yes 0.65658 0.00013 0.6563 0.6568
DGX-1 V100 1 FP32 No 0.65662 0.00011 0.6564 0.6568
DGX-1 V100 1 AMP Yes 0.65664 0.00011 0.6564 0.6568
DGX-1 V100 1 AMP No 0.65658 0.00011 0.6564 0.6568
DGX-1 V100 8 FP32 Yes 0.65668 0.00016 0.6564 0.6570
DGX-1 V100 8 FP32 No 0.65665 0.00019 0.6564 0.6570
DGX-1 V100 8 AMP Yes 0.65655 0.00012 0.6563 0.6567
DGX-1 V100 8 AMP No 0.65654 0.00013 0.6563 0.6567
Impact of mixed precision on training accuracy

The accuracy of training, measured with MAP@12 on the evaluation set after the final epoch metric was not impacted by enabling mixed precision. The obtained results were statistically similar. The similarity was measured according to the following procedure:

The model was trained 20 times for default settings (FP32 or TF32 for Volta and Ampere architecture respectively) and 20 times for AMP. After the last epoch, the accuracy score MAP@12 was calculated on the evaluation set.

Distributions for four configurations: architecture (A100, V100) and single/multi gpu for NVTabular dataset are presented below.


Figure 4. Influence of AMP on MAP@12 distribution for DGX A100 and DGX-1 V100 for single and multi gpu training.

Distribution scores for full precision training and AMP training were compared in terms of mean, variance and Kolmogorov–Smirnov test to state statistical difference between full precision and AMP results. Refer to the expandable table below.

Full tabular data for AMP influence on MAP@12
GPUs XLA Mean MAP@12 for Full precision (TF32 for A100, FP32 for V100) Std MAP@12 for Full precision (TF32 for A100, FP32 for V100) Mean MAP@12 for AMP Std MAP@12 for AMP KS test value: statistics, p-value
DGX A100 1 Yes 0.65656 0.00016 0.65654 0.00010 0.10000 (0.99999)
DGX A100 8 Yes 0.65672 0.00012 0.65665 0.00014 0.40000 (0.08106)
DGX A100 1 No 0.65662 0.00013 0.65656 0.00011 0.35000 (0.17453)
DGX A100 8 No 0.65671 0.00013 0.65655 0.00012 0.35000 (0.17453)
DGX-1 V100 1 Yes 0.65658 0.00013 0.65664 0.00011 0.25000 (0.57134)
DGX-1 V100 8 Yes 0.65668 0.00016 0.65655 0.00012 0.30000 (0.33559)
DGX-1 V100 1 No 0.65662 0.00011 0.65658 0.00011 0.20000 (0.83197)
DGX-1 V100 8 No 0.65665 0.00019 0.65654 0.00013 0.40000 (0.08106)

Training performance results

Training performance: NVIDIA DGX A100 (8x A100 80GB)

Our results were obtained by running the benchmark script (main.py --benchmark) in the TensorFlow2 NGC container on NVIDIA DGX A100 with (8x A100 80GB) GPUs.

GPUs Batch size / GPU XLA Throughput - TF32 (samples/s) Throughput - mixed precision (samples/s) Throughput speedup (TF32 - mixed precision) Strong scaling - TF32 Strong scaling - mixed precision
1 131,072 Yes 2026524 3069487 1.51 1.00 1.00
1 131,072 No 1379960 1928375 1.40 1.00 1.00
8 16,384 Yes 6892010 7574174 1.10 3.40 2.47
8 16,384 No 5124054 5120040 1.00 3.71 2.66
Training performance: NVIDIA DGX-1 (8x V100 32GB)

Our results were obtained by running the benchmark script (main.py --benchmark) in the TensorFlow2 NGC container on NVIDIA DGX-1 with (8x V100 32GB) GPUs.

GPUs Batch size / GPU XLA Throughput - FP32 (samples/s) Throughput - mixed precision (samples/s) Throughput speedup (FP32 - mixed precision) Strong scaling - FP32 Strong scaling - mixed precision
1 131,072 Yes 378918 1405633 3.71 1.00 1.00
1 131,072 No 323817 969824 2.99 1.00 1.00
8 16,384 Yes 2196648 4332939 1.97 5.80 3.08
8 16,384 No 1772485 3058944 1.73 5.47 3.15

Evaluation performance results

Evaluation performance: NVIDIA DGX A100 (8x A100 80GB)

Our results were obtained by running the benchmark script (main.py --evaluate --benchmark) in the TensorFlow2 NGC container on NVIDIA DGX A100 with 8x A100 80GB GPUs.

GPUs Batch size / GPU XLA Throughput [samples/s] TF32 Throughput [samples/s] AMP Throughput speedup AMP to TF32
1 4096 NO 1107650 1028782 0.93
1 8192 NO 1783848 1856528 1.04
1 16384 NO 2295874 2409601 1.05
1 32768 NO 2367142 2583293 1.09
1 65536 NO 3044662 3471619 1.14
1 131072 NO 3229625 3823612 1.18
8 4096 NO 5503985 5333228 0.97
8 8192 NO 12251675 12386870 1.01
8 16384 NO 16020973 16438269 1.03
8 32768 NO 17225168 18667798 1.08
8 65536 NO 19969248 22270424 1.12
8 131072 NO 19929457 22496045 1.13

For more results go to the expandable table below.

Full tabular data for evaluation performance results for DGX A100
GPUs Batch size / GPU XLA Throughput [samples/s] TF32 Throughput [samples/s] AMP Throughput speedup AMP to TF32
1 4096 YES 1344225 1501677 1.12
1 4096 NO 1107650 1028782 0.93
1 8192 YES 2220721 2545781 1.15
1 8192 NO 1783848 1856528 1.04
1 16384 YES 2730441 3230949 1.18
1 16384 NO 2295874 2409601 1.05
1 32768 YES 2527368 2974417 1.18
1 32768 NO 2367142 2583293 1.09
1 65536 YES 3163906 3935731 1.24
1 65536 NO 3044662 3471619 1.14
1 131072 YES 3171670 4064426 1.28
1 131072 NO 3229625 3823612 1.18
8 4096 YES 6243348 6553485 1.05
8 4096 NO 5503985 5333228 0.97
8 8192 YES 14995914 16222429 1.08
8 8192 NO 12251675 12386870 1.01
8 16384 YES 14584474 16543902 1.13
8 16384 NO 16020973 16438269 1.03
8 32768 YES 17840220 21537660 1.21
8 32768 NO 17225168 18667798 1.08
8 65536 YES 20732672 24082577 1.16
8 65536 NO 19969248 22270424 1.12
8 131072 YES 20104010 24157900 1.20
8 131072 NO 19929457 22496045 1.13
Evaluation performance: NVIDIA DGX-1 (8x V100 32GB)

Our results were obtained by running the benchmark script (main.py --evaluate --benchmark) in the TensorFlow2 NGC container on NVIDIA DGX-1 with (8x V100 32GB) GPUs.

GPUs Batch size / GPU XLA Throughput [samples/s] FP32 Throughput [samples/s] AMP Throughput speedup AMP to FP32
1 4096 NO 499442 718163 1.44
1 8192 NO 670906 1144640 1.71
1 16384 NO 802366 1599006 1.99
1 32768 NO 856130 1795285 2.10
1 65536 NO 934394 2221221 2.38
1 131072 NO 965293 2403829 2.49
8 4096 NO 2840155 3602516 1.27
8 8192 NO 4810100 7912019 1.64
8 16384 NO 5939908 10876135 1.83
8 32768 NO 6489446 12593087 1.94
8 65536 NO 6614453 14742844 2.23
8 131072 NO 7133219 15524549 2.18

For more results go to the expandable table below.

Full tabular data for evaluation performance for DGX-1 V100 results
GPUs Batch size / GPU XLA Throughput [samples/s] FP32 Throughput [samples/s] AMP Throughput speedup AMP to FP32
1 4096 YES 573285 919150 1.60
1 4096 NO 499442 718163 1.44
1 8192 YES 753993 1486867 1.97
1 8192 NO 670906 1144640 1.71
1 16384 YES 859699 1945700 2.26
1 16384 NO 802366 1599006 1.99
1 32768 YES 904255 1995194 2.21
1 32768 NO 856130 1795285 2.10
1 65536 YES 982448 2608010 2.65
1 65536 NO 934394 2221221 2.38
1 131072 YES 926734 2621095 2.83
1 131072 NO 965293 2403829 2.49
8 4096 YES 3102948 4083015 1.32
8 4096 NO 2840155 3602516 1.27
8 8192 YES 5536556 10094905 1.82
8 8192 NO 4810100 7912019 1.64
8 16384 YES 5722386 10524548 1.84
8 16384 NO 5939908 10876135 1.83
8 32768 YES 6813318 14356608 2.11
8 32768 NO 6489446 12593087 1.94
8 65536 YES 6918413 16227668 2.35
8 65536 NO 6614453 14742844 2.23
8 131072 YES 6910518 16423342 2.38
8 131072 NO 7133219 15524549 2.18

Release notes

Changelog

February 2021

  • Initial release

November 2021

  • Refresh release with performance optimizations
  • Updated NVTabular to v0.6.1
  • Replaced native TF dataloader with NVTabular counterpart
  • Removed spark CPU preprocessing
  • Updated readme numbers
  • Changed V100 cards from 16GB to 32GB

Known issues

  • In this model the TF32 precision can in some cases be as fast as the FP16 precision on Ampere GPUs. This is because TF32 also uses Tensor Cores and doesn't need any additional logic such as maintaining FP32 master weights and casts. However, please note that W&D is, by modern recommender standards, a very small model. Larger models should still see significant benefits of using FP16 math.