[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10,"_11":12},"actionData","errors","root",{"_87":256,"_257":258,"_259":260,"_261":262,"_263":264,"_265":30,"_266":267,"_268":269,"_270":30,"_271":30,"_272":30,"_273":30,"_274":275,"_276":-5,"_277":278,"_279":280,"_281":282,"_283":114,"_284":116,"_285":286,"_108":-5,"_287":30,"_288":-5,"_289":290,"_291":292,"_138":293,"_294":295,"_296":297,"_298":299,"_300":301,"_302":30,"_25":26,"_27":303,"_29":30,"_31":304,"_33":305,"_35":306},"routes/orgs/$orgName/$teamName/$resourceType/$resourceId/$version/route",{"_123":124,"_125":126,"_127":128,"_129":130,"_131":30,"_132":133,"_134":30,"_135":30,"_136":137,"_138":21,"_139":-5,"_140":141,"_142":-5,"_143":144,"_145":146,"_147":30,"_148":149,"_25":26,"_27":150,"_29":30,"_31":151,"_33":152,"_35":153},"routes/orgs/$orgName/$teamName/$resourceType/$resourceId/$version/_access/route",{"_84":85,"_86":-5,"_87":88,"_89":90,"_25":26,"_27":91,"_29":30,"_31":92,"_33":93,"_35":94},"routes/orgs/$orgName/$teamName/$resourceType/$resourceId/$version/file-browser/route",{"_13":14,"_15":16,"_17":16,"_18":19,"_20":21,"_22":21,"_23":24,"_25":26,"_27":28,"_29":30,"_31":32,"_33":34,"_35":36},"fileTree",[71],"totalFiles",1,"page","pageSize",50,"unresolvedVersion","22.01.0","resolvedVersion","expandedFolders",[],"traceId","02d088b82267b0f1f39d9fa68cd100ba","traceStartTime",1787757155047,"isStaleResponse",false,"notifications",[],"toasts",[],"apiCalls",[37,38,39,40,41,42,43],{"_44":59,"_46":69,"_48":61,"_50":62,"_52":70,"_54":55},{"_44":59,"_46":68,"_48":61,"_50":62,"_52":19,"_54":55},{"_44":59,"_46":66,"_48":61,"_50":62,"_52":67,"_54":55},{"_44":59,"_46":64,"_48":61,"_50":62,"_52":65,"_54":55},{"_44":59,"_46":60,"_48":61,"_50":62,"_52":63,"_54":55},{"_44":56,"_46":57,"_48":49,"_50":51,"_52":58,"_54":55},{"_44":45,"_46":47,"_48":49,"_50":51,"_52":53,"_54":55},"apiTag","subscription-api","path","/v2/users/me/subscriptions","status","error","httpStatus",401,"durationMs",8,"retryCount",0,"oms-api","/v2/users/me",3,"models-api","/v2/resources/org/nvidia/team/dle/sim_tf2/collections?page-number=0&page-size=1","success",200,54,"/v2/resources/nvidia/dle/sim_tf2/versions/22.01.0/files?page-size=50&page-number=0",61,"/v2/resources/nvidia/dle/sim_tf2/versions/22.01.0",60,"/v2/resources/nvidia/dle/sim_tf2/versions?page-size=100&page-number=0&sort-order=CREATED_DATE_DESC","/v2/resources/nvidia/dle/sim_tf2?resolve-labels=true&remove-unresolved-labels=true",51,{"_72":73,"_73":74},"type","file",{"_75":76,"_77":76,"_46":76,"_78":79,"_80":81,"_82":83},"id","sim_tf2.md","name","size",83524,"formattedSize","81.57 KB","updatedAt","2022-09-22T14:41:56.297Z","canDownload",true,"accessDeniedReason","appConfig",{"_106":107,"_108":109},"debugAccessInfo",{"_102":85,"_103":104,"_105":30},1787757154976,[],[],[95,96,97,98,99,100],{"_44":59,"_46":69,"_48":61,"_50":62,"_52":70,"_54":55},{"_44":59,"_46":68,"_48":61,"_50":62,"_52":19,"_54":55},{"_44":59,"_46":66,"_48":101,"_50":-5,"_52":-5,"_54":55},{"_44":59,"_46":60,"_48":61,"_50":62,"_52":63,"_54":55},{"_44":56,"_46":57,"_48":101,"_50":-5,"_52":-5,"_54":55},{"_44":45,"_46":47,"_48":49,"_50":51,"_52":53,"_54":55},"pending","canGuestAccess","requiredProducts",[],"hasLicenseRequirement","urls",{"_111":112,"_113":114,"_115":116,"_117":118,"_119":120,"_121":122},"user",{"_110":30},"loggedIn","build","https://build.nvidia.com","ngcApi","https://api.ngc.nvidia.com","ngcHelm","https://helm.ngc.nvidia.com","ngcRegistry","https://registry.ngc.nvidia.com","orgUi","https://org.ngc.nvidia.com","ngcUi","https://ngc.nvidia.com","resource",{"_75":203,"_204":205,"_206":207,"_208":209,"_161":162,"_163":164,"_210":211,"_167":212,"_213":214,"_172":215,"_216":217,"_218":199,"_219":-7,"_220":-7,"_221":222,"_223":224,"_225":85,"_226":227,"_228":-5,"_229":-7,"_230":-7,"_231":-7,"_232":-7,"_233":-7,"_234":-7,"_235":236,"_237":238,"_239":240,"_241":242,"_243":244,"_245":30},"html","

This resource is using open-source code maintained in github (see the quick-start-guide section) and available for download from NGC

Search-based Interest Model (SIM) is a system for predicting user behavior given sequences of previous interactions. The model is based on Search-based User Interest Modeling with Lifelong Sequential Behavior Data for Click-Through Rate Prediction paper which reports that it has been deployed at Alibaba in the display advertising system. This repository provides a reimplementation of the code-base provided originally for SIM and DIEN models (SIM model's inner component).

There are several differences between this and the original SIM model implementation. First, this model is implemented in TensorFlow 2 using Python 3.8 instead of TensorFlow 1 in Python 2.7. Second, this implementation utilizes the user dimension (identifiers), which enables to train a personalized recommender system. Finally, the training code uses data preprocessed to TFRecord format, which improves data loading. We also include scripts necessary to preprocess Amazon Reviews dataset used in experiments.

The table below provides a fine-grained summary of the differences between this repository and the original implementation.

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n

tr:nth-child(odd)]:bg-surface-sunken\">

Mode	Original implementation	This repository
Python	2.7	3.8
Dataset size	135K samples	12M samples
Dataset format	CSV	TFRecord
Model	- user id feature not included - batch normalization included but not used correctly - two-dimensional softmax output - hardcoded features cardinalities	- includes user id feature - doesn`t include batch normalization - one-dimensional sigmoid output - features cardinalities deducted from dataset

In the author's SIM implementation, the internals of submodels differs slightly between code and original papers (DIN, DIEN, SIM). Our implementation core is based on the paper's modules. For exact implementation details, refer to the list below.

\nList of implementation differences between original SIM code and DIN/DIEN/SIM papers\n

Batch normalization before MLP is not included in papers.
Batch normalization in code used trainable=False during the training phase.
ItemItemInteraction in DIN`s attention module in SIM implementation didn't correspond to activation unit inside DIN paper.\n
- Element-wise subtraction and multiplications are fed to MLP, skipping outer product operation.
- Sigmoids are used instead of PReLU/DICE in MLP.
\n
Soft search MLP is missing a middle layer in implementation.
In the ESU part, multi-head attention is implemented as a DIN interaction block instead of a typical multi-head attention.
ESU part adds additional embedding by summing all the embedding passed from the GSU part.
DIEN auxiliary loss uses auxiliary network instead of the sigmoid of concatenated embeddings from the DIEN paper.

The model enables you to train a high-quality, personalized, sequential neural network-based recommender system.

This model is trained with mixed precision using Tensor Cores on NVIDIA Volta and the NVIDIA Ampere GPU architectures. Therefore, researchers can get results 1.48x 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

SIM model consists of two components: General Search Unit (GSU) and the Exact Search Unit (ESU). The goal of the former is to filter down possibly long historical user behavior sequence to a shorter and relevant sequence. On the other hand, ESU utilizes the most recent user behaviors for a candidate item, for example, estimate click-through rate for a candidate ad. Both parts are trained jointly using data on past user behaviors.

A model architecture diagram is presented below.

\n \n
\nFigure 1. The architecture of the model.\n

Embeddings in model architecture diagram are obtained by passing each feature from the dataset through the Embedding Layer. Item features from target item, short behavior history and long behavior history share embedding tables.

\n \n
\nFigure 2. Embedding of input features.\n

Default configuration

The following features are implemented in this model:

general\n
- dynamic loss scaling for Tensor Cores (mixed precision) training
- data-parallel multi-GPU training
\n
preprocessing\n
- dataset preprocessing using NVtabular library
\n

The following performance optimizations were implemented in this model:

dataloader based on TFRecords

Feature support matrix

This model supports the following features:

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n

tr:nth-child(odd)]:bg-surface-sunken\">

Feature	SIM v1.0 TF2
Horovod Multi-GPU (NCCL)	Yes
Accelerated Linear Algebra (XLA)	Yes
Automatic mixed precision (AMP)	Yes
Preprocessing on GPU with NVTabular	Yes
BYO dataset	Yes

Features

Multi-GPU training with Horovod\nOur model uses Horovod to implement efficient multi-GPU training with NCCL. For details, refer to the example sources in this repository or refer to the TensorFlow tutorial.

Accelerated Linear Algebra (XLA)\nXLA 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.

Automatic Mixed Precision (AMP)\nAMP enables mixed precision training without any changes to the code-base by performing automatic graph rewrites and loss scaling controlled by an environmental variable.

Preprocessing on GPU with NVTabular\nPreprocessing on GPU with NVTabular - Amazon Reviews dataset preprocessing can be conducted using NVTabular. For more information on the framework, refer to this blog post.

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 NVIDIA Volta, and following with 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:

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

This can now be achieved using Automatic Mixed Precision (AMP) for TensorFlow to enable the full mixed precision methodology in your existing TensorFlow model code. AMP enables mixed precision training on NVIDIA Volta, and NVIDIA Ampere GPU architectures automatically. The TensorFlow framework code makes all necessary model changes internally.

In TF-AMP, the computational graph is optimized to use as few casts as necessary and maximize the use of FP16, and the loss scaling is automatically applied inside of supported optimizers. AMP can be configured to work with the existing tf.contrib loss scaling manager by disabling the AMP scaling with a single environment variable to perform only the automatic mixed-precision optimization. It accomplishes this by automatically rewriting all computation graphs with the necessary operations to enable mixed precision training and automatic loss scaling.

For information about:

How to train using mixed precision, refer to the Mixed Precision Training paper and Training With Mixed Precision documentation.
Techniques used for mixed precision training, refer to the Mixed-Precision Training of Deep Neural Networks blog.
How to access and enable AMP for TensorFlow, refer to Using TF-AMP from the TensorFlow User Guide.

Enabling mixed precision

To enable SIM training to use mixed precision, use --amp flag for 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 the matrix math, also called 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 NVIDIA 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 a 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.

BYO dataset functionality overview

This section describes how you can train the DeepLearningExamples RecSys models on your own datasets without changing\nthe model or data loader and with similar performance to the one published in each repository.\nThis can be achieved thanks to Dataset Feature Specification, which describes how the dataset, data loader, and model\ninteract with each other during training, inference, and evaluation.\nDataset Feature Specification has a consistent format across all recommendation models in NVIDIA's DeepLearningExamples\nrepository, regardless of dataset file type and the data loader,\ngiving you the flexibility to train RecSys models on your own datasets.

BYO dataset glossary

The Dataset Feature Specification consists of three mandatory and one optional section:

feature_spec provides a base of features that may be referenced in other sections, along with their metadata.\nFormat: dictionary (feature name) => (metadata name => metadata value)

source_spec provides information necessary to extract features from the files that store them.\nFormat: dictionary (mapping name) => (list of chunks)

Mappings are used to represent different versions of the dataset (think: train/validation/test, k-fold splits). A mapping is a list of chunks.
Chunks are subsets of features that are grouped together for saving. For example, some formats may constrain data saved in one file to a single data type. In that case, each data type would correspond to at least one chunk. Another example where this might be used is to reduce file size and enable more parallel loading. Chunk description is a dictionary of three keys:
\n
- type provides information about the format in which the data is stored. Not all formats are supported by all models.
- features is a list of features that are saved in a given chunk. The order of this list may matter: for some formats, it is crucial for assigning read data to the proper feature.
- files is a list of paths to files where the data is saved. For Feature Specification in yaml format, these paths are assumed to be relative to the yaml file's directory (basename). Order of this list matters: It is assumed that rows 1 to i appear in the first file, rows i+1 to j in the next one, etc.
\n

channel_spec determines how features are used. It is a mapping (channel name) => (list of feature names).

Channels are model-specific magic constants. In general, data within a channel is processed using the same logic. Example channels: model output (labels), categorical ids, numerical inputs, user data, and item data.

metadata is a catch-all, wildcard section: If there is some information about the saved dataset that does not fit into the other sections, you can store it here.

Dataset feature specification

Data flow can be described abstractly:\nInput data consists of a list of rows. Each row has the same number of columns; each column represents a feature.\nThe columns are retrieved from the input files, loaded, aggregated into channels and supplied to the model/training script.

FeatureSpec contains metadata to configure this process and can be divided into three parts:

\n
Specification of how data is organized on disk (source_spec). It describes which feature (from feature_spec) is stored in which file and how files are organized on disk.
\n
\n
Specification of features (feature_spec). Describes a dictionary of features, where key is the feature name and values are the features' characteristics such as dtype and other metadata (for example, cardinalities for categorical features)
\n
\n
Specification of model's inputs and outputs (channel_spec). Describes a dictionary of model's inputs where keys specify model channel's names and values specify lists of features to be loaded into that channel. Model's channels are groups of data streams to which common model logic is applied, for example categorical/continuous data, and user/item ids. Required/available channels depend on the model
\n

The FeatureSpec is a common form of description regardless of underlying dataset format, dataset data loader form, and model.

Data flow in NVIDIA Deep Learning Examples recommendation models

The typical data flow is as follows:

S.0. Original dataset is downloaded to a specific folder.
S.1. Original dataset is preprocessed into Intermediary Format. For each model, the preprocessing is done differently, using different tools. The Intermediary Format also varies (for example, for DLRM PyTorch, the Intermediary Format is a custom binary one.)
S.2. The Preprocessing Step outputs Intermediary Format with dataset split into training and validation/testing parts along with the Dataset Feature Specification yaml file. Metadata in the preprocessing step is automatically calculated.
S.3. Intermediary Format data, together with the Dataset Feature Specification, are fed into training/evaluation scripts. The data loader reads Intermediary Format and feeds the data into the model according to the description in the Dataset Feature Specification.
S.4. The model is trained and evaluated

\n \n
\n

Figure 3. Data flow in Recommender models in NVIDIA Deep Learning Examples repository. Channels of the model are drawn in green.

Example of dataset feature specification

As an example, let's consider a Dataset Feature Specification for a small CSV dataset for some abstract model.

feature_spec:\n  user_gender:\n    dtype: torch.int8\n    cardinality: 3 #M,F,Other\n  user_age: #treated as numeric value\n    dtype: torch.int8\n  user_id:\n    dtype: torch.int32\n    cardinality: 2655\n  item_id:\n    dtype: torch.int32\n    cardinality: 856\n  label:\n    dtype: torch.float32\n\nsource_spec:\n  train:\n    - type: csv\n      features:\n        - user_gender\n        - user_age\n      files:\n        - train_data_0_0.csv\n        - train_data_0_1.csv\n    - type: csv\n      features:\n        - user_id\n        - item_id\n        - label\n      files:\n        - train_data_1.csv\n  test:\n    - type: csv\n      features:\n        - user_id\n        - item_id\n        - label\n        - user_gender\n        - user_age\n        \n      files:\n        - test_data.csv\n\nchannel_spec:\n  numeric_inputs: \n    - user_age\n  categorical_user_inputs: \n    - user_gender\n    - user_id\n  categorical_item_inputs: \n    - item_id\n  label_ch: \n    - label\n

The data contains five features: (user_gender, user_age, user_id, item_id, label). Their data types and necessary metadata are described in the feature specification section.

In the source mapping section, two mappings are provided: one describes the layout of the training data, and the other of the testing data. The layout for training data has been chosen arbitrarily to showcase the flexibility.\nThe train mapping consists of two chunks. The first one contains user_gender and user_age, saved as a CSV, and is further broken down into two files. For specifics of the layout, refer to the following example and consult the glossary. The second chunk contains the remaining columns and is saved in a single file. Notice that the order of columns is different in the second chunk - this is alright, as long as the order matches the order in that file (that is, columns in the .csv are also switched)

Let's break down the train source mapping. The table contains example data color-paired to the files containing it.

\n\n

The channel spec describes how the data will be consumed. Four streams will be produced and available to the script/model.\nThe feature specification does not specify what happens further: names of these streams are only lookup constants defined by the model/script.\nBased on this example, we can speculate that the model has three input channels: numeric_inputs, categorical_user_inputs,\ncategorical_item_inputs, and one output channel: label.\nFeature names are internal to the FeatureSpec and can be freely modified.

BYO dataset functionality

In order to train any Recommendation model in NVIDIA Deep Learning Examples, one can follow one of three possible ways:

\n
One delivers preprocessed datasets in the Intermediary Format supported by data loader used by the training script\n(different models use different data loaders) together with FeatureSpec yaml file describing at least specification of dataset, features, and model channels
\n
\n
One uses a transcoding script (not supported in SIM model yet)
\n
\n
One delivers datasets in non-preprocessed form and uses preprocessing scripts that are a part of the model repository.\nIn order to use already existing preprocessing scripts, the format of the dataset needs to match one of the original datasets.\nThis way, the FeatureSpec file will be generated automatically, but the user will have the same preprocessing as in the original model repository.
\n

Glossary

Auxiliary loss is used to improve DIEN (so SIM as well) model training. It is constructed based on consecutive user actions from their short behavior history.

DIEN model was proposed in Deep Interest Evolution Network for Click-Through Rate Prediction paper as an extension of the DIN model. It can also be used as a backbone for processing short interaction sequences in the SIM model.

DIN model was proposed in Deep Interest Network for Click-Through Rate Prediction paper. It can be used as a backbone for processing short interaction sequences in the SIM model.

Long user behavior history is the record of past user interactions. They are processed by the General Search Unit part of the SIM model (refer to Figure 1). This typically is a lightweight model aimed at processing longer sequences.

Short user behavior history is the record of the most recent user interactions. They are processed by a more computationally intensive Exact Search Unit part of the SIM model (refer to Figure 1).

User behaviors are users' interactions with given items of interest. Example interactions include reviewed items for Amazon Reviews dataset or clicks in the e-commerce domain. All the systems contained in this repository focus on modeling user interactions.

","resourceVersions",[195,196,197],"resourceVersionsTotal",2,"hasMoreVersions","versionMetadata",{"_166":21,"_167":-5,"_48":168,"_169":-5,"_170":171,"_172":-5,"_173":79,"_174":16,"_175":30,"_176":-5,"_177":30,"_178":179,"_180":85,"_181":182,"_183":85,"_184":185,"_186":85,"_187":188,"_189":85,"_190":191,"_192":85,"_193":194},"versionNotFound","isLoggedIn","latestVersionPath","/orgs/nvidia/dle/resources/sim_tf2/-","version","imageReference","cliDownloadCommand","ngc registry resource download-version nvidia/dle/sim_tf2:22.01.0","helmFetchCommand","viewAllVersionsPath","/orgs/nvidia/dle/resources/sim_tf2/22.01.0/version-history","accessCheckResult",{"_165":85},"hasRelatedCollections","context",{"_161":162,"_163":164},1787757154988,[],[],[154,155,156,157,158,159,160],{"_44":59,"_46":69,"_48":61,"_50":62,"_52":70,"_54":55},{"_44":59,"_46":68,"_48":61,"_50":62,"_52":19,"_54":55},{"_44":59,"_46":66,"_48":61,"_50":62,"_52":67,"_54":55},{"_44":59,"_46":64,"_48":101,"_50":-5,"_52":-5,"_54":55},{"_44":59,"_46":60,"_48":61,"_50":62,"_52":63,"_54":55},{"_44":56,"_46":57,"_48":49,"_50":51,"_52":58,"_54":55},{"_44":45,"_46":47,"_48":49,"_50":51,"_52":53,"_54":55},"orgName","nvidia","teamName","dle","allowed","versionId","description","UPLOAD_COMPLETE","releaseType","createdDate",["D",1663857714586],"updatedDate","totalSizeInBytes","totalFileCount","isSigned","malwareScanStatus","isScanned","releaseNotes","### Changelog\n\nMay 2022\n- Initial release\n\n### Known issues\n\n- While benchmarking inference on a single GPU, sometimes throughput drops drastically in the middle of the epoch and remains low until the end of the epoch.\n- On a multi-GPU setup, the summary of throughput (in the last line of the logfile) is lower than it would result from each step`s throughput (sample/s). It is probably the case when a single GPU is slower than the one on the logging node. In this case, the overhead for synchronization before the final throughput calculation is higher than usual.\n- The SIM model results are non-deterministic, even using the same random seed. The reason for this non-determinism is the [tf.math.unsorted_segment_sum](https://www.tensorflow.org/api_docs/python/tf/math/unsorted_segment_sum) operation called within an optimization step. Its influence depends on categorical data distribution within a batch, and this issue is more severe for momentum-based optimizers. A potential solution is to use a deterministic version of this op which allows perfect reproduction, but is up to six times slower training.","hasReleaseNotes","setup","The following section lists the requirements that you need to meet in order to start training the SIM model.\n\n### Requirements\n\nThis repository contains a Dockerfile that extends the TensorFflow2 NGC container and encapsulates some dependencies. Aside from these dependencies, ensure you have the following components:\n- [NVIDIA Docker](https://github.com/NVIDIA/nvidia-docker)\n- [TensorFlow2 22.01-py3](https://ngc.nvidia.com/catalog/containers/nvidia:tensorflow/tags) NGC container\n- Supported GPUs:\n - [NVIDIA Volta architecture](https://www.nvidia.com/en-us/data-center/volta-gpu-architecture/)\n - [NVIDIA Ampere architecture](https://www.nvidia.com/en-us/data-center/nvidia-ampere-gpu-architecture/)\n\nFor more information about how to get started with NGC containers, refer to the following sections from the NVIDIA GPU Cloud Documentation and the Deep Learning Documentation:\n- [Getting Started Using NVIDIA GPU Cloud](https://docs.nvidia.com/ngc/ngc-getting-started-guide/index.html)\n- [Accessing And Pulling From The NGC Container Registry](https://docs.nvidia.com/deeplearning/frameworks/user-guide/index.html#accessing_registry)\n \nFor those unable to use the Tensorflow2 NGC container, to set up the required environment or create your own container, refer to the versioned [NVIDIA Container Support Matrix](https://docs.nvidia.com/deeplearning/frameworks/support-matrix/index.html).","hasSetup","quickStartGuide","To train your model using mixed or TF32 precision with Tensor Cores or using FP32, perform the following steps using the default parameters of the SIM model on the Amazon Reviews dataset. For the specifics concerning training and inference, refer to the [Advanced](#advanced) section.\n\n1. Clone the repository.\n ```bash\n git clone https://github.com/NVIDIA/DeepLearningExamples\n cd DeepLearningExamples/TensorFlow2/Recommendation/SIM\n ```\n\n2. Build the SIM Tensorflow2 container.\n ```bash\n docker build -t sim_tf2 .\n ```\n\n3. Start an interactive session in the NGC container to run preprocessing, training, or inference (Amazon Books dataset can be mounted if it has already been downloaded, otherwise, refer to point 4). The SIM TensorFlow2 container can be launched with:\n ```bash\n docker run --runtime=nvidia -it --rm --ipc=host --security-opt seccomp=unconfined -v ${AMAZON_DATASET_PATH}:${RAW_DATASET_PATH} sim_tf2 bash\n ```\n\n4. (Optional) Download Amazon Books dataset:\n\n ```bash\n scripts/download_amazon_books_2014.sh\n export RAW_DATASET_PATH=/data/amazon_books_2014\n ```\n\n5. Start preprocessing.\n\n For details of the required file format and certain preprocessing parameters refer to [BYO dataset](#byo-dataset).\n\n ```bash\n python preprocessing/sim_preprocessing.py \\\n --amazon_dataset_path ${RAW_DATASET_PATH} \\\n --output_path ${PARQUET_PATH}\n\n python preprocessing/parquet_to_tfrecord.py \\\n --amazon_dataset_path ${PARQUET_PATH} \\\n --tfrecord_output_dir ${TF_RECORD_PATH}\n ```\n\n6. Start training (`${GPU}` is an arbitrary number of GPUs to be used).\n ```bash\n mpiexec --allow-run-as-root --bind-to socket -np ${GPU} python main.py \\\n --dataset_dir ${TF_RECORD_PATH} \\\n --mode train \\\n --model_type sim \\\n --embedding_dim 16 \\\n --drop_remainder \\\n --optimizer adam \\\n --lr 0.01 \\\n --epochs 3 \\\n --global_batch_size 131072 \\\n --amp\n ```\n\n7. Start inference.\n ```bash\n mpiexec --allow-run-as-root --bind-to socket -np ${GPU} python main.py \\\n --dataset_dir ${TF_RECORD_PATH} \\\n --mode inference \\\n --model_type sim \\\n --embedding_dim 16 \\\n --global_batch_size 131072 \\\n --amp\n ```\n\nFor the explanation of output logs, refer to [Log format](#log-format) section.\n\nNow that you have your model trained and evaluated, you can choose to compare your training results with our [Training accuracy results](#training-accuracy-results). You can also choose to benchmark your performance to [Training performance benchmark](#training-performance-results), or [Inference performance benchmark](#inference-performance-results). Following the steps in these sections will ensure that you achieve the same accuracy and performance results as stated in the [Results](#results) section.","hasQuickStartGuide","advanced","The following sections provide greater details of the dataset, running training and inference, and the training results.\n\n### Scripts and sample code\n\nThe `main.py` script provides an entry point to all the provided functionalities. This includes running training and inference modes. The behavior of the script is controlled by command-line arguments listed below in the [Parameters](#parameters) section. The `preprocessing` folder contains scripts to prepare data. In particular, the `preprocessing/sim_preprocessing.py` script can be used to preprocess the Amazon Reviews dataset while `preprocessing/parquet_to_tfrecord.py` transforms parquet files to TFRecords for loading data efficiently.\n\nModels are implemented in corresponding modules in the `sim/models` subdirectory, for example, `sim/models/sim_model.py` for the SIM model. The `sim/layers` module contains definitions of different building blocks for the models. Finally, the `sim/data` subdirectory provides modules for the dataloader. Other useful utilities are contained in the `sim/utils` module.\n\n### Parameters\n\nThe `main.py` script parameters are detailed in the following table.\n\n| Scope | Parameter | Description | Default Value |\n|-----------------|---------------------------|-------------------------------------------------------------------------|---------------------------|\n| model | model_type | Model type | sim |\n| model | embedding_dim | Embedding dimension for different entities (users, items & categories) | 16 |\n| model | stage_one_mlp_dims | MLP hidden dimensions for the stage one component | 200 |\n| model | stage_two_mlp_dims | MLP hidden dimensions for the stage two component | 200,80 |\n| model | aux_mlp_dims | MLP hidden dimensions for the auxiliary loss | 100,50 |\n| datasets | dataset_dir | Path to the directory containing feature specification file and dataset splits | -- |\n| datasets | feature_spec | Name of the feature spec file in the dataset directory | feature_spec.yaml |\n| training | optimizer | Optimizer to use | adam |\n| training | lr | Learning rate for the optimizer selected | 0.01 |\n| training | weight_decay | Parameters decay of the optimizer selected | 0 |\n| training | epochs | Train for the following number of epochs | 3 |\n| training | global_batch_size | Batch size used for training, evaluation and inference | 131072 |\n| training | dropout_rate | Dropout rate for all the classification MLPs | -1 (disabled) |\n| training | amp | Enable automatic mixed precision training (flag) | False |\n| training | xla | Enable XLA conversion (flag) | False |\n| training | drop_remainder | Drop remainder batch for training set (flag) | False |\n| training | disable_cache | Disable dataset caching after the first time it is iterated over (flag) | False |\n| training | repeat_count | Repeat training dataset this number of times | 0 |\n| training | prefetch_train_size | Number of batches to prefetch in training. | 10 |\n| training | prefetch_test_size | Number of batches to prefetch in evaluation. | 2 |\n| training | long_seq_length | Determines the long history - short history split of history features | 90 |\n| training | prebatch_train_size | Batch size of batching applied during preprocessing to train dataset. | 0 |\n| training | prebatch_test_size | Batch size of batching applied during preprocessing to test dataset. | 0 |\n| results | results_dir | Path to the model result files storage | /tmp/sim |\n| results | log_filename | Name of the file to store logger output | log.json |\n| results | save_checkpoint_path | Directory to save model checkpoints | \"\" |\n| results | load_checkpoint_path | Directory to restore model checkpoints from | \"\" |\n| run mode | mode | One of: train, inference. | train |\n| run mode | benchmark | Perform benchmark measurements for, e.g., throughput calculation (flag) | False |\n| run mode | benchmark_warmup_steps | Number of warmup steps to use for performance benchmarking | 20 |\n| run mode | benchmark_steps | Number of steps to use for performance benchmarking | 200 |\n| run mode | affinity | Type of CPU affinity | socket_unique_interleaved |\n| run mode | inter_op_parallelism | Number of inter op threads | 0 |\n| run mode | intra_op_parallelism | Number of intra op threads | 0 |\n| run mode | num_parallel_calls | Parallelism level for tf.data API. If None, heuristic based on number of CPUs and number of GPUs will be used | None |\n| reproducibility | seed | Random seed | -1 |\n\n### Command-line options\n\nTo view the full list of available options and their descriptions, use the `--help` command-line option, for example:\n```bash\npython main.py --help\n```\n\n### Getting the data\n\nThe SIM model was trained on the Books department subset of [Amazon Reviews](https://snap.stanford.edu/data/web-Amazon.html) dataset. The dataset is split into two parts: training and test data. The test set for evaluation was constructed using the last user interaction from user behavior sequences. All the preceding interactions are used for training.\n\nThis repository contains the `scripts/download_amazon_books_2014.sh`, which can be used to download the dataset.\n\n#### Dataset guidelines\n\nThe preprocessing steps applied to the raw data include:\n- Sampling negatives randomly (out of all possible items)\n- Choosing the last category as the item category (in case more than one is available)\n- Determining embedding table sizes for categorical features needed to construct a model\n- Filter users for training split based on their number of interactions (discard users with less than 20 interactions)\n\n#### Prebatching\n\nPreprocessing scripts allow to apply batching prior to the model`s dataloader. This reduces the size of produced TFrecord files and speeds up dataloading.\nTo do so, specify `--prebatch_train_size` and `--prebatch_test_size` while converting data using `scripts/parquet_to_tfrecord.py`. Later, while using the `main.py` script, pass the information about applied prebatch size via the same parameters.\n\nExample\n\nStart preprocessing from step 5. from [Quick Start Guide](#quick-start-guide):\n\n```bash\npython preprocessing/sim_preprocessing.py \\\n--amazon_dataset_path ${RAW_DATASET_PATH} \\\n--output_path ${PARQUET_PATH}\n\npython preprocessing/parquet_to_tfrecord.py \\\n--amazon_dataset_path ${PARQUET_PATH} \\\n--tfrecord_output_dir ${TF_RECORD_PATH} \\\n--prebatch_train_size ${PREBATCH_TRAIN_SIZE} \\\n--prebatch_train_size ${PREBATCH_TEST_SIZE}\n```\n\nAnd then train the model (step 6.):\n\n```bash\nmpiexec --allow-run-as-root --bind-to socket -np ${GPU} python main.py \\\n--dataset_dir ${TF_RECORD_PATH} \\\n--mode train \\\n--model_type sim \\\n--embedding_dim 16 \\\n--drop_remainder \\\n--optimizer adam \\\n--lr 0.01 \\\n--epochs 3 \\\n--global_batch_size 131072 \\\n--amp \\\n--prebatch_train_size ${PREBATCH_TRAIN_SIZE} \\\n--prebatch_train_size ${PREBATCH_TEST_SIZE}\n```\n\n

Prebatching details

\n\n- The last batch for each split will pe saved to the separate file `remainder.tfrecord` unless there are enough samples to form a full batch.\n- Final batch size used in main script can be a multiple of prebatch size.\n- Final batch size used in main script can be a divider of prebatch size. In this case, when using multi GPU training, the number of batches received by each worker can be greater than 1 thus resulting in error during allgather operation. Dataset size, batch size and prebatch size have to be chosen with that limitation in mind.\n- For the orignal Amazon Books Dataset, parameters were set to PREBATCH_TRAIN_SIZE = PREBATCH_TEST_SIZE = 4096 for performance benchmarking purposes.\n

\n\n \n\n#### BYO dataset \n\nThis implementation supports using other datasets thanks to BYO dataset functionality. \nBYO dataset functionality allows users to plug in their dataset in a common fashion for all Recommender models \nthat support this functionality. Using BYO dataset functionality, the user does not have to modify the source code of \nthe model thanks to the Feature Specification file. For general information on how the BYO dataset works, refer to the \n[BYO dataset overview section](#byo-dataset-functionality-overview).\n\nFor usage of preprocessing scripts refer to [Quick Start Guide](#quick-start-guide)\n\nThere are currently two ways to plug in the user's dataset:\n\n

1. Provide preprocessed dataset in parquet format, then use parquet_to_tfrecord.py script to convert it to Intermediary Format and automatically generate FeatureSpec.

\n\nParquet $DATASET needs to have the following directory structure (or change the name with script arguments):\n\n```\nDATASET:\n metadata.json\n test:\n part.0.parquet\n part.1.parquet\n .\n .\n .\n train:\n part.0.parquet\n part.1.parquet\n .\n .\n .\n```\n\n`metadata.json` should contain cardinalities of each categorical feature present in the dataset and be of the following structure: (for features `uid`, `item`, `cat`)\n\n```yaml\n{\n \"cardinalities\": [\n {\"name\": \"uid\", \"value\": 105925}, \n {\"name\": \"item\", \"value\": 1209081}, \n {\"name\": \"cat\", \"value\": 2330}\n ]\n}\n```\n\nMake sure the dataset's columns are in the same order as entries in `metadata.json` (for user features and **item features in each channel**)\n\nColumns of parquets files must be organized in a specific order:\n- one column with `label` values\n- `number_of_user_features` (to be specified in script argument) columns follow with each **user feature** in the separate column\n- `number_of_item_features` columns. One column for each feature of **target (query) item**\n- `number_of_item_features` columns. Column with index i contains **sequence of item_feature_{i}** of **positive_history**\n- `number_of_item_features` columns. Column with index i contains **sequence of item_feature_{i} **of **negative_history**\n\n

\n\n

2. Provide preprocessed dataset in tfrecord format with feature_spec.yaml describing the details.

\n\nRequired channels and sample layout can be found in the configuration shown below. This is the file layout and feature specification for the original Amazon dataset.\n\nFiles layout:\n```\nTF_RECORD_PATH:\n feature_spec.yaml\n test.tfrecord\n train.tfrecord\n```\n\nfeature_spec.yaml:\n```yaml\nchannel_spec:\n label:\n - label\n negative_history:\n - item_id_neg\n - cat_id_neg\n positive_history:\n - item_id_pos\n - cat_id_pos\n target_item_features:\n - item_id_trgt\n - cat_id_trgt\n user_features:\n - user_id\nfeature_spec:\n item_id_neg:\n cardinality: 1209081\n dimensions:\n - 100\n dtype: int64\n item_id_pos:\n cardinality: 1209081\n dimensions:\n - 100\n dtype: int64\n item_id_trgt:\n cardinality: 1209081\n dtype: int64\n cat_id_neg:\n cardinality: 2330\n dimensions:\n - 100\n dtype: int64\n cat_id_pos:\n cardinality: 2330\n dimensions:\n - 100\n dtype: int64\n cat_id_trgt:\n cardinality: 2330\n dtype: int64\n label:\n dtype: bool\n user_id:\n cardinality: 105925\n dtype: int64\nmetadata: {}\nsource_spec:\n test:\n - features: &id001\n - label\n - user_id\n - item_id_trgt\n - cat_id_trgt\n - item_id_pos\n - cat_id_pos\n - item_id_neg\n - cat_id_neg\n files:\n - test.tfrecord\n type: tfrecord\n train:\n - features: *id001\n files:\n - train.tfrecord\n type: tfrecord\n```\n\n`dimensions` should contain the length of the sequencial features.\n\nNote that corresponsive features in `negative_history`, `positive_history`, `target_item_features` need to be listed in the same order in channel spec in each channel since they share embedding tables in the model. (for example `item_id` needs to be first and `cat_id` second). \n\n

\n\n \n\n##### Channel definitions and requirements\n\nThis model defines five channels:\n\n- label, accepting a single feature\n- negative_history, accepting a categorical ragged tensor for an arbitrary number of features\n- positive_history, accepting a categorical ragged tensor for an arbitrary number of features\n- target_item_features, accepting an arbitrary number of categorical features\n- user_features, accepting an arbitrary number of categorical features\n\nFeatures in `negative_history`, `positive_history` and `target_item_features` channels must be equal in number and must be defined in the same order in channel spec.\n\nThe training script expects two mappings:\n\n- train\n- test\n\nFor performance reasons, the only supported dataset type is tfrecord.\n\n### Training process\n\nTraining can be run using `main.py` script by specifying the `--mode train` parameter. The speed of training is measured by throughput, that is, the number of samples processed per second. Evaluation is based on the [Area under ROC Curve (ROC AUC)](https://en.wikipedia.org/wiki/Receiver_operating_characteristic) metric. Model checkpoints may be stored using Checkpoint manager via the `--save_checkpoint_path` and `--load_checkpoint_path` parameters. Training and inference logs are saved to a directory specified via the `--results_dir` parameter. Mixed precision training is supported via the `--amp` flag. Multi-GPU training is performed using mpiexec and Horovod libraries.\n\n### Inference process\n\nInference can be run using `main.py` script by specifying the `--mode inference` parameter. It is performed using a dummy model initialized randomly, and it is intended to measure inference throughput. The most important parameter for inference is the batch size.\n\nExample usage of training and inference are demonstrated in [Quick Start Guide](#quick-start-guide).\n\n### Log format\n\nThere are three type of log lines during model execution. Each of them have `step` value, however it is formatted differently based on the type of log:\n- step log - step value is in format `[epoch, step]`:\n\nDLLL {\"timestamp\": ..., \"datetime\": ..., \"elapsedtime\": ..., \"type\": ..., `\"step\": [2, 79]`, \"data\": ...}\n\n- end of epoch log - step value is in format `[epoch]`:\n\nDLLL {\"timestamp\": ..., \"datetime\": ..., \"elapsedtime\": ..., \"type\": ..., `\"step\": [2]`, \"data\": ...}\n\n- summary log - logged once at the end of script execution. Step value is in fomat `[]`:\n\nDLLL {\"timestamp\": ..., \"datetime\": ..., \"elapsedtime\": ..., \"type\": ..., `\"step\": []`, \"data\": ...}\n\nIn those logs, `data` field contains dictonary in form `{metric: value}`. Metrics logged differ based on log type (step, end of epoch, summary) and model mode (training, inference).\n\n#### Training log data\n- step log \n - classification_loss - loss at the final output of the model.\n - dien_aux_loss - loss at the output of auxiliary model.\n - total_loss - sum of the above.\n - samples/s - estimated throughput in samples per second.\n- end of epoch log \n - throughput - average throughput during epoch in samples/s.\n - time - epoch time in seconds.\n - train_auc - AUC during evaluation on train set.\n - test_auc - AUC during evaluation on test set.\n - train_loss - loss during evaluation on train set.\n - test_loss - loss during evaluation on test set.\n - latency_[mean, p90, p95, p99] - latencies in miliseconds.\n- summary log \n - time_to_train - total training time in seconds.\n - train_auc, test_auc, train_loss, test_loss - results from the last epoch (see above).\n\n#### Inference log data\n- step log \n - samples/s - estimated throughput in samples per second.\n- end of epoch log is not present\n- summary log \n - throughput - average throughput during epoch in samples/s. \n - time - total execution time in seconds.\n - latency_[mean, p90, p95, p99] - latencies in miliseconds.","hasAdvanced","performance","The performance measurements in this document were conducted at the time of publication and may not reflect the performance achieved from NVIDIA's latest software release. For the most up-to-date performance measurements, go to [NVIDIA Data Center Deep Learning Product Performance](https://developer.nvidia.com/deep-learning-performance-training-inference).\n\n### Benchmarking\n\nThe following section shows how to run benchmarks measuring the model performance in training and inference modes.\n\n#### Training performance benchmark\n\nTo benchmark the training performance on a specific batch size, run:\n\n```bash\nmpiexec --allow-run-as-root --bind-to socket -np ${GPU} python main.py \\\n --dataset_dir ${TF_RECORD_PATH} \\\n --mode train \\\n --model_type sim \\\n --global_batch_size 131072 \\\n --drop_remainder \\\n --amp \\\n --benchmark \\\n --prebatch_train_size ${PREBATCH_TRAIN_SIZE} \\\n --prebatch_test_size ${PREBATCH_TEST_SIZE}\n```\n\nEquivalent:\n```bash\nscripts/run_model.sh \\\n --data_path ${TF_RECORD_PATH} \\\n --gpus ${GPU} \\\n --amp 1 \\\n --benchmark 1 \\\n --prebatch_train_size ${PREBATCH_TRAIN_SIZE} \\\n --prebatch_test_size ${PREBATCH_TEST_SIZE}\n```\n\n#### Inference performance benchmark\n\nTo benchmark the inference performance on a specific batch size, run:\n\n```bash\nmpiexec --allow-run-as-root --bind-to socket -np ${GPU} python main.py \\\n --dataset_dir ${TF_RECORD_PATH} \\\n --mode inference \\\n --model_type sim \\\n --global_batch_size 131072 \\\n --amp \\\n --benchmark \\\n --prebatch_train_size ${PREBATCH_TRAIN_SIZE} \\\n --prebatch_test_size ${PREBATCH_TEST_SIZE}\n```\n\nEquivalent:\n```bash\nscripts/run_model.sh \\\n --data_path ${TF_RECORD_PATH} \\\n --gpus ${GPU} \\\n --amp 1 \\\n --benchmark 1 \\\n --prebatch_train_size ${PREBATCH_TRAIN_SIZE} \\\n --prebatch_test_size ${PREBATCH_TEST_SIZE}\n```\n\n### Results\n\nThe following sections provide details on how we achieved our performance and accuracy in training and inference.\n\n#### Training accuracy results\n\nOur results were obtained by running the `run_model.sh` bash script in the TensorFlow2 21.10-py3 NGC container. Experiments were run on 1 and 8 GPUs, with FP32/TF32 Precision and AMP and with XLA-OFF/XLA-ON. Dataset was prebatched with the size of 16384. Other parameters were set to defaults.\n\nThere were 10 runs for each configuration. In the `Training accuracy` sections, average values are reported. In the `Training stability` sections, values from all runs are included in plots.\n\n##### Training accuracy: NVIDIA DGX A100 (8x A100 80GB)\n\n|GPUs | XLA |Time to train - TF32 (seconds) | Time to train - mixed precision (seconds) | AUC - TF32 | AUC - mixed precision | Time to train speedup (TF32 to mixed precision) |\n|-----|----------|-------|--------|---------|-------------|---------------|\n|1 |XLA-OFF |133.62 | 109.29 |0.82 |0.811 | 1.22 |\n|1 |XLA-ON |132.31 | 113.91 |0.811 |0.822 | 1.16 |\n|8 |XLA-OFF |35.17 | 34.08 |0.813 |0.808 | 1.03 |\n|8 |XLA-ON |39.19 | 40.16 |0.814 |0.811 | 0.98 |\n\n##### Training accuracy: NVIDIA DGX-1 (8x V100 32GB)\n\n|GPUs | XLA |Time to train - FP32 (seconds) | Time to train - mixed precision (seconds) | AUC - FP32 | AUC - mixed precision | Time to train speedup (FP32 to mixed precision) |\n|-----|----------|-------|--------|---------|-------------|---------------|\n|1 |XLA-OFF | 210.70 | 154.54 |0.815 |0.817 | 1.36 |\n|1 |XLA-ON | 203.61 | 159.80 | 0.816 |0.813 | 1.27 |\n|8 |XLA-OFF | 48.643 | 44.02 | 0.811 | 0.817 | 1.11|\n|8 |XLA-ON | 55.26 | 54.33 | 0.814 | 0.817 | 1.02|\n\n##### Training stability test\n\nTraining stability was tested over 10 runs for each configuration of double precision / AMP, XLA-ON / XLA-OFF on 1 GPU and 8 GPUs for both Volta and Ampere architectures. \nEach run used the same random seed and default values of training hyperparameters.\nTraining was performed on DGX A100 80GB and DGX-1 V100 32GB setups.\nAUC metric achieved on test set after training is presented in the following plots.\n\n(Plot represents XLA-OFF results, for XLA-ON results, check expandable part below)\n\n

\n \n
\nFigure 4. Training stability plot, distribution of AUC across different configurations with XLA-OFF. \n

\n\n

\nTraining stability with XLA-ON\n

\n\n

\n \n
\nFigure 5. Training stability plot, distribution of AUC across different configurations with XLA-ON.\n

\n\n

\nComplete list of training stability results\n

\n\n| |GPUs |Precision |XLA |Mean AUC |Std AUC |Min AUC |Max AUC |\n|----------|-----|----------|----------|----------|----------|----------|----------|\n|DGX A100 |1 |TF32 |XLA-OFF |0.8195 |0.0083 |0.7981 |0.8307 |\n|DGX A100 |1 |TF32 |XLA-ON |0.8106 |0.0066 |0.8012 |0.8211 |\n|DGX A100 |1 |AMP |XLA-OFF |0.8110 |0.0103 |0.7939 |0.8244 |\n|DGX A100 |1 |AMP |XLA-ON |0.8224 |0.0067 |0.8115 |0.8397 |\n|DGX A100 |8 |TF32 |XLA-OFF |0.8127 |0.0070 |0.8027 |0.8285 |\n|DGX A100 |8 |TF32 |XLA-ON |0.8143 |0.0079 |0.8012 |0.8251 |\n|DGX A100 |8 |AMP |XLA-OFF |0.8084 |0.0121 |0.7850 |0.8203 |\n|DGX A100 |8 |AMP |XLA-ON |0.8109 |0.0077 |0.8018 |0.8281 |\n|DGX-1 V100|1 |FP32 |XLA-OFF |0.8152 |0.0075 |0.8006 |0.8255 |\n|DGX-1 V100|1 |FP32 |XLA-ON |0.8158 |0.0055 |0.8060 |0.8261 |\n|DGX-1 V100|1 |AMP |XLA-OFF |0.8172 |0.0045 |0.8097 |0.8237 |\n|DGX-1 V100|1 |AMP |XLA-ON |0.8133 |0.0070 |0.7987 |0.8234 |\n|DGX-1 V100|8 |FP32 |XLA-OFF |0.8112 |0.0055 |0.8027 |0.8182 |\n|DGX-1 V100|8 |FP32 |XLA-ON |0.8144 |0.0087 |0.8037 |0.8281 |\n|DGX-1 V100|8 |AMP |XLA-OFF |0.8173 |0.0061 |0.8080 |0.8277 |\n|DGX-1 V100|8 |AMP |XLA-ON |0.8169 |0.0109 |0.7952 |0.8326 |\n\n

\n \n\nFor both NVIDIA Ampere and NVIDIA Volta, even though the same seed was used for each run, there is a still noticeable variance. The reason for that are built-in non-deterministic GPU kernels in [tf.math.unsorted_segment_sum](https://www.tensorflow.org/api_docs/python/tf/math/unsorted_segment_sum) operation. However, since it is six times faster than the deterministic implementation of this operation, this is the preferable solution.\n\n##### Impact of mixed precision on training accuracy\nResults in this section present the impact of enabling AMP on the AUC. Models were trained using default parameters, on 1/8 GPUs and on Volta/Ampere architecture.\n\nAUC is measured on test set after model training.\n\n(Plot represents XLA-OFF results, for XLA-ON results, check expandable part below)\n\n

\n \n
\nFigure 6. Impact of AMP on test set AUC (XLA-OFF)\n

\n\n

\nImpact of AMP on AUC for XLA-ON\n

\n\n

\n \n
\nFigure 7. Impact of AMP on test set AUC (XLA-ON)\n

\n\n

\n \n\nDistribution 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.\n\n

\nFull tabular data for AMP influence on AUC\n

\n\n| |GPUs |XLA |Mean AUC for Full precision (TF32 for A100, FP32 for V100)|Std AUC for Full precision (TF32 for A100, FP32 for V100)|Mean AUC for AMP |Std AUC for AMP |KS test value: statistics, p-value|\n|------|-----|-----|------------------------------|------------------------------|--------------------|--------------------|--------------------|\n|DGX A100|1 |XLA-OFF|0.8195 |0.0083 |0.8110 |0.0103 |0.6000, 0.0524 |\n|DGX A100|1 |XLA-ON|0.8106 |0.0066 |0.8224 |0.0067 |0.7000, 0.0123 |\n|DGX A100|8 |XLA-OFF|0.8127 |0.0070 |0.8084 |0.0121 |0.2000, 0.9945 |\n|DGX A100|8 |XLA-ON|0.8143 |0.0079 |0.8109 |0.0077 |0.4000, 0.4175 |\n|DGX-1 V100|1 |XLA-OFF|0.8152 |0.0075 |0.8172 |0.0045 |0.2000, 0.9945 |\n|DGX-1 V100|1 |XLA-ON|0.8158 |0.0055 |0.8133 |0.0070 |0.2000, 0.9945 |\n|DGX-1 V100|8 |XLA-OFF|0.8112 |0.0055 |0.8173 |0.0061 |0.4000, 0.4175 |\n|DGX-1 V100|8 |XLA-ON|0.8144 |0.0087 |0.8169 |0.0109 |0.4000, 0.4175 |\n\n

\n \n\n#### Training accuracy plot\n\nModels trained with FP32, TF32, and Automatic Mixed Precision (AMP) achieve similar accuracy.\n\nPlot represents ROC AUC on the test set for 1 and 8 GPUs, with precision FP32/TF32 (for Volta/Ampere) and AMP. All other training parameters are default.\n\n

\n \n
\nFigure 8. ROC curve for different configurations of Ampere/Volta, 1/8 GPUs, double precision / AMP. (XLA-OFF)\n

\n\n#### Training performance results\n\nOur results were obtained by running the `scripts/run_model.sh` script in the TensorFlow2 21.10-py3 NGC container. Dataset was prebatched with the size of 16384.\n\nNumbers were averaged over 10 separate runs for each configuration.\n\nFor each run, performance numbers (in samples per second) were averaged over training steps from one epoch which gives reliable estimates of the throughput. We also exclude the first 20 steps of training as a warmup phase.\n\nThe cumulative batch size of all GPUs in performance tests was set to 131072.\n\nTo achieve these same results, follow the steps in the [Quick Start Guide](#quick-start-guide).\n\n##### Training performance: NVIDIA DGX A100 (8x A100 80GB)\n\n| GPUs | XLA | Throughput - TF32 (samples/s) | Throughput - mixed precision (samples/s) | Throughput speedup (mixed precision / TF32) | Strong scaling - TF32 | Strong scaling - mixed precision |\n|-------:|------:|--------------------------------:|-------------------------------------------:|----------------------------------------------:|------------------------:|-----------------------------------:|\n| 1 | OFF | 377254.65 | 479921.54 | 1.27 | 1.00 | 1.00 |\n| 1 | ON | 455724.01 | 565221.04 | 1.24 | 1.00 | 1.00 |\n| 8 | OFF | 2161681.55 | 2603489.60 | 1.20 | 5.73 | 5.42 |\n| 8 | ON | 2662368.18 | 2979441.80 | 1.12 | 5.84 | 5.27 |\n\n

\nNVIDIA DGX A100 XLA-ON / XLA-OFF training speedup\n

\n\nFor each configuration of parameters present in the table, the `Speedup` column shows the speedup achieved by turning on XLA.\n\n|GPUs |Precision |Speedup |\n|-----|---------------|--------|\n|1 |TF32 |1.208 |\n|1 |AMP |1.178 |\n|8 |TF32 |1.232 |\n|8 |AMP |1.119 |\n

\n\n \n\n##### Training performance: NVIDIA DGX-2 (16x V100 32GB)\n\n| GPUs | XLA | Throughput - FP32 (samples/s) | Throughput - mixed precision (samples/s) | Throughput speedup (mixed precision / FP32) | Strong scaling - FP32 | Strong scaling - mixed precision |\n|-------:|------:|--------------------------------:|-------------------------------------------:|----------------------------------------------:|------------------------:|-----------------------------------:|\n| 1 | OFF | 209376.38 | 309752.48 | 1.48 | 1.00 | 1.00 |\n| 1 | ON | 245414.62 | 348945.59 | 1.42 | 1.00 | 1.00 |\n| 8 | OFF | 1310239.01 | 1689602.79 | 1.29 | 6.26 | 5.45 |\n| 8 | ON | 1483120.32 | 1962226.32 | 1.32 | 6.04 | 5.62 |\n| 16 | OFF | 2127221.65 | 2555926.79 | 1.20 | 10.16 | 8.25 |\n| 16 | ON | 2450499.40 | 2788997.07 | 1.14 | 9.99 | 7.99 |\n\n

\nNVIDIA DGX-2 XLA-ON / XLA-OFF training speedup\n

\n\nFor each configuration of parameters present in the table, the `Speedup` column shows the speedup achieved by turning on XLA.\n\n|GPUs |AMP |Speedup |\n|-----|--------------------|---------------|\n|1 |FP32 |1.172 |\n|1 |AMP |1.127 |\n|8 |FP32 |1.132 |\n|8 |AMP |1.161 |\n|16 |FP32 |1.152 |\n|16 |AMP |1.091 |\n

\n\n \n\n

\nNVIDIA DGX A100 / DGX-2 (Ampere / Volta) training speedup\n

\n\n| GPUs | XLA | Precision | Speedup |\n|-------:|------:|:------------|----------:|\n| 1 | OFF | TF32/FP32 | 1.802 |\n| 1 | OFF | AMP | 1.549 |\n| 1 | ON | TF32/FP32 | 1.857 |\n| 1 | ON | AMP | 1.620 |\n| 8 | OFF | TF32/FP32 | 1.650 |\n| 8 | OFF | AMP | 1.541 |\n| 8 | ON | TF32/FP32 | 1.795 |\n| 8 | ON | AMP | 1.518 |\n\n

\n\n \n\n#### Inference performance results\n\nOur results were obtained by running the `scripts/run_model.sh` script in the TensorFlow2 21.10-py3 NGC container. \n\nNumbers were averaged over 10 separate runs for each configuration.\n\nFor each run, performance numbers (in samples per second) were averaged over training steps from one epoch which gives reliable estimates of the throughput. We also exclude the first 20 steps of training as a warmup phase.\n\nTo achieve these same results, follow the steps in the [Quick Start Guide](#quick-start-guide).\n\n##### Inference performance: NVIDIA DGX A100 (8x A100 80GB)\n\n| Batch Size | XLA | Throughput - TF32 (samples/s) | Throughput - mixed precision (samples/s) | Throughput speedup (mixed precision / TF32) |\n|--------------------:|------:|--------------------------------:|-------------------------------------------:|----------------------------------------------:|\n| 4096 | ON | 618547.45 | 669640.65 | 1.08 |\n| 8192 | ON | 722801.14 | 849101.88 | 1.17 |\n| 16384 | ON | 859418.77 | 1051361.67 | 1.22 |\n| 32768 | ON | 976771.70 | 1269000.97 | 1.30 |\n| 65536 | ON | 1082688.51 | 1444729.52 | 1.33 |\n| 131072 | ON | 1094733.64 | 1483542.86 | 1.36 |\n\n

Complete table of DGX A100 inference performance results

\n\n| Batch Size | XLA | Precision | Throughput (samples/s) |\n|-------------:|:------|:------------|:--------------------------|\n| 4096 | OFF | TF32 | 708349.73 ± 14161.58 |\n| 8192 | OFF | TF32 | 873335.82 ± 8539.56 |\n| 16384 | OFF | TF32 | 937987.79 ± 12114.34 |\n| 32768 | OFF | TF32 | 943313.07 ± 8631.81 |\n| 65536 | OFF | TF32 | 960794.46 ± 7388.45 |\n| 131072 | OFF | TF32 | 966245.27 ± 8637.82 |\n| 4096 | OFF | AMP | 645394.94 ± 14844.27 |\n| 8192 | OFF | AMP | 919410.07 ± 11355.28 |\n| 16384 | OFF | AMP | 1136346.66 ± 14529.91 |\n| 32768 | OFF | AMP | 1216810.45 ± 21013.12 |\n| 65536 | OFF | AMP | 1287305.05 ± 19373.18 |\n| 131072 | OFF | AMP | 1298478.97 ± 10733.67 |\n| 4096 | ON | TF32 | 618547.45 ± 6569.97 |\n| 8192 | ON | TF32 | 722801.14 ± 9448.19 |\n| 16384 | ON | TF32 | 859418.77 ± 10012.61 |\n| 32768 | ON | TF32 | 976771.70 ± 13377.36 |\n| 65536 | ON | TF32 | 1082688.51 ± 8523.55 |\n| 131072 | ON | TF32 | 1094733.64 ± 11157.18 |\n| 4096 | ON | AMP | 669640.65 ± 9319.68 |\n| 8192 | ON | AMP | 849101.88 ± 14068.04 |\n| 16384 | ON | AMP | 1051361.67 ± 15310.42 |\n| 32768 | ON | AMP | 1269000.97 ± 23971.56 |\n| 65536 | ON | AMP | 1444729.52 ± 18011.54 |\n| 131072 | ON | AMP | 1483542.86 ± 6751.29 |\n\n

\n\n

\nDGX A100 XLA-ON / XLA-OFF inference Speedup\n

\n\nFor each configuration of parameters present in the table, the `Speedup` column shows the speedup achieved by turning on XLA.\n\n|Batch Size |Precision |Speedup |\n|--------------------|---------------|--------|\n|4096 |TF32 |0.873 |\n|8192 |TF32 |0.828 |\n|16384 |TF32 |0.916 |\n|32768 |TF32 |1.035 |\n|65536 |TF32 |1.127 |\n|131072 |TF32 |1.133 |\n|4096 |AMP |1.038 |\n|8192 |AMP |0.924 |\n|16384 |AMP |0.925 |\n|32768 |AMP |1.043 |\n|65536 |AMP |1.187 |\n|131072 |AMP |1.143 |\n\n

\n\n \n\n##### Inference performance: NVIDIA DGX-2 (16x V100 32GB)\n\n| Batch Size | XLA | Throughput - FP32 (samples/s) | Throughput - mixed precision (samples/s) | Throughput speedup (mixed precision / FP32) |\n|--------------------:|------:|--------------------------------:|-------------------------------------------:|----------------------------------------------:|\n| 4096 | ON | 444532.22 | 541975.24 | 1.22 |\n| 8192 | ON | 505047.64 | 642784.48 | 1.27 |\n| 16384 | ON | 549325.54 | 727077.63 | 1.32 |\n| 32768 | ON | 587452.73 | 788606.35 | 1.34 |\n| 65536 | ON | 605187.67 | 832651.59 | 1.38 |\n| 131072 | ON | 599557.03 | 840602.90 | 1.40 |\n\n

\nComplete table of DGX-2 inference performance results\n

\n\n| Batch Size | XLA | Precision | Throughput (samples/s) |\n|-------------:|:------|:------------|:--------------------------|\n| 4096 | OFF | FP32 | 459175.30 ± 23184.33 |\n| 8192 | OFF | FP32 | 499179.20 ± 15967.26 |\n| 16384 | OFF | FP32 | 525180.72 ± 2521.56 |\n| 32768 | OFF | FP32 | 532042.10 ± 4020.44 |\n| 65536 | OFF | FP32 | 534307.20 ± 7276.26 |\n| 131072 | OFF | FP32 | 532311.44 ± 6195.16 |\n| 4096 | OFF | AMP | 581771.66 ± 6163.50 |\n| 8192 | OFF | AMP | 665048.04 ± 4607.95 |\n| 16384 | OFF | AMP | 716355.19 ± 7174.98 |\n| 32768 | OFF | AMP | 741642.61 ± 4981.04 |\n| 65536 | OFF | AMP | 755141.25 ± 6175.05 |\n| 131072 | OFF | AMP | 744459.46 ± 8183.17 |\n| 4096 | ON | FP32 | 444532.22 ± 6239.01 |\n| 8192 | ON | FP32 | 505047.64 ± 6543.06 |\n| 16384 | ON | FP32 | 549325.54 ± 2841.21 |\n| 32768 | ON | FP32 | 587452.73 ± 2366.43 |\n| 65536 | ON | FP32 | 605187.67 ± 3740.07 |\n| 131072 | ON | FP32 | 599557.03 ± 11811.28 |\n| 4096 | ON | AMP | 541975.24 ± 4441.93 |\n| 8192 | ON | AMP | 642784.48 ± 4721.08 |\n| 16384 | ON | AMP | 727077.63 ± 5332.80 |\n| 32768 | ON | AMP | 788606.35 ± 11705.36 |\n| 65536 | ON | AMP | 832651.59 ± 10401.17 |\n| 131072 | ON | AMP | 840602.90 ± 16358.73 |\n

\n\n

\nDGX-2 XLA-ON / XLA-OFF inference speedup\n

\n\nFor each configuration of parameters present in the table, the `Speedup` column shows the speedup achieved by turning on XLA.\n\n|Batch Size |Precision |Speedup |\n|--------------------|---------------|--------|\n|4096 |TF32 |0.968 |\n|8192 |TF32 |1.012 |\n|16384 |TF32 |1.046 |\n|32768 |TF32 |1.104 |\n|65536 |TF32 |1.133 |\n|131072 |TF32 |1.126 |\n|4096 |AMP |0.932 |\n|8192 |AMP |0.967 |\n|16384 |AMP |1.384 |\n|32768 |AMP |1.063 |\n|65536 |AMP |1.103 |\n|131072 |AMP |1.129 |\n

\n\n \n\n

\nNVIDIA A100 / DGX-2 (Ampere / Volta) inference speedup\n

\n\n| Batch Size | XLA | Precision | Speedup |\n|-------------:|:------|:------------|----------:|\n| 4096 | OFF | TF32/FP32 | 1.54 |\n| 8192 | OFF | TF32/FP32 | 1.75 |\n| 16384 | OFF | TF32/FP32 | 1.79 |\n| 32768 | OFF | TF32/FP32 | 1.77 |\n| 65536 | OFF | TF32/FP32 | 1.80 |\n| 131072 | OFF | TF32/FP32 | 1.81 |\n| 4096 | OFF | AMP | 1.11 |\n| 8192 | OFF | AMP | 1.38 |\n| 16384 | OFF | AMP | 1.59 |\n| 32768 | OFF | AMP | 1.64 |\n| 65536 | OFF | AMP | 1.71 |\n| 131072 | OFF | AMP | 1.74 |\n| 4096 | ON | TF32/FP32 | 1.39 |\n| 8192 | ON | TF32/FP32 | 1.43 |\n| 16384 | ON | TF32/FP32 | 1.56 |\n| 32768 | ON | TF32/FP32 | 1.66 |\n| 65536 | ON | TF32/FP32 | 1.79 |\n| 131072 | ON | TF32/FP32 | 1.83 |\n| 4096 | ON | AMP | 1.24 |\n| 8192 | ON | AMP | 1.32 |\n| 16384 | ON | AMP | 1.45 |\n| 32768 | ON | AMP | 1.61 |\n| 65536 | ON | AMP | 1.74 |\n| 131072 | ON | AMP | 1.76 |\n

\n\n ","hasPerformance","policies",[],{"_75":201,"_77":202,"_169":-5,"_175":30,"_177":30,"_193":200,"_180":85},{"_75":199,"_77":199,"_169":-5,"_175":30,"_177":30,"_193":200,"_180":85},{"_75":21,"_77":21,"_169":-5,"_175":30,"_177":30,"_193":198,"_180":85},[],"22.01.1",[],"-","Latest (22.01.1)","nvidia/dle/sim_tf2","key","nvidia/dle/resources/sim_tf2","resourceType","resources","resourceId","sim_tf2","shortDescription","Search-based Interest Model (SIM) is a system for predicting user behavior given sequences of previous interactions.","This resource is using open-source code maintained in github (see the quick-start-guide section) and available for download from [NGC](https://catalog.ngc.nvidia.com/orgs/nvidia/teams/dle/resources/sim-tf2)\n\n\nSearch-based Interest Model (SIM) is a system for predicting user behavior given sequences of previous interactions. The model is based on [Search-based User Interest Modeling with Lifelong Sequential Behavior Data for Click-Through Rate Prediction](https://arxiv.org/abs/2006.05639) paper which reports that it has been deployed at [Alibaba](https://www.alibaba.com) in the display advertising system. This repository provides a reimplementation of the code-base provided originally for [SIM](https://github.com/tttwwy/sim) and [DIEN](https://github.com/mouna99/dien) models (SIM model's inner component).\n\nThere are several differences between this and the original SIM model implementation. First, this model is implemented in TensorFlow 2 using Python 3.8 instead of TensorFlow 1 in Python 2.7. Second, this implementation utilizes the user dimension (identifiers), which enables to train a personalized recommender system. Finally, the training code uses data preprocessed to [TFRecord](https://www.tensorflow.org/tutorials/load_data/tfrecord) format, which improves data loading. We also include scripts necessary to preprocess [Amazon Reviews dataset](https://snap.stanford.edu/data/web-Amazon.html) used in experiments.\n\nThe table below provides a fine-grained summary of the differences between this repository and the original implementation.\n\n| Mode | Original implementation | This repository |\n|----------------|---------------------------------------------------------------------------------------------------|----------------------------------------------------------------------------------------------------|\n| Python | 2.7 | 3.8 |\n| Dataset size | 135K samples | 12M samples |\n| Dataset format | CSV | TFRecord |\n| Model | - user id feature not included
- batch normalization included but not used correctly
- two-dimensional softmax output
- hardcoded features cardinalities
| - includes user id feature
- doesn`t include batch normalization
- one-dimensional sigmoid output
- features cardinalities deducted from dataset| \n\nIn the author's SIM implementation, the internals of submodels differs slightly between code and original papers (DIN, DIEN, SIM). Our implementation core is based on the paper's modules. For exact implementation details, refer to the list below.\n\n

\nList of implementation differences between original SIM code and DIN/DIEN/SIM papers\n

\n\n- Batch normalization before MLP is not included in papers.\n- Batch normalization in code used `trainable=False` during the training phase.\n- ItemItemInteraction in DIN`s attention module in SIM implementation didn't correspond to activation unit inside DIN paper.\n - Element-wise subtraction and multiplications are fed to MLP, skipping outer product operation.\n - Sigmoids are used instead of PReLU/DICE in MLP.\n- Soft search MLP is missing a middle layer in implementation.\n- In the ESU part, multi-head attention is implemented as a DIN interaction block instead of a typical multi-head attention.\n- ESU part adds additional embedding by summing all the embedding passed from the GSU part.\n- DIEN auxiliary loss uses auxiliary network instead of the sigmoid of concatenated embeddings from the DIEN paper.\n\n

\n\n \n\nThe model enables you to train a high-quality, personalized, sequential neural network-based recommender system.\n\nThis model is trained with mixed precision using Tensor Cores on NVIDIA Volta and the NVIDIA Ampere GPU architectures. Therefore, researchers can get results 1.48x 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.\n\n### Model architecture\n\nSIM model consists of two components: General Search Unit (GSU) and the Exact Search Unit (ESU). The goal of the former is to filter down possibly long historical user behavior sequence to a shorter and relevant sequence. On the other hand, ESU utilizes the most recent user behaviors for a candidate item, for example, estimate click-through rate for a candidate ad. Both parts are trained jointly using data on past user behaviors.\n\nA model architecture diagram is presented below.\n\n

\n \n
\nFigure 1. The architecture of the model.\n

\n\nEmbeddings in model architecture diagram are obtained by passing each feature from the dataset through the Embedding Layer. Item features from target item, short behavior history and long behavior history share embedding tables.\n\n

\n \n
\nFigure 2. Embedding of input features.\n

\n\n### Default configuration\n\nThe following features are implemented in this model:\n- general\n - dynamic loss scaling for Tensor Cores (mixed precision) training\n - data-parallel multi-GPU training\n- preprocessing\n - dataset preprocessing using [NVtabular](https://github.com/NVIDIA-Merlin/NVTabular) library\n\nThe following performance optimizations were implemented in this model:\n- dataloader based on [TFRecords](https://www.tensorflow.org/tutorials/load_data/tfrecord)\n \n### Feature support matrix\n\nThis model supports the following features: \n\n| Feature | SIM v1.0 TF2 |\n|----------------------------------------------------------------------------|--------------|\n| Horovod Multi-GPU (NCCL) | Yes |\n| Accelerated Linear Algebra (XLA) | Yes |\n| Automatic mixed precision (AMP) | Yes |\n| Preprocessing on GPU with NVTabular | Yes |\n| BYO dataset | Yes |\n\n#### Features\n\n**Multi-GPU training with Horovod**\nOur model uses Horovod to implement efficient multi-GPU training with NCCL. For details, refer to the example sources in this repository or refer to the [TensorFlow tutorial](https://github.com/horovod/horovod/#usage).\n\n**Accelerated Linear Algebra (XLA)**\nXLA 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.\n\n**Automatic Mixed Precision (AMP)**\nAMP enables mixed precision training without any changes to the code-base by performing automatic graph rewrites and loss scaling controlled by an environmental variable.\n\n**Preprocessing on GPU with NVTabular**\nPreprocessing on GPU with NVTabular - Amazon Reviews dataset preprocessing can be conducted using [NVTabular](https://github.com/NVIDIA-Merlin/NVTabular). For more information on the framework, refer to [this blog post](https://developer.nvidia.com/blog/announcing-the-nvtabular-open-beta-with-multi-gpu-support-and-new-data-loaders).\n\n### Mixed precision training\n\nMixed precision is the combined use of different numerical precisions in a computational method. [Mixed precision](https://arxiv.org/abs/1710.03740) 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](https://developer.nvidia.com/tensor-cores) in NVIDIA Volta, and following with 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](https://docs.nvidia.com/deeplearning/performance/mixed-precision-training/index.html) previously required two steps:\n1. Porting the model to use the FP16 data type where appropriate. \n2. Adding loss scaling to preserve small gradient values.\n\nThis can now be achieved using Automatic Mixed Precision (AMP) for TensorFlow to enable the full [mixed precision methodology](https://docs.nvidia.com/deeplearning/sdk/mixed-precision-training/index.html#tensorflow) in your existing TensorFlow model code. AMP enables mixed precision training on NVIDIA Volta, and NVIDIA Ampere GPU architectures automatically. The TensorFlow framework code makes all necessary model changes internally.\n\nIn TF-AMP, the computational graph is optimized to use as few casts as necessary and maximize the use of FP16, and the loss scaling is automatically applied inside of supported optimizers. AMP can be configured to work with the existing `tf.contrib` loss scaling manager by disabling the AMP scaling with a single environment variable to perform only the automatic mixed-precision optimization. It accomplishes this by automatically rewriting all computation graphs with the necessary operations to enable mixed precision training and automatic loss scaling.\n\nFor information about:\n- How to train using mixed precision, refer to the [Mixed Precision Training](https://arxiv.org/abs/1710.03740) paper and [Training With Mixed Precision](https://docs.nvidia.com/deeplearning/performance/mixed-precision-training/index.html) documentation.\n- Techniques used for mixed precision training, refer to the [Mixed-Precision Training of Deep Neural Networks](https://devblogs.nvidia.com/mixed-precision-training-deep-neural-networks/) blog.\n- How to access and enable AMP for TensorFlow, refer to [Using TF-AMP](https://docs.nvidia.com/deeplearning/dgx/tensorflow-user-guide/index.html#tfamp) from the TensorFlow User Guide.\n\n#### Enabling mixed precision\n\nTo enable SIM training to use mixed precision, use `--amp` flag for the training script. Refer to the [Quick Start Guide](#quick-start-guide) for more information.\n\n#### Enabling TF32\n\nTensorFloat-32 (TF32) is the new math mode in [NVIDIA A100](https://www.nvidia.com/en-us/data-center/a100/) GPUs for handling the matrix math, also called 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 NVIDIA Volta GPUs. \n\nTF32 Tensor Cores can speed up networks using FP32, typically with no loss of accuracy. It is more robust than FP16 for models which require a high dynamic range for weights or activations.\n\nFor more information, refer to the [TensorFloat-32 in the A100 GPU Accelerates AI Training, HPC up to 20x](https://blogs.nvidia.com/blog/2020/05/14/tensorfloat-32-precision-format/) blog post.\n\nTF32 is supported in the NVIDIA Ampere GPU architecture and is enabled by default.\n\n### BYO dataset functionality overview\n\nThis section describes how you can train the DeepLearningExamples RecSys models on your own datasets without changing\nthe model or data loader and with similar performance to the one published in each repository.\nThis can be achieved thanks to Dataset Feature Specification, which describes how the dataset, data loader, and model\ninteract with each other during training, inference, and evaluation.\nDataset Feature Specification has a consistent format across all recommendation models in NVIDIA's DeepLearningExamples\nrepository, regardless of dataset file type and the data loader,\ngiving you the flexibility to train RecSys models on your own datasets.\n\n- [BYO dataset glossary](#byo-dataset-glossary)\n- [Dataset Feature Specification](#dataset-feature-specification)\n- [Data Flow in Recommendation Models in DeepLearning examples](#data-flow-in-nvidia-deep-learning-examples-recommendation-models)\n- [Example of Dataset Feature Specification](#example-of-dataset-feature-specification)\n- [BYO dataset functionality](#byo-dataset-functionality)\n\n#### BYO dataset glossary\n\nThe Dataset Feature Specification consists of three mandatory and one optional section:\n\nfeature_spec provides a base of features that may be referenced in other sections, along with their metadata.\n\tFormat: dictionary (feature name) => (metadata name => metadata value)
\n\nsource_spec provides information necessary to extract features from the files that store them. \n\tFormat: dictionary (mapping name) => (list of chunks)
\n\n* Mappings are used to represent different versions of the dataset (think: train/validation/test, k-fold splits). A mapping is a list of chunks.
\n* Chunks are subsets of features that are grouped together for saving. For example, some formats may constrain data saved in one file to a single data type. In that case, each data type would correspond to at least one chunk. Another example where this might be used is to reduce file size and enable more parallel loading. Chunk description is a dictionary of three keys:
\n * type provides information about the format in which the data is stored. Not all formats are supported by all models.
\n * features is a list of features that are saved in a given chunk. The order of this list may matter: for some formats, it is crucial for assigning read data to the proper feature.
\n * files is a list of paths to files where the data is saved. For Feature Specification in yaml format, these paths are assumed to be relative to the yaml file's directory (basename). Order of this list matters: It is assumed that rows 1 to i appear in the first file, rows i+1 to j in the next one, etc.
\n\nchannel_spec determines how features are used. It is a mapping (channel name) => (list of feature names). \n\nChannels are model-specific magic constants. In general, data within a channel is processed using the same logic. Example channels: model output (labels), categorical ids, numerical inputs, user data, and item data.\n\nmetadata is a catch-all, wildcard section: If there is some information about the saved dataset that does not fit into the other sections, you can store it here.\n\n#### Dataset feature specification\n\nData flow can be described abstractly:\nInput data consists of a list of rows. Each row has the same number of columns; each column represents a feature.\nThe columns are retrieved from the input files, loaded, aggregated into channels and supplied to the model/training script. \n\nFeatureSpec contains metadata to configure this process and can be divided into three parts:\n\n* Specification of how data is organized on disk (source_spec). It describes which feature (from feature_spec) is stored in which file and how files are organized on disk.\n\n* Specification of features (feature_spec). Describes a dictionary of features, where key is the feature name and values are the features' characteristics such as dtype and other metadata (for example, cardinalities for categorical features)\n\n* Specification of model's inputs and outputs (channel_spec). Describes a dictionary of model's inputs where keys specify model channel's names and values specify lists of features to be loaded into that channel. Model's channels are groups of data streams to which common model logic is applied, for example categorical/continuous data, and user/item ids. Required/available channels depend on the model\n\nThe FeatureSpec is a common form of description regardless of underlying dataset format, dataset data loader form, and model. \n\n#### Data flow in NVIDIA Deep Learning Examples recommendation models\n\nThe typical data flow is as follows:\n* S.0. Original dataset is downloaded to a specific folder.\n* S.1. Original dataset is preprocessed into Intermediary Format. For each model, the preprocessing is done differently, using different tools. The Intermediary Format also varies (for example, for DLRM PyTorch, the Intermediary Format is a custom binary one.)\n* S.2. The Preprocessing Step outputs Intermediary Format with dataset split into training and validation/testing parts along with the Dataset Feature Specification yaml file. Metadata in the preprocessing step is automatically calculated.\n* S.3. Intermediary Format data, together with the Dataset Feature Specification, are fed into training/evaluation scripts. The data loader reads Intermediary Format and feeds the data into the model according to the description in the Dataset Feature Specification.\n* S.4. The model is trained and evaluated\n\n

\n \n
\n\nFigure 3. Data flow in Recommender models in NVIDIA Deep Learning Examples repository. Channels of the model are drawn in green.\n

\n\n#### Example of dataset feature specification\n\nAs an example, let's consider a Dataset Feature Specification for a small CSV dataset for some abstract model.\n\n```yaml\nfeature_spec:\n user_gender:\n dtype: torch.int8\n cardinality: 3 #M,F,Other\n user_age: #treated as numeric value\n dtype: torch.int8\n user_id:\n dtype: torch.int32\n cardinality: 2655\n item_id:\n dtype: torch.int32\n cardinality: 856\n label:\n dtype: torch.float32\n\nsource_spec:\n train:\n - type: csv\n features:\n - user_gender\n - user_age\n files:\n - train_data_0_0.csv\n - train_data_0_1.csv\n - type: csv\n features:\n - user_id\n - item_id\n - label\n files:\n - train_data_1.csv\n test:\n - type: csv\n features:\n - user_id\n - item_id\n - label\n - user_gender\n - user_age\n \n files:\n - test_data.csv\n\nchannel_spec:\n numeric_inputs: \n - user_age\n categorical_user_inputs: \n - user_gender\n - user_id\n categorical_item_inputs: \n - item_id\n label_ch: \n - label\n```\n\nThe data contains five features: (user_gender, user_age, user_id, item_id, label). Their data types and necessary metadata are described in the feature specification section.\n\nIn the source mapping section, two mappings are provided: one describes the layout of the training data, and the other of the testing data. The layout for training data has been chosen arbitrarily to showcase the flexibility.\nThe train mapping consists of two chunks. The first one contains user_gender and user_age, saved as a CSV, and is further broken down into two files. For specifics of the layout, refer to the following example and consult the glossary. The second chunk contains the remaining columns and is saved in a single file. Notice that the order of columns is different in the second chunk - this is alright, as long as the order matches the order in that file (that is, columns in the .csv are also switched)\n\nLet's break down the train source mapping. The table contains example data color-paired to the files containing it.\n\n

\n\n

\n\nThe channel spec describes how the data will be consumed. Four streams will be produced and available to the script/model.\nThe feature specification does not specify what happens further: names of these streams are only lookup constants defined by the model/script.\nBased on this example, we can speculate that the model has three input channels: numeric_inputs, categorical_user_inputs,\ncategorical_item_inputs, and one output channel: label.\nFeature names are internal to the FeatureSpec and can be freely modified.\n\n#### BYO dataset functionality\n\nIn order to train any Recommendation model in NVIDIA Deep Learning Examples, one can follow one of three possible ways:\n* One delivers preprocessed datasets in the Intermediary Format supported by data loader used by the training script\n(different models use different data loaders) together with FeatureSpec yaml file describing at least specification of dataset, features, and model channels\n\n* One uses a transcoding script (**not supported in SIM model yet**)\n\n* One delivers datasets in non-preprocessed form and uses preprocessing scripts that are a part of the model repository.\nIn order to use already existing preprocessing scripts, the format of the dataset needs to match one of the original datasets.\nThis way, the FeatureSpec file will be generated automatically, but the user will have the same preprocessing as in the original model repository. \n\n### Glossary\n\n**Auxiliary loss** is used to improve DIEN (so SIM as well) model training. It is constructed based on consecutive user actions from their short behavior history.\n\n**DIEN model** was proposed in [Deep Interest Evolution Network for Click-Through Rate Prediction](https://arxiv.org/abs/1809.03672) paper as an extension of the DIN model. It can also be used as a backbone for processing short interaction sequences in the SIM model.\n\n**DIN model** was proposed in [Deep Interest Network for Click-Through Rate Prediction](https://arxiv.org/abs/1706.06978) paper. It can be used as a backbone for processing short interaction sequences in the SIM model.\n\n**Long user behavior history** is the record of past user interactions. They are processed by the General Search Unit part of the SIM model (refer to Figure 1). This typically is a lightweight model aimed at processing longer sequences.\n\n**Short user behavior history** is the record of the most recent user interactions. They are processed by a more computationally intensive Exact Search Unit part of the SIM model (refer to Figure 1).\n\n**User behaviors** are users' interactions with given items of interest. Example interactions include reviewed items for Amazon Reviews dataset or clicks in the e-commerce domain. All the systems contained in this repository focus on modeling user interactions.","displayName","SIM for TensorFlow2",["D",1680631901616],"publisher","NVIDIA Deep Learning Examples","latestVersion","versions","collectionCounts","imageUrl","/api/image/nvidia/dle/resources/sim_tf2","rawImageUrl","https://assets.nvidiagrid.net/ngc/logos/Deep-Learning-Examples.png","canGuestDownload","latestVersionSizeInBytes",74736,"licenseTerms","isMultinodeEnabled","isMultiArchitecture","bias","explainability","privacy","safetyAndSecurity","featureLabels",[],"displayLabels",[246],"policyLabels",[],"systemLabels",[],"products",[],"hasNimMetadata",{"_213":247,"_248":249,"_250":251,"_252":253},"Recommendation","value","uscs_recommedation","filter",{"_254":255,"_248":249},"category","label","field","application",{"_106":1236,"_108":1237},"ngcApps",[1189],"csaApiBaseUrl","https://www.nvidia.com/ai-agents/csa-agent/dcsm","datadogRum",{"_1170":1171,"_1172":1173,"_1174":1175,"_1176":1177,"_1178":267,"_138":293,"_1179":1180,"_1181":55,"_1182":85,"_1183":85,"_1184":1185,"_1186":85,"_1187":1188},"amplitude",{"_1167":1168,"_807":-7,"_1169":30},"disableJs","environment","prod","featureFlagsBlob","JTdCJTIydGVzdEZsYWclMjIlM0FmYWxzZSUyQyUyMnNob3dTZWFyY2hMYWJlbFNldEZpbHRlcnMlMjIlM0FmYWxzZSUyQyUyMmVuYWJsZUNzYVdpZGdldCUyMiUzQWZhbHNlJTJDJTIyZW5hYmxlUHJvZHVjdEdyb3VwaW5nQnlTdWJzY3JpcHRpb24lMjIlM0FmYWxzZSUyQyUyMmNvbnRhaW5lclNjYW5uaW5nUmVkZXNpZ25NMiUyMiUzQWZhbHNlJTdE","featureFlagsGuarded","isDevLikeEnv","isPlaywright","isMobileDevice","categorizedOrgs",{"_1161":1162,"_1163":1164,"_1165":1166},"currentOrgContext","currentContextRoles",[],"enterpriseOrgs",[],"manageableExternalUserGroups",[],"ngcApiOrigin","ngcHelmOrigin","origin","https://catalog.ngc.nvidia.com","usesMocks","theme","year",2026,"now",1787757154912,"1.12.0","defaultSearchSuggestions",{"_1047":374,"_1048":1049,"_1050":1051,"_1052":1053},"announcements",[],"consentState",{"_331":1037,"_1038":85,"_1039":30,"_1040":30,"_1041":1042},"consentConfig",{"_313":314,"_315":316,"_317":318,"_319":320,"_321":322,"_323":324,"_325":326,"_327":328,"_329":330,"_331":332},"showContactAdmin",1787757154913,[],[],[307,308,309,310,311,312],{"_44":59,"_46":69,"_48":101,"_50":-5,"_52":-5,"_54":55},{"_44":59,"_46":68,"_48":101,"_50":-5,"_52":-5,"_54":55},{"_44":59,"_46":66,"_48":101,"_50":-5,"_52":-5,"_54":55},{"_44":59,"_46":60,"_48":101,"_50":-5,"_52":-5,"_54":55},{"_44":56,"_46":57,"_48":101,"_50":-5,"_52":-5,"_54":55},{"_44":45,"_46":47,"_48":101,"_50":-5,"_52":-5,"_54":55},"mainInfoText","We and our third-party partners (including social media, advertising, and analytics partners) use cookies and other tracking technologies to collect, store, monitor, and process certain information about you when you visit our website. The information collected might relate to you, your preferences, or your device. We use that information to make the site work, analyze performance and traffic on our website, provide a more personalized web experience, and assist in our marketing efforts.

Under certain privacy laws, you have the right to direct us not to \"sell\" or \"share\" your personal information for targeted advertising. To opt-out of the \"sale\" and \"sharing\" of personal information through cookies, you must opt-out of optional cookies using the toggles below. To opt out of the \"sale\" and \"sharing\" of data collected by other means (e.g., online forms) you must also update your data sharing preferences through the NVIDIA Preference Center.

Click on the different category headings below to find out more and change the settings according to your preference. You cannot opt out of Required Cookies as they are deployed to ensure the proper functioning of our website (such as prompting the cookie banner and remembering your settings, etc.). By clicking \"Save and Accept\" or \"Decline All\" at the bottom, you consent to the use of cookies and other tools as described in our Cookie Policy in accordance with your settings and accept our Terms of Service (which contains important waivers). For more information about our privacy practices, please see our Privacy Policy.","mainText","Cookie Settings","alertNoticeText",{"_1029":1030,"_1031":1032,"_1033":1034,"_1035":1036},"bannerAcceptLabel","Accept All","bannerRejectLabel","Reject Optional","bannerSettingsLabel","Manage Settings","preferenceSaveLabel","Save Choices","preferenceRejectLabel","Decline All","alwaysActiveText","Always Active","groups",[333,334,335,336,337],{"_338":719,"_77":720,"_167":721,"_342":85,"_343":16,"_345":722,"_347":723},{"_338":622,"_77":623,"_167":624,"_342":30,"_343":58,"_345":625,"_347":626},{"_338":529,"_77":530,"_167":531,"_342":30,"_343":532,"_345":533,"_347":534},{"_338":349,"_77":350,"_167":351,"_342":30,"_343":352,"_345":353,"_347":354},{"_338":339,"_77":340,"_167":341,"_342":30,"_343":344,"_345":346,"_347":348},"optanonGroupId","C0005","Social Media Cookies","These cookies are set by a range of social media services that we have added to the site to enable you to share our content with your friends and networks. They are capable of tracking your browser across other sites and building up a profile of your interests. This may impact the content and messages you see on other websites you visit. If you do not allow these cookies you may not be able to use or see these sharing tools.","alwaysActive","order",11,"cookies",[],"hosts",[],"C0004","Advertising Cookies","These cookies record your visit to our websites, the pages you have visited and the links you have followed to influence the advertisements that you see on other websites. These cookies and the information they collect may be managed by other companies, including our advertising partners, and may be used to build a profile of your interests and show you relevant advertising on other sites. We and our advertising partners will use this information to make our websites and the advertising displayed on it, more relevant to your interests.",5,[471,472,473,474,475,476,477,478,479,480,481,482,483,484,485,486,487,488,489,490,491,492],[355,356,357,358,359,360,361,362,363,364,365,366,367,368,369,370,371],{"_372":466,"_213":466,"_167":374,"_345":467},{"_372":462,"_213":462,"_167":374,"_345":463},{"_372":457,"_213":457,"_167":374,"_345":458},{"_372":453,"_213":453,"_167":374,"_345":454},{"_372":448,"_213":448,"_167":374,"_345":449},{"_372":443,"_213":443,"_167":374,"_345":444},{"_372":435,"_213":435,"_167":374,"_345":436},{"_372":427,"_213":427,"_167":374,"_345":428},{"_372":422,"_213":422,"_167":374,"_345":423},{"_372":417,"_213":417,"_167":374,"_345":418},{"_372":412,"_213":412,"_167":374,"_345":413},{"_372":407,"_213":407,"_167":374,"_345":408},{"_372":402,"_213":402,"_167":374,"_345":403},{"_372":395,"_213":395,"_167":374,"_345":396},{"_372":390,"_213":390,"_167":374,"_345":391},{"_372":383,"_213":383,"_167":374,"_345":384},{"_372":373,"_213":373,"_167":374,"_345":375},"hostName","amazon-adsystem.com","",[376,377],{"_77":382,"_379":373,"_380":30,"_167":381},{"_77":378,"_379":373,"_380":30,"_167":381},"ad-privacy","host","isSession","This domain is owned by online retailer Amazon and is used as part of its affiliate marketing programme.","ad-id","doubleclick.net",[385,386],{"_77":389,"_379":383,"_380":30,"_167":388},{"_77":387,"_379":383,"_380":30,"_167":388},"IDE","This domain is owned by Doubleclick (Google). The main business activity is: Doubleclick is Googles real time bidding advertising exchange","test_cookie","bing.com",[392],{"_77":393,"_379":390,"_380":30,"_167":394},"MUID","This domain is owned by Mircosoft - it is the site for the search engine Bing.","assets.adobedtm.com",[397,398],{"_77":401,"_379":395,"_380":30,"_167":400},{"_77":399,"_379":395,"_380":30,"_167":400},"kndctr_F207D74D549850760A4C98C6_AdobeOrg_*","If you know anything about this host and its uses please add your contributions here. ","kndctr_F207D74D549850760A4C98C6_AdobeOrg_consent","tiktok.com",[404],{"_77":405,"_379":402,"_380":30,"_167":406},"_ttp","TikTok is a social media platform owned by ByteDance, known for short-form video sharing. The service is used to create and share videos, and it utilizes cookies for various purposes including advertising and analytics.","t.co",[409],{"_77":410,"_379":407,"_380":30,"_167":411},"muc_ads","This domain is owned by Twitter. It is used as a URL shortening service, which allows Twitter to collect data on how links are shared.","snowplow.com",[414],{"_77":415,"_379":412,"_380":30,"_167":416},"sp","Testing","google.com",[419],{"_77":420,"_379":417,"_380":30,"_167":421},"NID","This domain is owned by Google Inc. Although Google is primarily known as a search engine, the company provides a diverse range of products and services. Its main source of revenue however is advertising. Google tracks users extensively both through its own products and sites, and the numerous technologies embedded into many millions of websites around the world. It uses the data gathered from most of these services to profile the interests of web users and sell advertising space to organisations based on such interest profiles as well as aligning adverts to the content on the pages where its customer's adverts appear.","linkedin.com",[424],{"_77":425,"_379":422,"_380":30,"_167":426},"bcookie","This domain is owned by LinkedIn, the business networking platform. It typically acts as a third party host where website owners have placed one of its content sharing buttons in their pages, although its content and services can be embedded in other ways. Although such buttons add functionality to the website they are on, cookies are set regardless of whether or not the visitor has an active Linkedin profile, or agreed to their terms and conditions. For this reason it is classified as a primarily tracking/targeting domain.","twitter.com",[429,430],{"_77":433,"_379":427,"_380":30,"_167":434},{"_77":431,"_379":427,"_380":30,"_167":432},"personalization_id","This domain is owned by Twitter. The main business activity is: Social Networking Services. Where twitter acts as a third party host, it collects data through a range of plug-ins and integrations, that is primarily used for tracking and targeting.","guest_*","This domain is owned by X (formerly Twitter). The main business activity is: Social Networking Services. Where X (formerly Twitter) acts as a third party host, it collects data through a range of plug-ins and integrations, that is primarily used for tracking and targeting.","youtube.com",[437,438],{"_77":441,"_379":435,"_380":30,"_167":442},{"_77":439,"_379":435,"_380":30,"_167":440},"VISITOR_INFO1_LIVE","This cookie is used as a unique identifier to track viewing of videos","__Secure-*","YouTube is a Google owned platform for hosting and sharing videos. YouTube collects user data through videos embedded in websites, which is aggregated with profile data from other Google services in order to display targeted advertising to web visitors across a broad range of their own and other websites.","demdex.net",[445],{"_77":446,"_379":443,"_380":30,"_167":447},"demdex","This cookie helps Adobe Audience Manger perform basic functions such as visitor identification, ID synchronization, segmentation, modeling, reporting, etc.","im-apps.net",[450],{"_77":451,"_379":448,"_380":30,"_167":452},"receive-cookie-deprecation","This domain is owned by im-apps.net which provides information and resources about various instant messaging (IM) applications to help users make informed decisions based on their communication needs.","influ2.com",[455],{"_77":456,"_379":453,"_380":30,"_167":416},"R","6sc.co",[459],{"_77":460,"_379":457,"_380":30,"_167":461},"6suuid","This domain is associated with the company 6Sense Insights, Inc., which provides predictive analytics and account engagement services. It is primarily used for tracking and targeting in marketing automation and advertising.","hushly.com",[464],{"_77":465,"_379":462,"_380":30,"_167":416},"_hly_gvid","t.contentsquare.net",[468],{"_77":469,"_379":466,"_380":85,"_167":470},"_cs_*","This is a session cookie. This is a pattern type cookie where a random 13-digit number is appended to the prefix _cs_.",{"_77":527,"_379":494,"_380":30,"_167":528},{"_77":525,"_379":494,"_380":30,"_167":526},{"_77":523,"_379":494,"_380":30,"_167":524},{"_77":405,"_379":494,"_380":30,"_167":522},{"_77":499,"_379":494,"_380":85,"_167":501},{"_77":521,"_379":494,"_380":30,"_167":374},{"_77":520,"_379":497,"_380":30,"_167":374},{"_77":518,"_379":494,"_380":30,"_167":519},{"_77":516,"_379":494,"_380":30,"_167":517},{"_77":515,"_379":494,"_380":30,"_167":374},{"_77":513,"_379":494,"_380":30,"_167":514},{"_77":499,"_379":512,"_380":85,"_167":501},{"_77":510,"_379":497,"_380":30,"_167":511},{"_77":510,"_379":500,"_380":30,"_167":511},{"_77":469,"_379":494,"_380":30,"_167":470},{"_77":508,"_379":494,"_380":30,"_167":509},{"_77":415,"_379":506,"_380":30,"_167":507},{"_77":504,"_379":494,"_380":30,"_167":505},{"_77":502,"_379":494,"_380":30,"_167":503},{"_77":499,"_379":500,"_380":85,"_167":501},{"_77":496,"_379":497,"_380":30,"_167":498},{"_77":493,"_379":494,"_380":85,"_167":495},"AMCVS_xxxxxAdobeOrg","nvidia.com","This is a pattern type cookie name associated with Adobe Marketing Cloud. It stores a unique visitor identifier, and uses an organisation identifier to allow a company to track users across their domains and services.This is a pattern type cookie name associated with Adobe Marketing Cloud. It stores a unique visitor identifier, and uses an organisation identifier.","_rdt_pn","www.nvidia.com","This cookie name is associated with www.mind.org.uk which is used for marketing purposes.","AMP_*","build.nvidia.com","Supports Amplitude product analytics.","_gcl_au","Used by Google AdSense for experimenting with advertisement efficiency across websites using their services","_mkto_trk","This cookie is associated with an email marketing service provided by Marketo. This tracking cookie allows a website to link visitor behaviour to the recipient of an email marketing campaign, to measure campaign effectiveness.","investor.nvidia.com","This cookie is associated with Eventbrite and is used to deliver content tailored to the end user's interests and improve content creation. This cookie is also used for event-booking purposes.","_gat_gtag_UA_140825383_2","Google Analytics","_rdt_uuid","This cookie is set by Reddit and is used for tracking the views of embedded videos and performance of the advertisements.","catalog.ngc.nvidia.com","_uetsid","This cookie is used by Bing to determine what ads should be shown that may be relevant to the end user perusing the site. ","kndctr_F207D74D549850760A4C98C6_AdobeOrg_bc_session_id","ttcsid_XXXXXXXXXXXXXXXXXXXXCategory","This cookie name is associated with TikTok which is used to track your activity on the platform, including how you interact with their embedded services on other websites.This cookie is associated with TikTok. It helps track your activity on the platform, including how you interact with their embedded services on other websites.","_fbp","Used by Facebook to deliver a series of advertisement products such as real time bidding from third party advertisers","_zitok","_twpid","This cookie is typically associated with the use of Taptap Digital, a company specializing in digital advertising. It is used for tracking and ad personalization purposes.","_tt_enable_cookie","This cookie name is associated with TikTok which is used to collect data about behavior and purchases on the website and to measure the effect of advertising. It is also used to track website activity to help optimize advertising efforts.","AMCV_xxxxxAdobeOrg","This cookie is used to identify a unique visitor for tracking purposes by Adobe Experience Cloud.This is a pattern type cookie name associated with Adobe Marketing Cloud. It stores a unique visitor identifier, and uses an organisation identifier to allow a company to track users across their domains and services.","ttcsid","This cookie name is associated with TikTok which is used to enable TikTok Ads session tracking by assigning a unique session ID to users who click on TikTok ads, allowing for the attribution of user behavior and conversions within the same session.This cookie is associated with TikTok. It helps track your activity on the platform, including how you interact with their embedded services on other websites.","C0003","Personalization Cookies","These cookies collect data about how you have interacted with our website to help us improve your web experience, such as which pages you have visited. These cookies may store a unique ID so that our system will remember you when you return. They may be set by us or by third party providers whose services we have added to our pages. These cookies enable us to provide enhanced website functionality and personalization as well as make the marketing messages we send to you more relevant to your interests. If you do not allow these cookies, then some or all of these services may not function properly.",4,[564,565,566,567,568,569,570,571,572,573,574,575,576,577,578,579,580,581,582,583,584,585,586,587,588],[535,536,537,538,539],{"_372":448,"_213":448,"_167":374,"_345":559},{"_372":422,"_213":422,"_167":374,"_345":554},{"_372":549,"_213":549,"_167":374,"_345":550},{"_372":544,"_213":544,"_167":374,"_345":545},{"_372":540,"_213":540,"_167":374,"_345":541},"nvidia.partners",[542],{"_77":543,"_379":540,"_380":30,"_167":416},"edge","app.hushly.com",[546],{"_77":547,"_379":544,"_380":30,"_167":548},"*CORS","This domain is associated with Hushly, a platform that provides lead conversion solutions for content marketing by using personalization and content engagement strategies.","github.com",[551],{"_77":552,"_379":549,"_380":30,"_167":553},"_octo","This domain is owned by Github. The main business activity is: Source code repository and collaboration platform",[555,556],{"_77":558,"_379":422,"_380":30,"_167":426},{"_77":557,"_379":422,"_380":30,"_167":426},"lidc","li_gc",[560,561],{"_77":563,"_379":448,"_380":30,"_167":452},{"_77":562,"_379":448,"_380":30,"_167":452},"imid_*","imid",{"_77":589,"_379":497,"_380":30,"_167":591},{"_77":620,"_379":494,"_380":30,"_167":621},{"_77":589,"_379":619,"_380":30,"_167":591},{"_77":617,"_379":494,"_380":85,"_167":618},{"_77":615,"_379":494,"_380":30,"_167":616},{"_77":614,"_379":597,"_380":85,"_167":374},{"_77":589,"_379":593,"_380":30,"_167":591},{"_77":612,"_379":494,"_380":30,"_167":613},{"_77":589,"_379":500,"_380":30,"_167":591},{"_77":600,"_379":497,"_380":30,"_167":374},{"_77":589,"_379":611,"_380":30,"_167":591},{"_77":609,"_379":494,"_380":30,"_167":610},{"_77":606,"_379":607,"_380":30,"_167":608},{"_77":604,"_379":605,"_380":30,"_167":374},{"_77":589,"_379":601,"_380":30,"_167":591},{"_77":603,"_379":494,"_380":30,"_167":374},{"_77":589,"_379":602,"_380":30,"_167":591},{"_77":600,"_379":601,"_380":30,"_167":374},{"_77":599,"_379":597,"_380":85,"_167":374},{"_77":589,"_379":598,"_380":30,"_167":591},{"_77":589,"_379":597,"_380":30,"_167":591},{"_77":589,"_379":596,"_380":30,"_167":591},{"_77":595,"_379":494,"_380":85,"_167":591},{"_77":592,"_379":593,"_380":85,"_167":594},{"_77":589,"_379":590,"_380":30,"_167":591},"_gd_*","info.nvidia.com","This cookie contains no identifying information and is used for diagnostic purposes by Impact Radius, a service we use to track signup commissions for our affiliate program. We have categorized it under Functionality because it is necessary for the operation of our affiliate program.","nv_country_code","developer.nvidia.com","This cookie is likely used to store the country code of the visitor to provide localized content and possibly for legal compliance with regional data privacy laws.","_gdxxxxxxxxxxxxx","nvidianews.nvidia.com","docs.api.nvidia.com","research.nvidia.com","readme_library","wcmmode","ai-podcast.nvidia.com","resources.nvidia.com","geo_locator","openedx-language-preference","learn.nvidia.com","doxygen_width","raytracing-docs.nvidia.com","This cookie is likely used to remember the width preferences of a sidebar or interface element within a documentation platform like Doxygen.","_sfid_eb78","This cookie is categorized as Functionality Cookie.","crowdsource.nvidia.com","mbox","Cookie set by Adobe SiteCatalyst analytics software - used for measuring the performance of page content using A/B split testing","readme_language","vid","This cookie name 'vid' is used for tracking the search request made by the user on the 'Find a practitioner' page.","at_check","Used by Adobe Target to check if cookies are enabled/supported on the browser","blogs.nvidia.com","nvHeadless","This cookie is likely used to track if a user is using the site in a 'headless' non-GUI mode, which could be for accessibility reasons or for automated processes.","C0002","Performance Cookies","These cookies are used to provide quantitative measures of our website visitors, such as the number of times you visit, time on page, your mouse movements, scrolling, clicks and keystroke activity on the websites; other browsing, search, or product research behavior; and what brought you to our site. These cookies may store a unique ID so that our system will remember you when you return. Information collected with these cookies is used to measure and find ways to improve website performance.",[647,648,649,650,651,652,653,654,655,656,657,658,659,660,661,662,663,664,665,666,667,668,669,670,671,672,673,674,675],[627,628,629,630],{"_372":642,"_213":642,"_167":374,"_345":643},{"_372":637,"_213":637,"_167":374,"_345":638},{"_372":435,"_213":435,"_167":374,"_345":634},{"_372":549,"_213":549,"_167":374,"_345":631},[632],{"_77":633,"_379":549,"_380":85,"_167":553},"_gh_sess",[635],{"_77":636,"_379":435,"_380":85,"_167":442},"YSC","bizible.com",[639],{"_77":640,"_379":637,"_380":30,"_167":641},"_BUID","Bizible is a marketing attribution software that tracks the performance of marketing efforts and connects them to revenue, primarily used for B2B environments.","a.usbrowserspeed.com",[644],{"_77":645,"_379":642,"_380":30,"_167":646},"tuid","This domain is owned by Ookla which provides internet speed testing services through its Speedtest application.",{"_77":717,"_379":494,"_380":30,"_167":718},{"_77":702,"_379":716,"_380":85,"_167":703},{"_77":685,"_379":497,"_380":30,"_167":610},{"_77":685,"_379":598,"_380":30,"_167":610},{"_77":714,"_379":494,"_380":30,"_167":715},{"_77":702,"_379":500,"_380":85,"_167":703},{"_77":712,"_379":494,"_380":30,"_167":713},{"_77":702,"_379":497,"_380":85,"_167":703},{"_77":710,"_379":597,"_380":30,"_167":711},{"_77":702,"_379":619,"_380":85,"_167":703},{"_77":682,"_379":497,"_380":30,"_167":684},{"_77":708,"_379":494,"_380":85,"_167":709},{"_77":706,"_379":494,"_380":30,"_167":707},{"_77":704,"_379":494,"_380":85,"_167":705},{"_77":702,"_379":593,"_380":85,"_167":703},{"_77":700,"_379":494,"_380":30,"_167":701},{"_77":698,"_379":494,"_380":30,"_167":699},{"_77":696,"_379":494,"_380":85,"_167":697},{"_77":694,"_379":602,"_380":30,"_167":695},{"_77":682,"_379":593,"_380":30,"_167":684},{"_77":692,"_379":494,"_380":30,"_167":693},{"_77":690,"_379":494,"_380":85,"_167":691},{"_77":688,"_379":494,"_380":30,"_167":689},{"_77":686,"_379":494,"_380":30,"_167":687},{"_77":685,"_379":494,"_380":30,"_167":610},{"_77":682,"_379":683,"_380":30,"_167":684},{"_77":680,"_379":494,"_380":30,"_167":681},{"_77":678,"_379":494,"_380":85,"_167":679},{"_77":676,"_379":494,"_380":30,"_167":677},"_ga_xxxxxxxxxx","Used by Google Analytics to identify and track an individual session with your device.","s_ppvl","This cookie is set by Adobe Analytics to store information on the percentage of the page displayed to the end user.","coveo_visitorId","This cookie name is associated with Coveo which is used to store a unique user ID for marketing purposes and to track the number of distinct users who submitted a query or clicked an item, counting them only once even if they did the same query or clicked the same item over the course of many visits.","_dd_s","marketplace.nvidia.com","Datadog Cookie used to group all events generated from a unique user session across multiple pages. It contains the current session ID, whether the session is excluded due to sampling, and the expiration date of the session. The cookie is extended for an extra 15 minutes every time the user interacts with the website, up to the maximum user session duration (4 hours).","nvweb_A","_ga","This cookie name is associated with Google Universal Analytics - which is a significant update to Google's more commonly used analytics service. This cookie is used to distinguish unique users by assigning a randomly generated number as a client identifier. It is included in each page request in a site and used to calculate visitor, session and campaign data for the sites analytics reports. By default it is set to expire after 2 years, although this is customisable by website owners.","_uetvid","This is a cookie utilised by Microsoft Bing Ads and is a tracking cookie. It allows us to engage with a user that has previously visited our website.","tp","This cookie is associated with Adobe Analytics site tracking. It registers if the website's scroll-depth detection is active. This function remembers how far the user has scrolled on the current session across the website's sub-pages.","_hly_vid","This cookie _hly_vid is used to record website and server performance with each visit to the site. The data is aggregated and analyzed; therefore, it remains anonymous upon use.","_session_id","This cookie creates an interim session ID used as an in-session user ID.\n","s_cc","This cookie is associated with the Adobe Site Catalyst. It determines whether cookies are enabled in the web browser.","s_ecid","Adobe Analytics cookie containing the the Adobe Experience Cloud 1st party visitor id if the Analytics AMCV visitor id is not found.","_hjSessionUser_xxxxxx","Hotjar cookie that is set when a user first lands on a page with the Hotjar script. It is used to persist the Hotjar User ID, unique to that site on the browser. This ensures that behavior in subsequent visits to the same site will be attributed to the same user ID.","_hjHasCachedUserAttributes","This cookie name is associated with Hotjar which is used to know whether the data set in _hjUserAttributes Local Storage item is up to date or not.","s_ppv","This cookie is associated with Adobe Analytics site tracking and is used to measure a user's scroll activity to see how much of a page they view before moving to another page.","_gid","This cookie name is associated with Google Universal Analytics. This appears to be a new cookie and as of Spring 2017 no information is available from Google. It appears to store and update a unique value for each page visited.","_hjTLDTest","When the Hotjar script executes we try to determine the most generic cookie path we should use, instead of the page hostname. This is done so that cookies can be shared across subdomains (where applicable). To determine this, we try to store the _hjTLDTest cookie for different URL substring alternatives until it fails. After this check, the cookie is removed.","ekfls-2025-03-27","This cookie name is associated with readme.io which is used to help maintain data for the API reference support page.","s_getNewRepeat","This cookie is used to determine whether a visitor is a new visitor or a repeat visitor to the site, and is used for analytical purposes.","_hly_sid","This cookie is likely associated with website analytics, as the prefix '_hly_' suggests a connection to Holistics, a business intelligence and analytics platform.","docs.nvidia.com","_hjSession_xxxxxx","A cookie that holds the current session data. This ensues that subsequent requests within the session window will be attributed to the same Hotjar session.","C0001","Required Cookies","These cookies enable core functionality such as security, network management, and accessibility. These cookies are required for the site to function and cannot be turned off.",[864,865,866,867,868,869,870,871,872,873,874,875,876,877,878,879,880,881,882,883,884,885,886,887,888,889,890,891,892,893,894,895,896,897,898,899,900,901,902,903,904,905,906,907,908,909,910,911,912,913,914,915,916,917,918,919,920,921,922,923,924,925,926,927,928,929,930,931,932,933,934,935,936],[724,725,726,727,728,729,730,731,732,733,734,735,736,737,738,739,740,741,742,743,744,745,746,747,748],{"_372":861,"_213":861,"_167":374,"_345":862},{"_372":855,"_213":855,"_167":374,"_345":856},{"_372":852,"_213":852,"_167":374,"_345":853},{"_372":840,"_213":840,"_167":374,"_345":841},{"_372":837,"_213":837,"_167":374,"_345":838},{"_372":834,"_213":834,"_167":374,"_345":835},{"_372":829,"_213":829,"_167":374,"_345":830},{"_372":435,"_213":435,"_167":374,"_345":826},{"_372":823,"_213":823,"_167":374,"_345":824},{"_372":427,"_213":427,"_167":374,"_345":821},{"_372":422,"_213":422,"_167":374,"_345":819},{"_372":814,"_213":814,"_167":374,"_345":815},{"_372":809,"_213":809,"_167":374,"_345":810},{"_372":804,"_213":804,"_167":374,"_345":805},{"_372":801,"_213":801,"_167":374,"_345":802},{"_372":549,"_213":549,"_167":374,"_345":798},{"_372":407,"_213":407,"_167":374,"_345":795},{"_372":544,"_213":544,"_167":374,"_345":788},{"_372":540,"_213":540,"_167":374,"_345":782},{"_372":776,"_213":776,"_167":374,"_345":777},{"_372":771,"_213":771,"_167":374,"_345":772},{"_372":766,"_213":766,"_167":374,"_345":767},{"_372":762,"_213":762,"_167":374,"_345":763},{"_372":757,"_213":757,"_167":374,"_345":758},{"_372":749,"_213":749,"_167":374,"_345":750},"polyfill.io",[751,752],{"_77":755,"_379":749,"_380":30,"_167":756},{"_77":753,"_379":749,"_380":30,"_167":754},"cf_ob_info","This cookie is associated with sites using CloudFlare. It is used to improve page load times and to override any security restrictions based on the visitors IP address. It does not contain any user identification information.","cf_use_ob","This cookie is associated with sites using CloudFlare. It is used to improve page load times and to override any security restrictions based on the visitors IP address. It does not contain any user identification information. ","www.recaptcha.net",[759],{"_77":760,"_379":757,"_380":30,"_167":761},"_GRECAPTCHA","This domain is used by Google reCAPTCHA, which provides spam protection services to websites by detecting if a user is a human or a bot.","api2.hcaptcha.com",[764],{"_77":765,"_379":762,"_380":30,"_167":416},"__cflb","instagram.com",[768],{"_77":769,"_379":766,"_380":30,"_167":770},"csrftoken","Instagram is a social networking service owned by Facebook, Inc. It's designed for sharing photos and videos from a smartphone. The service also uses cookies for various purposes, including analytics, personalization, and advertising.","readme.io",[773],{"_77":774,"_379":771,"_380":30,"_167":775},"__cf_bm","The __cf_bm cookie is a cookie necessary to support Cloudflare Bot Management, currently in private beta. As part of our bot management service, this cookie helps manage incoming traffic that matches criteria associated with bots.","nvidiacorp.us-5.evergage.com",[778,779],{"_77":781,"_379":776,"_380":30,"_167":416},{"_77":780,"_379":776,"_380":30,"_167":416},"AWSALBTG","AWSALBTGCORS",[783,784],{"_77":786,"_379":540,"_380":30,"_167":787},{"_77":785,"_379":540,"_380":30,"_167":416},"ak_bmsc","bm_mi","This domain is likely associated with NVIDIA's partner program, which could involve marketing, sales, and support services for partners.",[789,790,791],{"_77":794,"_379":544,"_380":30,"_167":416},{"_77":793,"_379":544,"_380":85,"_167":416},{"_77":792,"_379":544,"_380":30,"_167":416},"AWSALBCORS","JSESSIONID","AWSALB",[796],{"_77":774,"_379":407,"_380":30,"_167":797},"This is a CloudFoundry cookie",[799],{"_77":800,"_379":549,"_380":30,"_167":553},"logged_in","hcaptcha.com",[803],{"_77":774,"_379":801,"_380":30,"_167":797},"tracking.intentsify.io",[806],{"_77":807,"_379":804,"_380":30,"_167":808},"userId","Used to track usage of the embedded video player.","static.reo.dev",[811],{"_77":812,"_379":809,"_380":30,"_167":813},"cookietest","Common cookie name could have a number of different origins. Where this is first party and a session cookie, its most likely to do with checking to see if the browser is set to block or allow cookies.","www.datadoghq-browser-agent.com",[816],{"_77":817,"_379":814,"_380":30,"_167":818},"dd_cookie_test_","dd_cookie_test",[820],{"_77":774,"_379":422,"_380":30,"_167":797},[822],{"_77":774,"_379":427,"_380":30,"_167":797},"supabase.co",[825],{"_77":774,"_379":823,"_380":30,"_167":797},[827],{"_77":828,"_379":435,"_380":30,"_167":442},"VISITOR_PRIVACY_METADATA","dosa-prd.azurewebsites.net",[831],{"_77":832,"_379":829,"_380":30,"_167":833},"access_token","This cookie is used to obtain the permissions for users of the portal to allow for API access in order to retrieve information required to perform portal functions.","w.hcaptcha.com",[836],{"_77":774,"_379":834,"_380":30,"_167":797},"cdn.readme.io",[839],{"_77":774,"_379":837,"_380":30,"_167":797},"queue-it.net",[842,843,844,845,846],{"_77":851,"_379":840,"_380":30,"_167":374},{"_77":850,"_379":840,"_380":30,"_167":374},{"_77":849,"_379":840,"_380":30,"_167":374},{"_77":848,"_379":840,"_380":30,"_167":374},{"_77":847,"_379":840,"_380":30,"_167":374},"Queue-it-{queueId}","Queue-it-{dmonvidia }________{eventId}","Queue-it-token","queueitsoftblock_{queueId}","Queue-it-{nvidia}________{eventId}","x.com",[854],{"_77":774,"_379":852,"_380":30,"_167":775},"api.hcaptcha.com",[857,858],{"_77":860,"_379":855,"_380":30,"_167":416},{"_77":765,"_379":855,"_380":30,"_167":859},"This domain is associated with hCaptcha, a service that provides CAPTCHA and reCAPTCHA alternatives for user verification and bot protection.","hmt_id","vimeo.com",[863],{"_77":774,"_379":861,"_380":30,"_167":797},{"_77":997,"_379":494,"_380":30,"_167":1028},{"_77":1026,"_379":590,"_380":85,"_167":1027},{"_77":1019,"_379":988,"_380":85,"_167":1020},{"_77":1025,"_379":683,"_380":85,"_167":374},{"_77":1023,"_379":494,"_380":30,"_167":1024},{"_77":781,"_379":1009,"_380":30,"_167":948},{"_77":956,"_379":1018,"_380":85,"_167":958},{"_77":780,"_379":947,"_380":30,"_167":948},{"_77":959,"_379":1018,"_380":30,"_167":960},{"_77":1021,"_379":494,"_380":85,"_167":1022},{"_77":1019,"_379":995,"_380":85,"_167":1020},{"_77":953,"_379":1018,"_380":30,"_167":955},{"_77":1016,"_379":494,"_380":30,"_167":1017},{"_77":780,"_379":939,"_380":30,"_167":948},{"_77":944,"_379":1015,"_380":30,"_167":946},{"_77":792,"_379":945,"_380":30,"_167":961},{"_77":794,"_379":945,"_380":30,"_167":1012},{"_77":959,"_379":954,"_380":30,"_167":960},{"_77":1013,"_379":593,"_380":30,"_167":1014},{"_77":794,"_379":619,"_380":30,"_167":1012},{"_77":1010,"_379":494,"_380":30,"_167":1011},{"_77":786,"_379":619,"_380":30,"_167":1005},{"_77":982,"_379":598,"_380":85,"_167":374},{"_77":780,"_379":1009,"_380":30,"_167":948},{"_77":1007,"_379":947,"_380":85,"_167":1008},{"_77":971,"_379":494,"_380":30,"_167":962},{"_77":944,"_379":494,"_380":30,"_167":946},{"_77":1006,"_379":506,"_380":30,"_167":374},{"_77":786,"_379":494,"_380":30,"_167":1005},{"_77":794,"_379":939,"_380":30,"_167":1004},{"_77":1002,"_379":602,"_380":85,"_167":1003},{"_77":1000,"_379":494,"_380":85,"_167":1001},{"_77":999,"_379":494,"_380":30,"_167":374},{"_77":997,"_379":945,"_380":30,"_167":998},{"_77":996,"_379":506,"_380":30,"_167":374},{"_77":987,"_379":995,"_380":85,"_167":989},{"_77":993,"_379":598,"_380":30,"_167":994},{"_77":991,"_379":683,"_380":30,"_167":992},{"_77":990,"_379":500,"_380":30,"_167":374},{"_77":987,"_379":988,"_380":85,"_167":989},{"_77":985,"_379":954,"_380":85,"_167":986},{"_77":983,"_379":494,"_380":30,"_167":984},{"_77":982,"_379":497,"_380":85,"_167":374},{"_77":953,"_379":957,"_380":30,"_167":955},{"_77":785,"_379":972,"_380":30,"_167":962},{"_77":980,"_379":497,"_380":30,"_167":981},{"_77":978,"_379":494,"_380":30,"_167":979},{"_77":976,"_379":494,"_380":30,"_167":977},{"_77":975,"_379":494,"_380":30,"_167":374},{"_77":817,"_379":497,"_380":30,"_167":974},{"_77":971,"_379":619,"_380":30,"_167":962},{"_77":774,"_379":590,"_380":30,"_167":973},{"_77":971,"_379":972,"_380":30,"_167":962},{"_77":970,"_379":506,"_380":30,"_167":374},{"_77":969,"_379":494,"_380":30,"_167":942},{"_77":967,"_379":954,"_380":30,"_167":968},{"_77":780,"_379":965,"_380":30,"_167":966},{"_77":964,"_379":506,"_380":30,"_167":374},{"_77":963,"_379":494,"_380":30,"_167":374},{"_77":956,"_379":954,"_380":85,"_167":958},{"_77":785,"_379":619,"_380":30,"_167":962},{"_77":792,"_379":619,"_380":30,"_167":961},{"_77":959,"_379":957,"_380":30,"_167":960},{"_77":956,"_379":957,"_380":85,"_167":958},{"_77":953,"_379":954,"_380":30,"_167":955},{"_77":951,"_379":947,"_380":85,"_167":952},{"_77":949,"_379":605,"_380":30,"_167":950},{"_77":781,"_379":947,"_380":30,"_167":948},{"_77":944,"_379":945,"_380":30,"_167":946},{"_77":774,"_379":597,"_380":30,"_167":943},{"_77":941,"_379":494,"_380":30,"_167":942},{"_77":792,"_379":939,"_380":30,"_167":940},{"_77":937,"_379":494,"_380":30,"_167":938},"LoginHints","Keeps users logged in across NVIDIA domains.","api.ngc.nvidia.com","This cookie is managed by AWS and is used for load balancing.","bm_so","This cookie name is used to load the site as quickly as possible and to recognize bots using Akamai.","The __cf_bm cookie is a cookie necessary to support Cloudflare Bot Management, currently in private beta. As part of our bot management service, this cookie helps manage incoming traffic that matches criteria associated with bots.\nThis is a CloudFoundry cookie","OptanonConsent","nvid.nvidia.com","This cookie is set by the cookie compliance solution from OneTrust. It stores information about the categories of cookies the site uses and whether visitors have given or withdrawn consent for the use of each category. This enables site owners to prevent cookies in each category from being set in the users browser, when consent is not given. The cookie has a normal lifespan of one year, so that returning visitors to the site will have their preferences remembered. It contains no information that can identify the site visitor.","events.nvidia.com","This cookie is associated with the Amazon Web Services Elastic Load Balancing functionality for routing client request on the server.Used to attribute commission to affiliates when you arrive at the website from an affiliate referral link. It is set when you click on one of our links and used to let the advertiser and us know the website from which you came.","sessionid","This is a very generic cookie name that may have different purposes on different sites, but generally it will be some kind of anonymous session identifier.","_csrf","This cookie is used to prevent Cross-Site Request Forgery (CSRF) attacks, ensuring that each request made to the server is from the user's browser and is intentional.","LSKey-c$CookieConsentPolicy","enterprise-support.nvidia.com","Used to apply end-user cookie consent preferences set by our client-side utility.Custom- Short Duration. Unique ID of an Enquiry- used to validate access to case, and load exisiting data.\n","renderCtx","academicgrants.nvidia.com","Used to store site parameters in the session for reuse across requests by a single client for functionality and performance reasons.","CookieConsentPolicy","Used to apply end-user cookie consent preferences set by our client-side utility.","This cookie is managed by AWS and is used for load balancing.Used to attribute commission to affiliates when you arrive at the website from an affiliate referral link. It is set when you click on one of our links and used to let the advertiser and us know the website from which you came.","This cookie is associated with Akamai and is used to differentiate between traffic from humans and bots.","nv_co","__uxq412__ses.88c1","learn.learn.nvidia.com","This cookie is associated with the Amazon Web Services Elastic Load Balancing functionality for routing client request on the server.","pctrk","Used to count page views by unauthenticated users against license usage.","bm_s","bpazaws52gukakzc__ctrl0_ctl36_uccaptcha","bm_sv","store.nvidia.com","The __cf_bm cookie is a cookie necessary to support Cloudflare Bot Management, currently in private beta. As part of our bot management service, this cookie helps manage incoming traffic that matches criteria associated with bots.\n","This cookie name is used by DataDog to determine if cookies are supported on the website.dd_cookie_test","AMP_TLDTEST_xxxxxxxx","SESSlsd","This cookie is likely to be a session identifier used to maintain user session state across page requests.","mboxEdgeCluster","This cookie name is associated with Adobe Target which is used to communicate the correct edge cluster to at.js so that Target profiles can stay in sync as users navigate across a site.","_rdt_em","This cookie name is associated with Reddit which is used to manage session state and user interactions.","megaNavCollapsed","_abck","The _abck cookie is associated with Akamai's Bot Manager, which is used to prevent fraud and abuse by safeguarding online interactions and detecting irregularities that could suggest malicious activity, such as bot-driven attacks or hacking attempts.","idccsrf","This cookie name is associated with Salesforce which is used to track CrossSiteRequestForgery validation for certain single sign-on (SSO) flows.","ApplicationGatewayAffinity","partners.nvidia.com","This cookie is used to maintain a user session by the server in a load balanced environment, to ensure the user has a consistent experience.","AMP_TEST_xxxxxxxx","bm_lso","This cookie name is associated with Akamai which is used to load the site as quickly as possible and to recognize bots.","SSESS###########","This cookie is likely to be a session cookie used to maintain an anonymized user session by the server in a secure environment, often seen in sites built with the Drupal content management system.","nvdeveloper.nvidia.com","__uxq412__id.88c1","OptanonAlertBoxClosed","This cookie is set by websites using certain versions of the cookie law compliance solution from OneTrust. It is set after visitors have seen a cookie information notice and in some cases only when they actively close the notice down. It enables the website not to show the message more than once to a user. The cookie has a normal lifespan of one year and contains no personal information.","QueueITAccepted-SDFrts345E-V3_{eventId}","TEST_AMCV_COOKIE_WRITE","This cookie is used to test whether cookies are enabled in the browser. It is a temporary cookie used to determine if the browser accepts cookies.","XSRF-TOKEN","This cookie is written to help with site security in preventing Cross-Site Request Forgery attacks.This cookie helps with site security in preventing Cross-Site Request Forgery attacks.","AWS ELB application load balancer","This cookie is associated with Akamai. It is used to analyse traffic to determine if it is automated traffic generated by bots or a human user.","bpazaws52gukakzc__ctrl0_ctl78_list_ctl09_systemcaptcha","PHPSESSID","PHP session cookie associated with embedded content from this domain.\n","base.manager.learn.nvidia.com","bm_ss","This cookie is used to help protect websites from automated bot traffic.","AWS ELB application load balancerUsed to attribute commission to affiliates when you arrive at the website from an affiliate referral link. It is set when you click on one of our links and used to let the advertiser and us know the website from which you came.","_devzone_session","This cookie is likely used to maintain a user session on the website by the server. It is common for session identifiers to be used to keep users logged in and track their interaction with the site during a session.","docs.ngc.nvidia.com","bm_sz","This cookie is used by Akamai to optimize site performance and security. It is typically associated with the delivery of web content and services.","vcalliance.nvidia.com","ASP.NET_SessionId","General purpose platform session cookie, used by sites written with Microsoft .NET based technologies. Usually used to maintain an anonymised user session by the server.","com.adobe.alloy.getTld","This cookie name is associated with the Adobe WebSDK which is used to determine the top-most domain that can accept cookies.","SESSls","This cookie is likely to be a session identifier, used to maintain a user's session state across page requests.","c_code","BIGipServerx","This cookie name is associated with the BIG-IP product suite from company F5. Usually associated with managing sessions on load balanced servers, to ensure user requests are routed consistently to the correct server. The common root is BIGipServer most commonly followed by a domain name, usually the one that it is hosted on, but not always.","This cookie is set by websites using certain versions of the cookie law compliance solution from OneTrust. It is set after visitors have seen a cookie information notice and in some cases only when they actively close the notice down. It enables the website not to show the message more than once to a user. The cookie has a one year lifespan and contains no personal information.","default","

NVIDIA uses cookies to improve your experience on our web site. We and our third-party partners also use cookies and other tools to collect and record information you provide as well as information about your interactions with our websites for performance improvement, analytics, and to assist in marketing efforts. By clicking \"Accept All\", you consent to our use of cookies and other tools as described in our Cookie Policy. You can manage your cookie settings by clicking on \"Manage Settings.\" By continuing to use this site or by clicking one of the buttons below, you agree to our Terms of Service (which contains important waivers). Please see our Privacy Policy for more information on our privacy practices.

","gpc","

We have detected the Global Privacy Control (GPC) signal and recorded your rejection of all optional cookies on this site for this browser. You can manage your cookie settings by clicking on \"Manage Settings\". Please see our Cookie Policy for more information. To opt out of non-cookie personal information \"sales\" / \"sharing\" for targeted advertising purposes, please visit the NVIDIA Preference Center. Please see our Privacy Policy for more information on our privacy practices.

","gpcLoggedIn","

We have detected the Global Privacy Control Signal (GPC) and have opted you out of all optional cookies on this browser. You can manage your cookie settings by clicking on \"Manage Settings\". Please see our Cookie Policy for more information. We have also opted you out of \"sharing\"/\"sales\" of personal information outside of cookies. You can manage these settings in the NVIDIA NVIDIA Preference Center. Please see our Privacy Policy for more information.

","gpcLoggedInOverridden","

We have detected the Global Privacy Control Signal (GPC) and have opted you out of all optional cookies on this browser. You can manage your cookie settings by clicking on \"Manage Settings\". Please see our Cookie Policy for more information. We have also opted you out of \"sharing\"/\"sales\" of personal information outside of cookies which overrides at least one of your previous settings. You can manage them in the NVIDIA Preference Center. Please see our Privacy Policy for more information.

",{"_719":16,"_622":55,"_529":55,"_349":55,"_339":55},"hasConsented","gpcApplied","showBanner","geolocation",{"_1043":1044,"_1045":1046},"country","US","state","OR","query","categories",[1066,1067,1068,1069,1070],"optionUrls",[1054,1055,1056,1057,1058,1059,1060,1061,1062,1063,1064,1065],"total",12,"/orgs/nim/deepseek-ai/containers/deepseek-r1/-","/orgs/nim/nvidia/containers/llama-3.1-nemotron-70b-instruct/-","/orgs/nvidia/-/containers/pytorch/-","/orgs/nvidia/omniverse/collections/kit/-","/orgs/nvidia/-/collections/deepstream_sdk/-","/orgs/nvidia/omniverse/collections/kit-appstreaming-collection/-","/orgs/nvidia/research/models/stylegan3/-","/orgs/nvidia/tao/models/peoplenet/-","/orgs/nvidia/tao/models/trafficcamnet/-","/orgs/nvidia/riva/resources/riva_quickstart/-","/orgs/nvidia/-/helm-charts/gpu-operator/-","/orgs/nvidia/-/helm-charts/k8s-nim-operator/-",{"_206":1145,"_1072":1146,"_1074":55},{"_206":1110,"_1072":1111,"_1074":58},{"_206":1093,"_1072":1094,"_1074":1095},{"_206":207,"_1072":1086,"_1074":1087},{"_206":1071,"_1072":1073,"_1074":1075},"helm-charts","items",[1076,1077],"startIndex",10,{"_206":1071,"_213":1083,"_161":162,"_163":-7,"_208":1084,"_216":1080,"_243":1085,"_1082":1064},{"_206":1071,"_213":1078,"_161":162,"_163":-7,"_208":1079,"_216":1080,"_243":1081,"_1082":1065},"NVIDIA NIM Operator","k8s-nim-operator","NVIDIA",[],"url","GPU Operator","gpu-operator",[],[1088],9,{"_206":207,"_213":1089,"_161":162,"_163":1090,"_208":1091,"_216":1080,"_243":1092,"_1082":1063},"Riva Skills Quick Start","riva","riva_quickstart",[],"models",[1096,1097,1098],6,{"_206":1093,"_213":1106,"_161":162,"_163":1107,"_208":1108,"_216":1080,"_243":1109,"_1082":1060},{"_206":1093,"_213":1103,"_161":162,"_163":1100,"_208":1104,"_216":1080,"_243":1105,"_1082":1061},{"_206":1093,"_213":1099,"_161":162,"_163":1100,"_208":1101,"_216":1080,"_243":1102,"_1082":1062},"TrafficCamNet","tao","trafficcamnet",[],"PeopleNet","peoplenet",[],"StyleGAN3 pretrained models","research","stylegan3",[],"collections",[1112,1113,1114],{"_206":1110,"_213":1142,"_161":162,"_163":1116,"_208":1143,"_216":1080,"_243":1144,"_1082":1057},{"_206":1110,"_213":1139,"_161":162,"_163":-7,"_208":1140,"_216":1080,"_243":1141,"_1082":1058},{"_206":1110,"_213":1115,"_161":162,"_163":1116,"_208":1117,"_216":1080,"_243":1118,"_1082":1059},"Omniverse Kit App Streaming","omniverse","kit-appstreaming-collection",[1119,1120,1121],{"_77":1133,"_213":1134,"_210":1135,"_221":1136,"_1126":1137},{"_77":1116,"_213":1129,"_210":1130,"_221":1125,"_1126":1131},{"_77":1122,"_213":1123,"_210":1124,"_221":1125,"_1126":1127},"omniverse-dev","NVIDIA Developer Program","Join the Developer Program for access to free tools, support, and tech resources.","https://store-images.s-microsoft.com/image/apps.33936.6f1bcf42-55d0-45d1-b4b6-f08c68a50bbe.e91ca707-db64-49d6-9761-4781590a9685.7ba64fc5-7c44-4864-a5e6-7bde05126935","supportedFeatures",[1128],"UNIFIED_CATALOG","NVIDIA Omniverse","NVIDIA Omniverse is a platform of APIs, services, and software development kits (SDKs) that enable developers to build generative AI-enabled tools and applications.",[1128,1132],"SHAREABLE_TO_USER_GROUPS","nv-ai-enterprise","NVIDIA AI Enterprise","Accelerate your AI agent development","https://assets.ngc.nvidia.com/products/catalog/images/product-NVAIE.jpg",[1128,1132,1138],"NGC_CONSOLE","DeepStream SDK","deepstream_sdk",[],"Omniverse Kit (FB)","kit",[1119,1120,1121],"containers",[1147,1148,1149],{"_206":1145,"_213":1157,"_161":1154,"_163":1158,"_208":1159,"_216":1080,"_243":1160,"_1082":1054},{"_206":1145,"_213":1153,"_161":1154,"_163":162,"_208":1155,"_216":1080,"_243":1156,"_1082":1055},{"_206":1145,"_213":1150,"_161":162,"_163":-7,"_208":1151,"_216":1080,"_243":1152,"_1082":1056},"PyTorch","pytorch",[],"Llama-3.1-Nemotron-70B-Instruct","nim","llama-3.1-nemotron-70b-instruct",[],"DeepSeek-R1","deepseek-ai","deepseek-r1",[],"ownerOrgs",[],"memberOrgs",[],"externalOrgs",[],"apiKey","2b44f86fea50e90ffd2db29ee2028419","consentGranted","applicationId","4c857e4f-5ffa-45fa-b75e-af773c22f0ad","clientToken","pubb6927db33dae60ba5031a186cde0242a","site","datadoghq.com","service","catalog-ui","env","sessionSampleRate",100,"sessionReplaySampleRate","compressIntakeRequests","storeContextsAcrossPages","defaultPrivacyLevel","mask-user-input","trackViewsManually","trackingConsent","not-granted",{"_1190":1191,"_1192":286,"_1193":30,"_1194":1195,"_1072":1196,"_1197":1198,"_1199":1191,"_1200":30,"_1201":30,"_1202":30,"_1203":1204},"appName","ngc-catalog","appOrigin","hideOnPrivateOrg","icon","ObjectsModules2",[1205,1206,1207,1208,1209,1210,1211],"menuIcon","editor-swatches","productName","requiresAllScopes","requiresLogin","requiresOrgLevelScopesOnly","title","Catalog",{"_1203":1233,"_46":1234,"_1197":1235},{"_1226":1227,"_1203":1228,"_46":1229,"_1199":1133},{"_1203":1224,"_46":1225,"_1197":1198},{"_1203":1221,"_46":1222,"_1197":1223},{"_1203":1218,"_46":1219,"_1197":1220},{"_1203":1215,"_46":1216,"_1197":1217},{"_1203":1212,"_46":1213,"_1197":1214},"Resources","/resources","misc-library","Models","/models","misc-neural-network","Helm Charts","/helm-charts","misc-ship-wheel","Containers","/containers","3d-cube","Collections","/collections","access",[1230],"Enterprise Catalog","/enterprise",[1231,1232],"NVAIE","READ","Explore Catalog","/","common-home",{"_111":112,"_113":114,"_115":116,"_117":118,"_119":120,"_121":122},{"_110":30}]