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

The GNMT v2 model is similar to the one discussed in the Google's Neural\nMachine Translation System: Bridging the Gap between Human and Machine\nTranslation paper.

The most important difference between the two models is in the attention\nmechanism. In our model, the output from the first LSTM layer of the decoder\ngoes into the attention module, then the re-weighted context is concatenated\nwith inputs to all subsequent LSTM layers in the decoder at the current\ntime step.

The same attention mechanism is also implemented in the default GNMT-like\nmodels from TensorFlow Neural Machine Translation\nTutorial and NVIDIA OpenSeq2Seq\nToolkit.

Model architecture

$\"ModelArchitecture\"$

Default configuration

The following features were implemented in this model:

general:\n
- encoder and decoder are using shared embeddings
- data-parallel multi-GPU training
- dynamic loss scaling with backoff for Tensor Cores (mixed precision)\ntraining
- trained with label smoothing loss (smoothing factor 0.1)
\n
encoder:\n
- 4-layer LSTM, hidden size 1024, first layer is bidirectional, the rest are\nunidirectional
- with residual connections starting from 3rd layer
- uses standard PyTorch nn.LSTM layer
- dropout is applied on input to all LSTM layers, probability of dropout is\nset to 0.2
- hidden state of LSTM layers is initialized with zeros
- weights and bias of LSTM layers is initialized with uniform(-0.1,0.1)\ndistribution
\n
decoder:\n
- 4-layer unidirectional LSTM with hidden size 1024 and fully-connected\nclassifier
- with residual connections starting from 3rd layer
- uses standard PyTorch nn.LSTM layer
- dropout is applied on input to all LSTM layers, probability of dropout is\nset to 0.2
- hidden state of LSTM layers is initialized with zeros
- weights and bias of LSTM layers is initialized with uniform(-0.1,0.1)\ndistribution
- weights and bias of fully-connected classifier is initialized with\nuniform(-0.1,0.1) distribution
\n
attention:\n
- normalized Bahdanau attention
- output from first LSTM layer of decoder goes into attention, then\nre-weighted context is concatenated with the input to all subsequent LSTM\nlayers of the decoder at the current timestep
- linear transform of keys and queries is initialized with uniform(-0.1,\n0.1), normalization scalar is initialized with 1.0/sqrt(1024),\nnormalization bias is initialized with zero
\n
inference:\n
- beam search with default beam size of 5
- with coverage penalty and length normalization, coverage penalty factor is\nset to 0.1, length normalization factor is set to 0.6 and length\nnormalization constant is set to 5.0
- de-tokenized BLEU computed by\nSacreBLEU
- motivation for choosing\nSacreBLEU
\n

When comparing the BLEU score, there are various tokenization approaches and\nBLEU calculation methodologies; therefore, ensure you align similar metrics.

Code from this repository can be used to train a larger, 8-layer GNMT v2 model.\nOur experiments show that a 4-layer model is significantly faster to train and\nyields comparable accuracy on the public WMT16\nEnglish-German dataset. The\nnumber of LSTM layers is controlled by the --num-layers parameter in the\ntrain.py training script.

Feature support matrix

The following features are supported by this model.

\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	GNMT v2
Apex AMP	Yes
Apex DistributedDataParallel	Yes

Features

Apex AMP - a tool that enables Tensor\nCore-accelerated training. Refer to the Enabling mixed\nprecision section for more details.

Apex\nDistributedDataParallel -\na module wrapper that enables easy multiprocess distributed data parallel\ntraining, similar to\ntorch.nn.parallel.DistributedDataParallel.\nDistributedDataParallel is optimized for use with\nNCCL. It achieves high performance by\noverlapping communication with computation during backward() and bucketing\nsmaller gradient transfers to reduce the total number of transfers required.

Mixed precision training

Mixed precision is the combined use of different numerical precisions in a\ncomputational method.\nMixed precision training offers significant\ncomputational speedup by performing operations in half-precision format, while\nstoring minimal information in single-precision to retain as much information\nas possible in critical parts of the network. Since the introduction of Tensor\nCores in Volta, and following with\nboth the Turing and Ampere architectures, significant training speedups are\nexperienced by switching to mixed precision -- up to 3x overall speedup on the\nmost arithmetically intense model architectures. Using mixed precision training\npreviously required two steps:

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

The ability to train deep learning networks with lower precision was introduced\nin the Pascal architecture and first supported in CUDA\n8 in the NVIDIA Deep\nLearning SDK.

For information about:

How to train using mixed precision, see the Mixed Precision\nTraining paper and Training With Mixed\nPrecision\ndocumentation.
Techniques used for mixed precision training, see the Mixed-Precision\nTraining of Deep Neural\nNetworks\nblog.
APEX tools for mixed precision training, see the NVIDIA Apex: Tools for Easy\nMixed-Precision Training in\nPyTorch\n.

Enabling mixed precision

Mixed precision is enabled in PyTorch by using the Automatic Mixed Precision\n(AMP), library from APEX that casts variables\nto half-precision upon retrieval, while storing variables in single-precision\nformat. Furthermore, to preserve small gradient magnitudes in backpropagation,\na loss\nscaling\nstep must be included when applying gradients. In PyTorch, loss scaling can be\neasily applied by using scale_loss() method provided by AMP. The scaling\nvalue to be used can be\ndynamic or fixed.

For an in-depth walk through on AMP, check out sample usage\nhere.\nAPEX is a PyTorch extension that contains\nutility libraries, such as AMP, which require minimal network code changes to\nleverage Tensor Cores performance.

The following steps were needed to enable mixed precision training in GNMT:

Import AMP from APEX (file: seq2seq/train/trainer.py):

from apex import amp\n

Initialize AMP and wrap the model and the optimizer (file:\nseq2seq/train/trainer.py, class: Seq2SeqTrainer):

self.model, self.optimizer = amp.initialize(\n    self.model,\n    self.optimizer,\n    cast_model_outputs=torch.float16,\n    keep_batchnorm_fp32=False,\n    opt_level='O2')\n

Apply scale_loss context manager (file: seq2seq/train/fp_optimizers.py,\nclass: AMPOptimizer):

with amp.scale_loss(loss, optimizer) as scaled_loss:\n    scaled_loss.backward()\n

Apply gradient clipping on single precision master weights (file:\nseq2seq/train/fp_optimizers.py, class: AMPOptimizer):

if self.grad_clip != float('inf'):\n    clip_grad_norm_(amp.master_params(optimizer), self.grad_clip)\n

Enabling TF32

TensorFloat-32 (TF32) is the new math mode in NVIDIA\nA100 GPUs for handling the\nmatrix math also called tensor operations. TF32 running on Tensor Cores in A100\nGPUs can provide up to 10x speedups compared to single-precision floating-point\nmath (FP32) on Volta GPUs.

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

For more information, refer to the TensorFloat-32 in the A100 GPU Accelerates\nAI Training, HPC up to\n20x\nblog post.

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

","resourceVersions",[231,232],"resourceVersionsTotal","hasMoreVersions","versionMetadata",{"_210":28,"_211":-5,"_52":116,"_113":-5,"_117":212,"_119":-5,"_213":115,"_124":16,"_121":34,"_214":-5,"_120":34,"_215":216,"_217":90,"_218":219,"_220":90,"_221":222,"_223":90,"_224":225,"_226":90,"_227":228,"_229":90,"_122":230},"versionNotFound","isLoggedIn","latestVersionPath","/orgs/nvidia/dle/resources/gnmt_pyt/-","-","imageReference","cliDownloadCommand","ngc registry resource download-version nvidia/dle/gnmt_pyt:20.06.0","helmFetchCommand","viewAllVersionsPath","/orgs/nvidia/dle/resources/gnmt_pyt/-/version-history","accessCheckResult",{"_209":90},"hasRelatedCollections","context",{"_205":206,"_207":208},1787694354093,[],[],[197,198,199,200,201,202,203],{"_48":63,"_50":73,"_52":65,"_54":66,"_56":74,"_58":59},{"_48":63,"_50":72,"_52":65,"_54":66,"_56":67,"_58":59},{"_48":63,"_50":70,"_52":65,"_54":66,"_56":71,"_58":59},{"_48":63,"_50":68,"_52":65,"_54":66,"_56":204,"_58":59},{"_48":63,"_50":64,"_52":65,"_54":66,"_56":67,"_58":59},{"_48":60,"_50":61,"_52":53,"_54":55,"_56":62,"_58":59},{"_48":49,"_50":51,"_52":53,"_54":55,"_56":57,"_58":59},57,"orgName","nvidia","teamName","dle","allowed","versionId","description",["D",1663857277458],"totalSizeInBytes","malwareScanStatus","releaseNotes","### Changelog\n* July 2020\n * Added support for NVIDIA DGX A100\n * Default container updated to NGC PyTorch 20.06-py3\n* June 2019\n * Default container updated to NGC PyTorch 19.05-py3\n * Mixed precision training implemented using APEX AMP\n * Added inference throughput and latency results on NVIDIA T4 and NVIDIA\n Tesla V100 16GB\n * Added option to run inference on user-provided raw input text from command\n line\n* February 2019\n * Different batching algorithm (bucketing with 5 equal-width buckets)\n * Additional dropouts before first LSTM layer in encoder and in decoder\n * Weight initialization changed to uniform (-0.1,0.1)\n * Switched order of dropout and concatenation with attention in decoder\n * Default container updated to NGC PyTorch 19.01-py3\n* December 2018\n * Added exponential warm-up and step learning rate decay\n * Multi-GPU (distributed) inference and validation\n * Default container updated to NGC PyTorch 18.11-py3\n * General performance improvements\n* August 2018\n * Initial release\n\n### Known issues\nThere are no known issues in this release.","hasReleaseNotes","setup","The following section lists the requirements in order to start training the\nGNMT v2 model.\n\n### Requirements\n\nThis repository contains `Dockerfile` which extends the PyTorch NGC container\nand encapsulates some dependencies. Aside from these dependencies, ensure you\nhave the following components:\n* [NVIDIA Docker](https://github.com/NVIDIA/nvidia-docker)\n* [PyTorch 20.06-py3 NGC container](https://ngc.nvidia.com/registry/nvidia-pytorch)\n* GPU architecture:\n * [NVIDIA Volta](https://www.nvidia.com/en-us/data-center/volta-gpu-architecture/)\n * [NVIDIA Turing](https://www.nvidia.com/en-us/geforce/turing/)\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, see the\nfollowing sections from the NVIDIA GPU Cloud Documentation and the Deep\nLearning DGX Documentation:\n\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/dgx/user-guide/index.html#accessing_registry),\n* [Running PyTorch](https://docs.nvidia.com/deeplearning/dgx/pytorch-release-notes/running.html#running).\n\nFor those unable to use the Pytorch NGC container, to set up the required\nenvironment or create your own container, see the versioned [NVIDIA Container\nSupport\nMatrix](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\nFP32, perform the following steps using the default parameters of the GNMT v2\nmodel on the WMT16 English German dataset. For the specifics concerning\ntraining and inference, see the [Advanced](#advanced) section.\n\n**1. Clone the repository.**\n\n```\ngit clone https://github.com/NVIDIA/DeepLearningExamples\ncd DeepLearningExamples/PyTorch/Translation/GNMT\n```\n\n**2. Build the GNMT v2 Docker container.**\n\n```\nbash scripts/docker/build.sh\n```\n\n**3. Start an interactive session in the container to run training/inference.**\n\n```\nbash scripts/docker/interactive.sh\n```\n\n**4. Download and preprocess the dataset.**\n\nData will be downloaded to the `data` directory (on the host). The `data`\ndirectory is mounted to the `/workspace/gnmt/data` location in the Docker\ncontainer.\n\n```\nbash scripts/wmt16_en_de.sh\n```\n\n**5. Start training.**\n\nThe training script saves only one checkpoint with the lowest value of the loss\nfunction on the validation dataset. All results and logs are saved to the\n`gnmt` directory (on the host) or to the `/workspace/gnmt/gnmt` directory\n(in the container). By default, the `train.py` script will launch mixed\nprecision training with Tensor Cores. You can change this behavior by setting:\n* the `--math fp32` flag to launch single precision training (for NVIDIA Volta\n and NVIDIA Turing architectures) or\n* the `--math tf32` flag to launch TF32 training with Tensor Cores (for NVIDIA\n Ampere architecture) \n\nfor the `train.py` training script.\n\nTo launch mixed precision training on 1, 4 or 8 GPUs, run:\n\n```\npython3 -m torch.distributed.launch --nproc_per_node=<#GPUs> train.py --seed 2 --train-global-batch-size 1024\n```\n\nTo launch mixed precision training on 16 GPUs, run:\n\n```\npython3 -m torch.distributed.launch --nproc_per_node=16 train.py --seed 2 --train-global-batch-size 2048\n```\n\nBy default, the training script will launch training with batch size 128 per\nGPU. If `--train-global-batch-size` is specified and larger than 128 times the\nnumber of GPUs available for the training then the training script will\naccumulate gradients over consecutive iterations and then perform the weight\nupdate. For example, 1 GPU training with `--train-global-batch-size 1024` will\naccumulate gradients over 8 iterations before doing the weight update with\naccumulated gradients.\n\n**6. Start evaluation.**\n\nThe training process automatically runs evaluation and outputs the BLEU score\nafter each training epoch. Additionally, after the training is done, you can\nmanually run inference on the test dataset with the checkpoint saved during the\ntraining.\n\nTo launch FP16 inference on the `newstest2014.en` test set, run:\n\n```\npython3 translate.py \\\n --input data/wmt16_de_en/newstest2014.en \\\n --reference data/wmt16_de_en/newstest2014.de \\\n --output /tmp/output \\\n --model gnmt/model_best.pth\n```\n\nThe script will load the checkpoint specified by the `--model` option, then it\nwill launch inference on the file specified by the `--input` option, and\ncompute BLEU score against the reference translation specified by the\n`--reference` option. Outputs will be stored to the location specified by the\n`--output` option.\n\nAdditionally, one can pass the input text directly from the command-line:\n\n```\npython3 translate.py \\\n --input-text \"The quick brown fox jumps over the lazy dog\" \\\n --model gnmt/model_best.pth\n```\n\nTranslated output will be printed to the console:\n\n```\n(...)\n0: Translated output:\nDer schnelle braune Fuchs springt über den faulen Hund\n```\n\nBy default, the `translate.py` script will launch FP16 inference with Tensor\nCores. You can change this behavior by setting:\n* the `--math fp32` flag to launch single precision inference (for NVIDIA Volta\n and NVIDIA Turing architectures) or\n* the `--math tf32` flag to launch TF32 inference with Tensor Cores (for NVIDIA\n Ampere architecture)\n\nfor the `translate.py` inference script.","hasQuickStartGuide","advanced","The following sections provide greater details of the dataset, running training\nand inference, and the training results.\n\n### Scripts and sample code\nIn the `root` directory, the most important files are:\n\n* `train.py`: serves as the entry point to launch the training\n* `translate.py`: serves as the entry point to launch inference\n* `Dockerfile`: container with the basic set of dependencies to run GNMT v2\n* `requirements.txt`: set of extra requirements for running GNMT v2\n\nThe `seq2seq/model` directory contains the implementation of GNMT v2 building\nblocks:\n\n* `attention.py`: implementation of normalized Bahdanau attention\n* `encoder.py`: implementation of recurrent encoder\n* `decoder.py`: implementation of recurrent decoder with attention\n* `seq2seq_base.py`: base class for seq2seq models\n* `gnmt.py`: implementation of GNMT v2 model\n\nThe `seq2seq/train` directory encapsulates the necessary tools to execute\ntraining:\n\n* `trainer.py`: implementation of training loop\n* `smoothing.py`: implementation of cross-entropy with label smoothing\n* `lr_scheduler.py`: implementation of exponential learning rate warmup and\n step decay\n* `fp_optimizers.py`: implementation of optimizers for various floating point\n precisions\n\nThe `seq2seq/inference` directory contains scripts required to run inference:\n\n* `beam_search.py`: implementation of beam search with length normalization and\n length penalty\n* `translator.py`: implementation of auto-regressive inference\n\nThe `seq2seq/data` directory contains implementation of components needed for\ndata loading:\n\n* `dataset.py`: implementation of text datasets\n* `sampler.py`: implementation of batch samplers with bucketing by sequence\n length\n* `tokenizer.py`: implementation of tokenizer (maps integer vocabulary indices\n to text)\n\n### Parameters\nTraining\n\nThe complete list of available parameters for the `train.py` training script\ncontains:\n\n```\ndataset setup:\n --dataset-dir DATASET_DIR\n path to the directory with training/test data\n (default: data/wmt16_de_en)\n --src-lang SRC_LANG source language (default: en)\n --tgt-lang TGT_LANG target language (default: de)\n --vocab VOCAB path to the vocabulary file (relative to DATASET_DIR\n directory) (default: vocab.bpe.32000)\n -bpe BPE_CODES, --bpe-codes BPE_CODES\n path to the file with bpe codes (relative to\n DATASET_DIR directory) (default: bpe.32000)\n --train-src TRAIN_SRC\n path to the training source data file (relative to\n DATASET_DIR directory) (default:\n train.tok.clean.bpe.32000.en)\n --train-tgt TRAIN_TGT\n path to the training target data file (relative to\n DATASET_DIR directory) (default:\n train.tok.clean.bpe.32000.de)\n --val-src VAL_SRC path to the validation source data file (relative to\n DATASET_DIR directory) (default:\n newstest_dev.tok.clean.bpe.32000.en)\n --val-tgt VAL_TGT path to the validation target data file (relative to\n DATASET_DIR directory) (default:\n newstest_dev.tok.clean.bpe.32000.de)\n --test-src TEST_SRC path to the test source data file (relative to\n DATASET_DIR directory) (default:\n newstest2014.tok.bpe.32000.en)\n --test-tgt TEST_TGT path to the test target data file (relative to\n DATASET_DIR directory) (default: newstest2014.de)\n --train-max-size TRAIN_MAX_SIZE\n use at most TRAIN_MAX_SIZE elements from training\n dataset (useful for benchmarking), by default uses\n entire dataset (default: None)\n\nresults setup:\n --save-dir SAVE_DIR path to directory with results, it will be\n automatically created if it does not exist (default:\n gnmt)\n --print-freq PRINT_FREQ\n print log every PRINT_FREQ batches (default: 10)\n\nmodel setup:\n --hidden-size HIDDEN_SIZE\n hidden size of the model (default: 1024)\n --num-layers NUM_LAYERS\n number of RNN layers in encoder and in decoder\n (default: 4)\n --dropout DROPOUT dropout applied to input of RNN cells (default: 0.2)\n --share-embedding use shared embeddings for encoder and decoder (use '--\n no-share-embedding' to disable) (default: True)\n --smoothing SMOOTHING\n label smoothing, if equal to zero model will use\n CrossEntropyLoss, if not zero model will be trained\n with label smoothing loss (default: 0.1)\n\ngeneral setup:\n --math {fp16,fp32,tf32,manual_fp16}\n precision (default: fp16)\n --seed SEED master seed for random number generators, if \"seed\" is\n undefined then the master seed will be sampled from\n random.SystemRandom() (default: None)\n --prealloc-mode {off,once,always}\n controls preallocation (default: always)\n --dllog-file DLLOG_FILE\n Name of the DLLogger output file (default:\n train_log.json)\n --eval run validation and test after every epoch (use '--no-\n eval' to disable) (default: True)\n --env print info about execution env (use '--no-env' to\n disable) (default: True)\n --cuda enables cuda (use '--no-cuda' to disable) (default:\n True)\n --cudnn enables cudnn (use '--no-cudnn' to disable) (default:\n True)\n --log-all-ranks enables logging from all distributed ranks, if\n disabled then only logs from rank 0 are reported (use\n '--no-log-all-ranks' to disable) (default: True)\n\ntraining setup:\n --train-batch-size TRAIN_BATCH_SIZE\n training batch size per worker (default: 128)\n --train-global-batch-size TRAIN_GLOBAL_BATCH_SIZE\n global training batch size, this argument does not\n have to be defined, if it is defined it will be used\n to automatically compute train_iter_size using the\n equation: train_iter_size = train_global_batch_size //\n (train_batch_size * world_size) (default: None)\n --train-iter-size N training iter size, training loop will accumulate\n gradients over N iterations and execute optimizer\n every N steps (default: 1)\n --epochs EPOCHS max number of training epochs (default: 6)\n --grad-clip GRAD_CLIP\n enables gradient clipping and sets maximum norm of\n gradients (default: 5.0)\n --train-max-length TRAIN_MAX_LENGTH\n maximum sequence length for training (including\n special BOS and EOS tokens) (default: 50)\n --train-min-length TRAIN_MIN_LENGTH\n minimum sequence length for training (including\n special BOS and EOS tokens) (default: 0)\n --train-loader-workers TRAIN_LOADER_WORKERS\n number of workers for training data loading (default:\n 2)\n --batching {random,sharding,bucketing}\n select batching algorithm (default: bucketing)\n --shard-size SHARD_SIZE\n shard size for \"sharding\" batching algorithm, in\n multiples of global batch size (default: 80)\n --num-buckets NUM_BUCKETS\n number of buckets for \"bucketing\" batching algorithm\n (default: 5)\n\noptimizer setup:\n --optimizer OPTIMIZER\n training optimizer (default: Adam)\n --lr LR learning rate (default: 0.002)\n --optimizer-extra OPTIMIZER_EXTRA\n extra options for the optimizer (default: {})\n\nmixed precision loss scaling setup:\n --init-scale INIT_SCALE\n initial loss scale (default: 8192)\n --upscale-interval UPSCALE_INTERVAL\n loss upscaling interval (default: 128)\n\nlearning rate scheduler setup:\n --warmup-steps WARMUP_STEPS\n number of learning rate warmup iterations (default:\n 200)\n --remain-steps REMAIN_STEPS\n starting iteration for learning rate decay (default:\n 0.666)\n --decay-interval DECAY_INTERVAL\n interval between learning rate decay steps (default:\n None)\n --decay-steps DECAY_STEPS\n max number of learning rate decay steps (default: 4)\n --decay-factor DECAY_FACTOR\n learning rate decay factor (default: 0.5)\n\nvalidation setup:\n --val-batch-size VAL_BATCH_SIZE\n batch size for validation (default: 64)\n --val-max-length VAL_MAX_LENGTH\n maximum sequence length for validation (including\n special BOS and EOS tokens) (default: 125)\n --val-min-length VAL_MIN_LENGTH\n minimum sequence length for validation (including\n special BOS and EOS tokens) (default: 0)\n --val-loader-workers VAL_LOADER_WORKERS\n number of workers for validation data loading\n (default: 0)\n\ntest setup:\n --test-batch-size TEST_BATCH_SIZE\n batch size for test (default: 128)\n --test-max-length TEST_MAX_LENGTH\n maximum sequence length for test (including special\n BOS and EOS tokens) (default: 150)\n --test-min-length TEST_MIN_LENGTH\n minimum sequence length for test (including special\n BOS and EOS tokens) (default: 0)\n --beam-size BEAM_SIZE\n beam size (default: 5)\n --len-norm-factor LEN_NORM_FACTOR\n length normalization factor (default: 0.6)\n --cov-penalty-factor COV_PENALTY_FACTOR\n coverage penalty factor (default: 0.1)\n --len-norm-const LEN_NORM_CONST\n length normalization constant (default: 5.0)\n --intra-epoch-eval N evaluate within training epoch, this option will\n enable extra N equally spaced evaluations executed\n during each training epoch (default: 0)\n --test-loader-workers TEST_LOADER_WORKERS\n number of workers for test data loading (default: 0)\n\ncheckpointing setup:\n --start-epoch START_EPOCH\n manually set initial epoch counter (default: 0)\n --resume PATH resumes training from checkpoint from PATH (default:\n None)\n --save-all saves checkpoint after every epoch (default: False)\n --save-freq SAVE_FREQ\n save checkpoint every SAVE_FREQ batches (default:\n 5000)\n --keep-checkpoints KEEP_CHECKPOINTS\n keep only last KEEP_CHECKPOINTS checkpoints, affects\n only checkpoints controlled by --save-freq option\n (default: 0)\n\nbenchmark setup:\n --target-perf TARGET_PERF\n target training performance (in tokens per second)\n (default: None)\n --target-bleu TARGET_BLEU\n target accuracy (default: None)\n```\n\nInference\n\nThe complete list of available parameters for the `translate.py` inference\nscript contains:\n\n```\ndata setup:\n -o OUTPUT, --output OUTPUT\n full path to the output file if not specified, then\n the output will be printed (default: None)\n -r REFERENCE, --reference REFERENCE\n full path to the file with reference translations (for\n sacrebleu, raw text) (default: None)\n -m MODEL, --model MODEL\n full path to the model checkpoint file (default: None)\n --synthetic use synthetic dataset (default: False)\n --synthetic-batches SYNTHETIC_BATCHES\n number of synthetic batches to generate (default: 64)\n --synthetic-vocab SYNTHETIC_VOCAB\n size of synthetic vocabulary (default: 32320)\n --synthetic-len SYNTHETIC_LEN\n sequence length of synthetic samples (default: 50)\n -i INPUT, --input INPUT\n full path to the input file (raw text) (default: None)\n -t INPUT_TEXT [INPUT_TEXT ...], --input-text INPUT_TEXT [INPUT_TEXT ...]\n raw input text (default: None)\n --sort sorts dataset by sequence length (use '--no-sort' to\n disable) (default: False)\n\ninference setup:\n --batch-size BATCH_SIZE [BATCH_SIZE ...]\n batch size per GPU (default: [128])\n --beam-size BEAM_SIZE [BEAM_SIZE ...]\n beam size (default: [5])\n --max-seq-len MAX_SEQ_LEN\n maximum generated sequence length (default: 80)\n --len-norm-factor LEN_NORM_FACTOR\n length normalization factor (default: 0.6)\n --cov-penalty-factor COV_PENALTY_FACTOR\n coverage penalty factor (default: 0.1)\n --len-norm-const LEN_NORM_CONST\n length normalization constant (default: 5.0)\n\ngeneral setup:\n --math {fp16,fp32,tf32} [{fp16,fp32,tf32} ...]\n precision (default: ['fp16'])\n --env print info about execution env (use '--no-env' to\n disable) (default: False)\n --bleu compares with reference translation and computes BLEU\n (use '--no-bleu' to disable) (default: True)\n --cuda enables cuda (use '--no-cuda' to disable) (default:\n True)\n --cudnn enables cudnn (use '--no-cudnn' to disable) (default:\n True)\n --batch-first uses (batch, seq, feature) data format for RNNs\n (default: True)\n --seq-first uses (seq, batch, feature) data format for RNNs\n (default: True)\n --save-dir SAVE_DIR path to directory with results, it will be\n automatically created if it does not exist (default:\n gnmt)\n --dllog-file DLLOG_FILE\n Name of the DLLogger output file (default:\n eval_log.json)\n --print-freq PRINT_FREQ, -p PRINT_FREQ\n print log every PRINT_FREQ batches (default: 1)\n\nbenchmark setup:\n --target-perf TARGET_PERF\n target inference performance (in tokens per second)\n (default: None)\n --target-bleu TARGET_BLEU\n target accuracy (default: None)\n --repeat REPEAT [REPEAT ...]\n loops over the dataset REPEAT times, flag accepts\n multiple arguments, one for each specified batch size\n (default: [1])\n --warmup WARMUP warmup iterations for performance counters (default:\n 0)\n --percentiles PERCENTILES [PERCENTILES ...]\n Percentiles for confidence intervals for\n throughput/latency benchmarks (default: (90, 95, 99))\n --tables print accuracy, throughput and latency results in\n tables (use '--no-tables' to disable) (default: False)\n```\n\n### Command-line options\nTo see the full list of available options and their descriptions, use the `-h`\nor `--help` command line option. For example, for training:\n\n```\npython3 train.py --help\n\nusage: train.py [-h] [--dataset-dir DATASET_DIR] [--src-lang SRC_LANG]\n [--tgt-lang TGT_LANG] [--vocab VOCAB] [-bpe BPE_CODES]\n [--train-src TRAIN_SRC] [--train-tgt TRAIN_TGT]\n [--val-src VAL_SRC] [--val-tgt VAL_TGT] [--test-src TEST_SRC]\n [--test-tgt TEST_TGT] [--save-dir SAVE_DIR]\n [--print-freq PRINT_FREQ] [--hidden-size HIDDEN_SIZE]\n [--num-layers NUM_LAYERS] [--dropout DROPOUT]\n [--share-embedding] [--smoothing SMOOTHING]\n [--math {fp16,fp32,tf32,manual_fp16}] [--seed SEED]\n [--prealloc-mode {off,once,always}] [--dllog-file DLLOG_FILE]\n [--eval] [--env] [--cuda] [--cudnn] [--log-all-ranks]\n [--train-max-size TRAIN_MAX_SIZE]\n [--train-batch-size TRAIN_BATCH_SIZE]\n [--train-global-batch-size TRAIN_GLOBAL_BATCH_SIZE]\n [--train-iter-size N] [--epochs EPOCHS]\n [--grad-clip GRAD_CLIP] [--train-max-length TRAIN_MAX_LENGTH]\n [--train-min-length TRAIN_MIN_LENGTH]\n [--train-loader-workers TRAIN_LOADER_WORKERS]\n [--batching {random,sharding,bucketing}]\n [--shard-size SHARD_SIZE] [--num-buckets NUM_BUCKETS]\n [--optimizer OPTIMIZER] [--lr LR]\n [--optimizer-extra OPTIMIZER_EXTRA] [--init-scale INIT_SCALE]\n [--upscale-interval UPSCALE_INTERVAL]\n [--warmup-steps WARMUP_STEPS] [--remain-steps REMAIN_STEPS]\n [--decay-interval DECAY_INTERVAL] [--decay-steps DECAY_STEPS]\n [--decay-factor DECAY_FACTOR]\n [--val-batch-size VAL_BATCH_SIZE]\n [--val-max-length VAL_MAX_LENGTH]\n [--val-min-length VAL_MIN_LENGTH]\n [--val-loader-workers VAL_LOADER_WORKERS]\n [--test-batch-size TEST_BATCH_SIZE]\n [--test-max-length TEST_MAX_LENGTH]\n [--test-min-length TEST_MIN_LENGTH] [--beam-size BEAM_SIZE]\n [--len-norm-factor LEN_NORM_FACTOR]\n [--cov-penalty-factor COV_PENALTY_FACTOR]\n [--len-norm-const LEN_NORM_CONST] [--intra-epoch-eval N]\n [--test-loader-workers TEST_LOADER_WORKERS]\n [--start-epoch START_EPOCH] [--resume PATH] [--save-all]\n [--save-freq SAVE_FREQ] [--keep-checkpoints KEEP_CHECKPOINTS]\n [--target-perf TARGET_PERF] [--target-bleu TARGET_BLEU]\n [--local_rank LOCAL_RANK]\n```\nFor example, for inference:\n\n```\npython3 translate.py --help\n\nusage: translate.py [-h] [-o OUTPUT] [-r REFERENCE] [-m MODEL] [--synthetic]\n [--synthetic-batches SYNTHETIC_BATCHES]\n [--synthetic-vocab SYNTHETIC_VOCAB]\n [--synthetic-len SYNTHETIC_LEN]\n [-i INPUT | -t INPUT_TEXT [INPUT_TEXT ...]] [--sort]\n [--batch-size BATCH_SIZE [BATCH_SIZE ...]]\n [--beam-size BEAM_SIZE [BEAM_SIZE ...]]\n [--max-seq-len MAX_SEQ_LEN]\n [--len-norm-factor LEN_NORM_FACTOR]\n [--cov-penalty-factor COV_PENALTY_FACTOR]\n [--len-norm-const LEN_NORM_CONST]\n [--math {fp16,fp32,tf32} [{fp16,fp32,tf32} ...]] [--env]\n [--bleu] [--cuda] [--cudnn] [--batch-first | --seq-first]\n [--save-dir SAVE_DIR] [--dllog-file DLLOG_FILE]\n [--print-freq PRINT_FREQ] [--target-perf TARGET_PERF]\n [--target-bleu TARGET_BLEU] [--repeat REPEAT [REPEAT ...]]\n [--warmup WARMUP]\n [--percentiles PERCENTILES [PERCENTILES ...]] [--tables]\n [--local_rank LOCAL_RANK]\n```\n\n### Getting the data\nThe GNMT v2 model was trained on the [WMT16\nEnglish-German](http://www.statmt.org/wmt16/translation-task.html) dataset.\nConcatenation of the newstest2015 and newstest2016 test sets are used as a\nvalidation dataset and the newstest2014 is used as a testing dataset.\n\nThis repository contains the `scripts/wmt16_en_de.sh` download script which\nautomatically downloads and preprocesses the training, validation and test\ndatasets. By default, data is downloaded to the `data` directory.\n\nOur download script is very similar to the `wmt16_en_de.sh` script from the\n[tensorflow/nmt](https://github.com/tensorflow/nmt/blob/master/nmt/scripts/wmt16_en_de.sh)\nrepository. Our download script contains an extra preprocessing step, which\ndiscards all pairs of sentences which can't be decoded by *latin-1* encoder.\nThe `scripts/wmt16_en_de.sh` script uses the\n[subword-nmt](https://github.com/rsennrich/subword-nmt) package to segment text\ninto subword units (Byte Pair Encodings -\n[BPE](https://en.wikipedia.org/wiki/Byte_pair_encoding)). By default, the\nscript builds the shared vocabulary of 32,000 tokens.\n\nIn order to test with other datasets, the script needs to be customized\naccordingly.\n\n#### Dataset guidelines\nThe process of downloading and preprocessing the data can be found in the\n`scripts/wmt16_en_de.sh` script.\n\nInitially, data is downloaded from [www.statmt.org](www.statmt.org). Then\n`europarl-v7`, `commoncrawl` and `news-commentary` corpora are concatenated to\nform the training dataset, similarly `newstest2015` and `newstest2016` are\nconcatenated to form the validation dataset. Raw data is preprocessed with\n[Moses](https://github.com/moses-smt/mosesdecoder), first by launching [Moses\ntokenizer](https://github.com/moses-smt/mosesdecoder/blob/master/scripts/tokenizer/tokenizer.perl)\n(tokenizer breaks up text into individual words), then by launching\n[clean-corpus-n.perl](https://github.com/moses-smt/mosesdecoder/blob/master/scripts/training/clean-corpus-n.perl)\nwhich removes invalid sentences and does initial filtering by sequence length.\n\nSecond stage of preprocessing is done by launching the\n`scripts/filter_dataset.py` script, which discards all pairs of sentences that\ncan't be decoded by latin-1 encoder.\n\nThird state of preprocessing uses the\n[subword-nmt](https://github.com/rsennrich/subword-nmt) package. First it\nbuilds shared [byte pair\nencoding](https://en.wikipedia.org/wiki/Byte_pair_encoding) vocabulary with\n32,000 merge operations (command `subword-nmt learn-bpe`), then it applies\ngenerated vocabulary to training, validation and test corpora (command\n`subword-nmt apply-bpe`).\n\n### Training process\nThe default training configuration can be launched by running the `train.py`\ntraining script. By default, the training script saves only one checkpoint with\nthe lowest value of the loss function on the validation dataset. An evaluation\nis then performed after each training epoch. Results are stored in the\n`gnmt` directory.\n\nThe training script launches data-parallel training with batch size 128 per GPU\non all available GPUs. We have tested reliance on up to 16 GPUs on a single\nnode.\nAfter each training epoch, the script runs an evaluation on the validation\ndataset and outputs a BLEU score on the test dataset (newstest2014). BLEU is\ncomputed by the [SacreBLEU](https://github.com/mjpost/sacreBLEU) package. Logs\nfrom the training and evaluation are saved to the `gnmt` directory.\n\nThe summary after each training epoch is printed in the following format:\n\n```\n0: Summary: Epoch: 3\tTraining Loss: 3.1336\tValidation Loss: 2.9587\tTest BLEU: 23.18\n0: Performance: Epoch: 3\tTraining: 418772 Tok/s\tValidation: 1445331 Tok/s\n```\n\nThe training loss is averaged over an entire training epoch, the validation\nloss is averaged over the validation dataset and the BLEU score is computed on\nthe test dataset. Performance is reported in total tokens per second. The\nresult is averaged over an entire training epoch and summed over all GPUs\nparticipating in the training.\n\nBy default, the `train.py` script will launch mixed precision training with\nTensor Cores. You can change this behavior by setting:\n* the `--math fp32` flag to launch single precision training (for NVIDIA Volta\n and NVIDIA Turing architectures) or\n* the `--math tf32` flag to launch TF32 training with Tensor Cores (for NVIDIA\n Ampere architecture)\n\nfor the `train.py` training script.\n\nTo view all available options for training, run `python3 train.py --help`.\n\n### Inference process\nInference can be run by launching the `translate.py` inference script,\nalthough, it requires a pre-trained model checkpoint and tokenized input.\n\nThe inference script, `translate.py`, supports batched inference. By default,\nit launches beam search with beam size of 5, coverage penalty term and length\nnormalization term. Greedy decoding can be enabled by setting the beam size to\n1.\n\nTo view all available options for inference, run `python3 translate.py --help`.","hasAdvanced","performance","The performance measurements in this document were conducted at the time of\npublication and may not reflect the performance achieved from NVIDIA's latest\nsoftware release. For the most up-to-date performance measurements, go to\n[NVIDIA Data Center Deep Learning Product\nPerformance](https://developer.nvidia.com/deep-learning-performance-training-inference).\n\n### Benchmarking\nThe following section shows how to run benchmarks measuring the model\nperformance in training and inference modes.\n\n#### Training performance benchmark\nTraining is launched on batches of text data, different batches have different\nsequence lengths (number of tokens in the longest sequence). Sequence length\nand batch efficiency (ratio of non-pad tokens to total number of tokens) affect\nperformance of the training, therefore it's recommended to run the training on\na large chunk of training dataset to get a stable and reliable average training\nperformance. Ideally at least one full epoch of training should be launched to\nget a good estimate of training performance.\n\nThe following commands will launch one epoch of training:\n\nTo launch mixed precision training on 1, 4 or 8 GPUs, run:\n```\npython3 -m torch.distributed.launch --nproc_per_node=<#GPUs> train.py --seed 2 --train-global-batch-size 1024 --epochs 1 --math fp16\n```\n\nTo launch mixed precision training on 16 GPUs, run:\n```\npython3 -m torch.distributed.launch --nproc_per_node=16 train.py --seed 2 --train-global-batch-size 2048 --epochs 1 --math fp16\n```\n\nChange `--math fp16` to `--math fp32` to launch single precision training (for\nNVIDIA Volta and NVIDIA Turing architectures) or to `--math tf32` to launch\nTF32 training with Tensor Cores (for NVIDIA Ampere architecture).\n\nAfter the training is completed, the `train.py` script prints a summary to\nstandard output. Performance results are printed in the following format:\n\n```\n(...)\n0: Performance: Epoch: 0\tTraining: 418926 Tok/s\tValidation: 1430828 Tok/s\n(...)\n```\n\n`Training: 418926 Tok/s` represents training throughput averaged over an entire\ntraining epoch and summed over all GPUs participating in the training.\n\n#### Inference performance benchmark\nThe inference performance and accuracy benchmarks require a checkpoint from a\nfully trained model.\n\nCommand to launch the inference accuracy benchmark on NVIDIA Volta or on NVIDIA\nTuring architectures:\n\n```\npython3 translate.py \\\n --model gnmt/model_best.pth \\\n --input data/wmt16_de_en/newstest2014.en \\\n --reference data/wmt16_de_en/newstest2014.de \\\n --output /tmp/output \\\n --math fp16 fp32 \\\n --batch-size 128 \\\n --beam-size 1 2 5 \\\n --tables\n```\n\nCommand to launch the inference accuracy benchmark on NVIDIA Ampere architecture:\n\n```\npython3 translate.py \\\n --model gnmt/model_best.pth \\\n --input data/wmt16_de_en/newstest2014.en \\\n --reference data/wmt16_de_en/newstest2014.de \\\n --output /tmp/output \\\n --math fp16 tf32 \\\n --batch-size 128 \\\n --beam-size 1 2 5 \\\n --tables\n```\n\nCommand to launch the inference throughput and latency benchmarks on NVIDIA\nVolta or NVIDIA Turing architectures:\n\n```\npython3 translate.py \\\n --model gnmt/model_best.pth \\\n --input data/wmt16_de_en/newstest2014.en \\\n --reference data/wmt16_de_en/newstest2014.de \\\n --output /tmp/output \\\n --math fp16 fp32 \\\n --batch-size 1 2 4 8 32 128 512 \\\n --repeat 1 1 1 1 2 8 16 \\\n --beam-size 1 2 5 \\\n --warmup 5 \\\n --tables\n```\n\nCommand to launch the inference throughput and latency benchmarks on NVIDIA\nAmpere architecture:\n\n```\npython3 translate.py \\\n --model gnmt/model_best.pth \\\n --input data/wmt16_de_en/newstest2014.en \\\n --reference data/wmt16_de_en/newstest2014.de \\\n --output /tmp/output \\\n --math fp16 tf32 \\\n --batch-size 1 2 4 8 32 128 512 \\\n --repeat 1 1 1 1 2 8 16 \\\n --beam-size 1 2 5 \\\n --warmup 5 \\\n --tables\n```\n\n### Results\nThe following sections provide details on how we achieved our performance and\naccuracy in training and inference.\n\n#### Training accuracy results\n\n##### Training accuracy: NVIDIA DGX A100 (8x A100 40GB)\nOur results were obtained by running the `train.py` script with the default\nbatch size = 128 per GPU in the pytorch-20.06-py3 NGC container on NVIDIA DGX\nA100 with 8x A100 40GB GPUs.\n\nCommand to launch the training:\n\n```\npython3 -m torch.distributed.launch --nproc_per_node=<#GPUs> train.py --seed 2 --train-global-batch-size 1024 --math fp16\n```\n\nChange `--math fp16` to `--math tf32` to launch TF32 training with Tensor Cores.\n\n| **GPUs** | **Batch Size / GPU** | **Accuracy - TF32 (BLEU)** | **Accuracy - Mixed precision (BLEU)** | **Time to Train - TF32 (minutes)** | **Time to Train - Mixed precision (minutes)** | **Time to Train Speedup (TF32 to Mixed precision)** |\n| --- | --- | ----- | ----- | ----- | ------ | ---- |\n| 8 | 128 | 24.46 | 24.60 | 34.7 | 22.7 | 1.53 |\n\nTo achieve these same results, follow the [Quick Start Guide](#quick-start-guide)\noutlined above.\n\n##### Training accuracy: NVIDIA DGX-1 (8x V100 16GB)\nOur results were obtained by running the `train.py` script with the default\nbatch size = 128 per GPU in the pytorch-20.06-py3 NGC container on NVIDIA DGX-1\nwith 8x V100 16GB GPUs.\n\nCommand to launch the training:\n\n```\npython3 -m torch.distributed.launch --nproc_per_node=<#GPUs> train.py --seed 2 --train-global-batch-size 1024 --math fp16\n```\n\nChange `--math fp16` to `--math fp32` to launch single precision training.\n\n| **GPUs** | **Batch Size / GPU** | **Accuracy - FP32 (BLEU)** | **Accuracy - Mixed precision (BLEU)** | **Time to Train - FP32 (minutes)** | **Time to Train - Mixed precision (minutes)** | **Time to Train Speedup (FP32 to Mixed precision)** |\n| --- | --- | ----- | ----- | ----- | ------ | ---- |\n| 1 | 128 | 24.41 | 24.42 | 810.0 | 224.0 | 3.62 |\n| 4 | 128 | 24.40 | 24.33 | 218.2 | 69.5 | 3.14 |\n| 8 | 128 | 24.45 | 24.38 | 112.0 | 38.6 | 2.90 |\n\nTo achieve these same results, follow the [Quick Start Guide](#quick-start-guide)\noutlined above.\n\n##### Training accuracy: NVIDIA DGX-2H (16x V100 32GB)\nOur results were obtained by running the `train.py` script with the default\nbatch size = 128 per GPU in the pytorch-20.06-py3 NGC container on NVIDIA DGX-2H\nwith 16x V100 32GB GPUs.\n\nTo launch mixed precision training on 16 GPUs, run:\n```\npython3 -m torch.distributed.launch --nproc_per_node=16 train.py --seed 2 --train-global-batch-size 2048 --math fp16\n```\n\nChange `--math fp16` to `--math fp32` to launch single precision training.\n\n| **GPUs** | **Batch Size / GPU** | **Accuracy - FP32 (BLEU)** | **Accuracy - Mixed precision (BLEU)** | **Time to Train - FP32 (minutes)** | **Time to Train - Mixed precision (minutes)** | **Time to Train Speedup (FP32 to Mixed precision)** |\n| --- | --- | ----- | ----- | ------ | ----- | ---- |\n| 16 | 128 | 24.41 | 24.38 | 52.1 | 19.4 | 2.69 |\n\nTo achieve these same results, follow the [Quick Start Guide](#quick-start-guide)\noutlined above.\n\n![TrainingLoss](https://raw.githubusercontent.com/NVIDIA/DeepLearningExamples/master/PyTorch/Translation/GNMT//img/training_loss.png)\n\n\n##### Training stability test\nThe GNMT v2 model was trained for 6 epochs, starting from 32 different initial\nrandom seeds. After each training epoch, the model was evaluated on the test\ndataset and the BLEU score was recorded. The training was performed in the\npytorch-20.06-py3 Docker container on NVIDIA DGX A100 with 8x A100 40GB GPUs.\nThe following table summarizes the results of the stability test.\n\nIn the following table, the BLEU scores after each training epoch for different\ninitial random seeds are displayed.\n\n| **Epoch** | **Average** | **Standard deviation** | **Minimum** | **Maximum** | **Median** |\n| --- | ------ | ----- | ------ | ------ | ------ |\n| 1 | 19.959 | 0.238 | 19.410 | 20.390 | 19.970 |\n| 2 | 21.772 | 0.293 | 20.960 | 22.280 | 21.820 |\n| 3 | 22.435 | 0.264 | 21.740 | 22.870 | 22.465 |\n| 4 | 23.167 | 0.166 | 22.870 | 23.620 | 23.195 |\n| 5 | 24.233 | 0.149 | 23.820 | 24.530 | 24.235 |\n| 6 | 24.416 | 0.131 | 24.140 | 24.660 | 24.390 |\n\n#### Training throughput results\n\n##### Training throughput: NVIDIA DGX A100 (8x A100 40GB)\nOur results were obtained by running the `train.py` training script in the\npytorch-20.06-py3 NGC container on NVIDIA DGX A100 with 8x A100 40GB GPUs.\nThroughput performance numbers (in tokens per second) were averaged over an\nentire training epoch.\n\n| **GPUs** | **Batch size / GPU** | **Throughput - TF32 (tok/s)** | **Throughput - Mixed precision (tok/s)** | **Throughput speedup (TF32 to Mixed precision)** | **Strong Scaling - TF32** | **Strong Scaling - Mixed precision** |\n| --- | --- | ------ | ------ | ----- | ----- | ----- |\n| 1 | 128 | 83214 | 140909 | 1.693 | 1.000 | 1.000 |\n| 4 | 128 | 278576 | 463144 | 1.663 | 3.348 | 3.287 |\n| 8 | 128 | 519952 | 822024 | 1.581 | 6.248 | 5.834 |\n\nTo achieve these same results, follow the [Quick Start Guide](#quick-start-guide)\noutlined above.\n\n##### Training throughput: NVIDIA DGX-1 (8x V100 16GB)\nOur results were obtained by running the `train.py` training script in the\npytorch-20.06-py3 NGC container on NVIDIA DGX-1 with 8x V100 16GB GPUs.\nThroughput performance numbers (in tokens per second) were averaged over an\nentire training epoch.\n\n| **GPUs** | **Batch size / GPU** | **Throughput - FP32 (tok/s)** | **Throughput - Mixed precision (tok/s)** | **Throughput speedup (FP32 to Mixed precision)** | **Strong Scaling - FP32** | **Strong Scaling - Mixed precision** |\n| --- | --- | ------ | ------ | ----- | ----- | ----- |\n| 1 | 128 | 21860 | 76438 | 3.497 | 1.000 | 1.000 |\n| 4 | 128 | 80224 | 249168 | 3.106 | 3.670 | 3.260 |\n| 8 | 128 | 154168 | 447832 | 2.905 | 7.053 | 5.859 |\n\nTo achieve these same results, follow the [Quick Start Guide](#quick-start-guide)\noutlined above.\n\n##### Training throughput: NVIDIA DGX-2H (16x V100 32GB)\nOur results were obtained by running the `train.py` training script in the\npytorch-20.06-py3 NGC container on NVIDIA DGX-2H with 16x V100 32GB GPUs.\nThroughput performance numbers (in tokens per second) were averaged over an\nentire training epoch.\n\n| **GPUs** | **Batch size / GPU** | **Throughput - FP32 (tok/s)** | **Throughput - Mixed precision (tok/s)** | **Throughput speedup (FP32 to Mixed precision)** | **Strong Scaling - FP32** | **Strong Scaling - Mixed precision** |\n| --- | --- | ------ | ------ | ----- | ------ | ------ |\n| 1 | 128 | 25583 | 87829 | 3.433 | 1.000 | 1.000 |\n| 4 | 128 | 91400 | 290640 | 3.180 | 3.573 | 3.309 |\n| 8 | 128 | 176616 | 522008 | 2.956 | 6.904 | 5.943 |\n| 16 | 128 | 351792 | 1010880 | 2.874 | 13.751 | 11.510 |\n\nTo achieve these same results, follow the [Quick Start Guide](#quick-start-guide)\noutlined above.\n\n#### Inference accuracy results\n\n##### Inference accuracy: NVIDIA A100 40GB\nOur results were obtained by running the `translate.py` script in the\npytorch-20.06-py3 NGC Docker container with NVIDIA A100 40GB GPU. Full\ncommand to launch the inference accuracy benchmark was provided in the\n[Inference performance benchmark](#inference-performance-benchmark) section.\n\n| **Batch Size** | **Beam Size** | **Accuracy - TF32 (BLEU)** | **Accuracy - FP16 (BLEU)** |\n| -------------: | ------------: | -------------------------: | -------------------------: |\n| 128 | 1 | 23.07 | 23.07 |\n| 128 | 2 | 23.81 | 23.81 |\n| 128 | 5 | 24.41 | 24.43 |\n\n##### Inference accuracy: NVIDIA Tesla V100 16GB\nOur results were obtained by running the `translate.py` script in the\npytorch-20.06-py3 NGC Docker container with NVIDIA Tesla V100 16GB GPU. Full\ncommand to launch the inference accuracy benchmark was provided in the\n[Inference performance benchmark](#inference-performance-benchmark) section.\n\n| **Batch Size** | **Beam Size** | **Accuracy - FP32 (BLEU)** | **Accuracy - FP16 (BLEU)** |\n| -------------: | ------------: | -------------------------: | -------------------------: |\n| 128 | 1 | 23.07 | 23.07 |\n| 128 | 2 | 23.81 | 23.79 |\n| 128 | 5 | 24.40 | 24.43 |\n\n##### Inference accuracy: NVIDIA T4\nOur results were obtained by running the `translate.py` script in the\npytorch-20.06-py3 NGC Docker container with NVIDIA Tesla T4. Full command to\nlaunch the inference accuracy benchmark was provided in the [Inference\nperformance benchmark](#inference-performance-benchmark) section.\n\n| **Batch Size** | **Beam Size** | **Accuracy - FP32 (BLEU)** | **Accuracy - FP16 (BLEU)** |\n| -------------: | ------------: | -------------------------: | -------------------------: |\n| 128 | 1 | 23.07 | 23.08 |\n| 128 | 2 | 23.81 | 23.80 |\n| 128 | 5 | 24.40 | 24.39 |\n\nTo achieve these same results, follow the [Quick Start Guide](#quick-start-guide)\noutlined above.\n\n#### Inference throughput results\nTables presented in this section show the average inference throughput (columns\n**Avg (tok/s)**) and inference throughput for various confidence intervals\n(columns **N% (ms)**, where `N` denotes the confidence interval). Inference\nthroughput is measured in tokens per second. Speedups reported in FP16\nsubsections are relative to FP32 (for NVIDIA Volta and NVIDIA Turing) and\nrelative to TF32 (for NVIDIA Ampere) numbers for corresponding configuration.\n\n##### Inference throughput: NVIDIA A100 40GB\nOur results were obtained by running the `translate.py` script in the\npytorch-20.06-py3 NGC Docker container with NVIDIA A100 40GB.\nFull command to launch the inference throughput benchmark was provided in the\n[Inference performance benchmark](#inference-performance-benchmark) section.\n\n**FP16**\n\n|**Batch Size**|**Beam Size**|**Avg (tok/s)**|**Speedup**|**90% (tok/s)**|**Speedup**|**95% (tok/s)**|**Speedup**|**99% (tok/s)**|**Speedup**|\n|-------------:|------------:|--------------:|----------:|--------------:|----------:|--------------:|----------:|--------------:|----------:|\n| 1| 1| 1291.6| 1.031| 1195.7| 1.029| 1165.8| 1.029| 1104.7| 1.030|\n| 1| 2| 882.7| 1.019| 803.4| 1.015| 769.2| 1.015| 696.7| 1.017|\n| 1| 5| 848.3| 1.042| 753.0| 1.037| 715.0| 1.043| 636.4| 1.033|\n| 2| 1| 2060.5| 1.034| 1700.8| 1.032| 1621.8| 1.032| 1487.4| 1.022|\n| 2| 2| 1445.7| 1.026| 1197.6| 1.024| 1132.5| 1.023| 1043.7| 1.033|\n| 2| 5| 1402.3| 1.063| 1152.4| 1.056| 1100.5| 1.053| 992.9| 1.053|\n| 4| 1| 3465.6| 1.046| 2838.3| 1.040| 2672.7| 1.043| 2392.8| 1.043|\n| 4| 2| 2425.4| 1.041| 2002.5| 1.028| 1898.3| 1.033| 1690.2| 1.028|\n| 4| 5| 2364.4| 1.075| 1930.0| 1.067| 1822.0| 1.065| 1626.1| 1.058|\n| 8| 1| 6151.1| 1.099| 5078.0| 1.087| 4786.5| 1.096| 4206.9| 1.090|\n| 8| 2| 4241.9| 1.075| 3494.1| 1.066| 3293.6| 1.066| 2970.9| 1.064|\n| 8| 5| 4117.7| 1.118| 3430.9| 1.103| 3224.5| 1.104| 2833.5| 1.110|\n| 32| 1| 18830.4| 1.147| 16210.0| 1.152| 15563.9| 1.138| 13973.2| 1.135|\n| 32| 2| 12698.2| 1.133| 10812.3| 1.114| 10256.1| 1.145| 9330.2| 1.101|\n| 32| 5| 11802.6| 1.355| 9998.8| 1.318| 9671.6| 1.329| 9058.4| 1.335|\n| 128| 1| 53394.5| 1.350| 48867.6| 1.342| 46898.5| 1.414| 40670.6| 1.305|\n| 128| 2| 34876.4| 1.483| 31687.4| 1.491| 30025.4| 1.505| 27677.1| 1.421|\n| 128| 5| 28201.3| 1.986| 25660.5| 1.997| 24306.0| 1.967| 23326.2| 2.007|\n| 512| 1| 119675.3| 1.904| 112400.5| 1.971| 109694.8| 1.927| 108781.3| 1.919|\n| 512| 2| 74514.7| 2.126| 69578.9| 2.209| 69348.1| 2.210| 69253.7| 2.212|\n| 512| 5| 47003.2| 2.760| 43348.2| 2.893| 43080.3| 2.884| 42878.4| 2.881|\n\n##### Inference throughput: NVIDIA T4\nOur results were obtained by running the `translate.py` script in the\npytorch-20.06-py3 NGC Docker container with NVIDIA T4.\nFull command to launch the inference throughput benchmark was provided in the\n[Inference performance benchmark](#inference-performance-benchmark) section.\n\n**FP16**\n\n|**Batch Size**|**Beam Size**|**Avg (tok/s)**|**Speedup**|**90% (tok/s)**|**Speedup**|**95% (tok/s)**|**Speedup**|**99% (tok/s)**|**Speedup**|\n|-------------:|------------:|--------------:|----------:|--------------:|----------:|--------------:|----------:|--------------:|----------:|\n| 1| 1| 1133.8| 1.266| 1059.1| 1.253| 1036.6| 1.251| 989.5| 1.242|\n| 1| 2| 793.9| 1.169| 728.3| 1.165| 698.1| 1.163| 637.1| 1.157|\n| 1| 5| 766.8| 1.343| 685.6| 1.335| 649.3| 1.335| 584.1| 1.318|\n| 2| 1| 1759.8| 1.233| 1461.6| 1.239| 1402.3| 1.242| 1302.1| 1.242|\n| 2| 2| 1313.3| 1.186| 1088.7| 1.185| 1031.6| 1.180| 953.2| 1.178|\n| 2| 5| 1257.2| 1.301| 1034.1| 1.316| 990.3| 1.313| 886.3| 1.265|\n| 4| 1| 2974.0| 1.261| 2440.3| 1.255| 2294.6| 1.257| 2087.7| 1.261|\n| 4| 2| 2204.7| 1.320| 1826.3| 1.283| 1718.9| 1.260| 1548.4| 1.260|\n| 4| 5| 2106.1| 1.340| 1727.8| 1.345| 1625.7| 1.353| 1467.7| 1.346|\n| 8| 1| 5076.6| 1.423| 4207.9| 1.367| 3904.4| 1.360| 3475.3| 1.355|\n| 8| 2| 3761.7| 1.311| 3108.1| 1.285| 2931.6| 1.300| 2628.7| 1.300|\n| 8| 5| 3578.2| 1.660| 2998.2| 1.614| 2812.1| 1.609| 2447.6| 1.523|\n| 32| 1| 14637.8| 1.636| 12702.5| 1.644| 12070.3| 1.634| 11036.9| 1.647|\n| 32| 2| 10627.3| 1.818| 9198.3| 1.818| 8431.6| 1.725| 8000.0| 1.773|\n| 32| 5| 8205.7| 2.598| 7117.6| 2.476| 6825.2| 2.497| 6293.2| 2.437|\n| 128| 1| 33800.5| 2.755| 30824.5| 2.816| 27685.2| 2.661| 26580.9| 2.694|\n| 128| 2| 20829.4| 2.795| 18665.2| 2.778| 17372.1| 2.639| 16820.5| 2.821|\n| 128| 5| 11753.9| 3.309| 10658.1| 3.273| 10308.7| 3.205| 9630.7| 3.328|\n| 512| 1| 44474.6| 3.327| 40108.1| 3.394| 39816.6| 3.378| 39708.0| 3.381|\n| 512| 2| 26057.9| 3.295| 23197.3| 3.294| 23019.8| 3.284| 22951.4| 3.284|\n| 512| 5| 12161.5| 3.428| 10777.5| 3.418| 10733.1| 3.414| 10710.5| 3.420|\n\nTo achieve these same results, follow the [Quick Start Guide](#quick-start-guide)\noutlined above.\n\n#### Inference latency results\nTables presented in this section show the average inference latency (columns **Avg\n(ms)**) and inference latency for various confidence intervals (columns **N%\n(ms)**, where `N` denotes the confidence interval). Inference latency is\nmeasured in milliseconds. Speedups reported in FP16 subsections are relative to\nFP32 (for NVIDIA Volta and NVIDIA Turing) and relative to TF32 (for NVIDIA\nAmpere) numbers for corresponding configuration.\n\n##### Inference latency: NVIDIA A100 40GB\nOur results were obtained by running the `translate.py` script in the\npytorch-20.06-py3 NGC Docker container with NVIDIA A100 40GB.\nFull command to launch the inference latency benchmark was provided in the\n[Inference performance benchmark](#inference-performance-benchmark) section.\n\n**FP16**\n\n|**Batch Size**|**Beam Size**|**Avg (ms)**|**Speedup**|**90% (ms)**|**Speedup**|**95% (ms)**|**Speedup**|**99% (ms)**|**Speedup**|\n|-------------:|------------:|-----------:|----------:|-----------:|----------:|-----------:|----------:|-----------:|----------:|\n| 1| 1| 44.69| 1.032| 74.04| 1.035| 84.61| 1.034| 99.14| 1.042|\n| 1| 2| 64.76| 1.020| 105.18| 1.018| 118.92| 1.019| 139.42| 1.023|\n| 1| 5| 67.06| 1.043| 107.56| 1.049| 121.82| 1.054| 143.85| 1.054|\n| 2| 1| 56.57| 1.034| 85.59| 1.037| 92.55| 1.038| 107.59| 1.046|\n| 2| 2| 80.22| 1.027| 119.22| 1.027| 128.43| 1.030| 150.06| 1.028|\n| 2| 5| 82.54| 1.063| 121.37| 1.067| 132.35| 1.069| 156.34| 1.059|\n| 4| 1| 67.29| 1.047| 92.69| 1.048| 100.08| 1.056| 112.63| 1.064|\n| 4| 2| 95.86| 1.041| 129.83| 1.040| 139.48| 1.044| 162.34| 1.045|\n| 4| 5| 98.34| 1.075| 133.83| 1.076| 142.70| 1.068| 168.30| 1.075|\n| 8| 1| 75.60| 1.099| 97.87| 1.103| 104.13| 1.099| 117.40| 1.102|\n| 8| 2| 109.38| 1.074| 137.71| 1.079| 147.69| 1.069| 168.79| 1.065|\n| 8| 5| 112.71| 1.116| 143.50| 1.104| 153.17| 1.118| 172.60| 1.113|\n| 32| 1| 98.40| 1.146| 117.02| 1.153| 123.42| 1.150| 129.01| 1.128|\n| 32| 2| 145.87| 1.133| 171.71| 1.159| 184.01| 1.127| 188.64| 1.141|\n| 32| 5| 156.82| 1.357| 189.10| 1.374| 194.95| 1.392| 196.65| 1.419|\n| 128| 1| 137.97| 1.350| 150.04| 1.348| 151.52| 1.349| 154.52| 1.434|\n| 128| 2| 211.58| 1.484| 232.96| 1.490| 237.46| 1.505| 239.86| 1.567|\n| 128| 5| 261.44| 1.990| 288.54| 2.017| 291.63| 2.052| 298.73| 2.136|\n| 512| 1| 245.93| 1.906| 262.51| 1.998| 264.24| 1.999| 265.23| 2.000|\n| 512| 2| 395.61| 2.129| 428.54| 2.219| 431.58| 2.224| 433.86| 2.227|\n| 512| 5| 627.21| 2.767| 691.72| 2.878| 696.01| 2.895| 702.13| 2.887|\n\n##### Inference latency: NVIDIA T4\nOur results were obtained by running the `translate.py` script in the\npytorch-20.06-py3 NGC Docker container with NVIDIA T4.\nFull command to launch the inference latency benchmark was provided in the\n[Inference performance benchmark](#inference-performance-benchmark) section.\n\n**FP16**\n\n|**Batch Size**|**Beam Size**|**Avg (ms)**|**Speedup**|**90% (ms)**|**Speedup**|**95% (ms)**|**Speedup**|**99% (ms)**|**Speedup**|\n|-------------:|------------:|-----------:|----------:|-----------:|----------:|-----------:|----------:|-----------:|----------:|\n| 1| 1| 51.08| 1.261| 84.82| 1.254| 97.45| 1.251| 114.6| 1.257|\n| 1| 2| 72.05| 1.168| 117.41| 1.165| 132.33| 1.170| 155.8| 1.174|\n| 1| 5| 74.20| 1.345| 119.45| 1.352| 135.07| 1.354| 160.3| 1.354|\n| 2| 1| 66.31| 1.232| 100.90| 1.232| 108.52| 1.235| 126.9| 1.238|\n| 2| 2| 88.35| 1.185| 131.47| 1.188| 141.46| 1.185| 164.7| 1.191|\n| 2| 5| 92.12| 1.305| 136.30| 1.310| 148.66| 1.309| 174.8| 1.320|\n| 4| 1| 78.54| 1.260| 108.53| 1.256| 117.19| 1.259| 133.7| 1.259|\n| 4| 2| 105.54| 1.315| 142.74| 1.317| 154.36| 1.307| 178.7| 1.303|\n| 4| 5| 110.43| 1.351| 150.62| 1.388| 161.61| 1.397| 191.2| 1.427|\n| 8| 1| 91.65| 1.418| 117.92| 1.421| 126.60| 1.405| 144.0| 1.411|\n| 8| 2| 123.39| 1.315| 156.00| 1.337| 167.34| 1.347| 193.4| 1.340|\n| 8| 5| 129.69| 1.666| 165.01| 1.705| 178.18| 1.723| 200.3| 1.765|\n| 32| 1| 126.53| 1.641| 153.23| 1.689| 159.58| 1.692| 167.0| 1.700|\n| 32| 2| 174.37| 1.822| 209.04| 1.899| 219.59| 1.877| 228.6| 1.878|\n| 32| 5| 226.15| 2.598| 277.38| 2.636| 290.27| 2.648| 299.4| 2.664|\n| 128| 1| 218.29| 2.755| 238.94| 2.826| 243.18| 2.843| 267.1| 2.828|\n| 128| 2| 354.83| 2.796| 396.63| 2.832| 410.53| 2.803| 433.2| 2.866|\n| 128| 5| 628.32| 3.311| 699.57| 3.353| 723.98| 3.323| 771.0| 3.337|\n| 512| 1| 663.07| 3.330| 748.62| 3.388| 753.20| 3.388| 758.0| 3.378|\n| 512| 2| 1134.04| 3.295| 1297.85| 3.283| 1302.25| 3.304| 1306.9| 3.308|\n| 512| 5| 2428.82| 3.428| 2771.72| 3.415| 2801.32| 3.427| 2817.6| 3.422|\n\nTo achieve these same results, follow the [Quick Start Guide](#quick-start-guide)\noutlined above.","hasPerformance",[],{"_85":181,"_112":234,"_113":-5,"_121":34,"_120":34,"_122":233,"_217":90},{"_85":28,"_112":28,"_113":-5,"_121":34,"_120":34,"_122":233,"_217":90},[],"Latest (20.06.0)","nvidia/dle/gnmt_pyt","key","nvidia/dle/resources/gnmt_pyt","resourceId","gnmt_pyt","shortDescription","The GNMT v2 model is an improved version of the first Google's Neural Machine Translation System with a modified attention mechanism.","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/gnmt-pyt)\n\nThe GNMT v2 model is similar to the one discussed in the [Google's Neural\nMachine Translation System: Bridging the Gap between Human and Machine\nTranslation](https://arxiv.org/abs/1609.08144) paper.\n\nThe most important difference between the two models is in the attention\nmechanism. In our model, the output from the first LSTM layer of the decoder\ngoes into the attention module, then the re-weighted context is concatenated\nwith inputs to all subsequent LSTM layers in the decoder at the current\ntime step.\n\nThe same attention mechanism is also implemented in the default GNMT-like\nmodels from [TensorFlow Neural Machine Translation\nTutorial](https://github.com/tensorflow/nmt) and [NVIDIA OpenSeq2Seq\nToolkit](https://github.com/NVIDIA/OpenSeq2Seq).\n\n### Model architecture\n![ModelArchitecture](https://raw.githubusercontent.com/NVIDIA/DeepLearningExamples/master/PyTorch/Translation/GNMT//img/diagram.png)\n\n\n### Default configuration\n\nThe following features were implemented in this model:\n\n* general:\n * encoder and decoder are using shared embeddings\n * data-parallel multi-GPU training\n * dynamic loss scaling with backoff for Tensor Cores (mixed precision)\n training\n * trained with label smoothing loss (smoothing factor 0.1)\n* encoder:\n * 4-layer LSTM, hidden size 1024, first layer is bidirectional, the rest are\n unidirectional\n * with residual connections starting from 3rd layer\n * uses standard PyTorch nn.LSTM layer\n * dropout is applied on input to all LSTM layers, probability of dropout is\n set to 0.2\n * hidden state of LSTM layers is initialized with zeros\n * weights and bias of LSTM layers is initialized with uniform(-0.1,0.1)\n distribution\n* decoder:\n * 4-layer unidirectional LSTM with hidden size 1024 and fully-connected\n classifier\n * with residual connections starting from 3rd layer\n * uses standard PyTorch nn.LSTM layer\n * dropout is applied on input to all LSTM layers, probability of dropout is\n set to 0.2\n * hidden state of LSTM layers is initialized with zeros\n * weights and bias of LSTM layers is initialized with uniform(-0.1,0.1)\n distribution\n * weights and bias of fully-connected classifier is initialized with\n uniform(-0.1,0.1) distribution\n* attention:\n * normalized Bahdanau attention\n * output from first LSTM layer of decoder goes into attention, then\n re-weighted context is concatenated with the input to all subsequent LSTM\n layers of the decoder at the current timestep\n * linear transform of keys and queries is initialized with uniform(-0.1,\n 0.1), normalization scalar is initialized with 1.0/sqrt(1024),\n normalization bias is initialized with zero\n* inference:\n * beam search with default beam size of 5\n * with coverage penalty and length normalization, coverage penalty factor is\n set to 0.1, length normalization factor is set to 0.6 and length\n normalization constant is set to 5.0\n * de-tokenized BLEU computed by\n [SacreBLEU](https://github.com/mjpost/sacrebleu)\n * [motivation](https://github.com/mjpost/sacrebleu#motivation) for choosing\n SacreBLEU\n\nWhen comparing the BLEU score, there are various tokenization approaches and\nBLEU calculation methodologies; therefore, ensure you align similar metrics.\n\nCode from this repository can be used to train a larger, 8-layer GNMT v2 model.\nOur experiments show that a 4-layer model is significantly faster to train and\nyields comparable accuracy on the public [WMT16\nEnglish-German](http://www.statmt.org/wmt16/translation-task.html) dataset. The\nnumber of LSTM layers is controlled by the `--num-layers` parameter in the\n`train.py` training script.\n\n### Feature support matrix\nThe following features are supported by this model.\n\n| **Feature** | **GNMT v2** |\n|:------------|------------:|\n|[Apex AMP](https://nvidia.github.io/apex/amp.html) | Yes |\n|[Apex DistributedDataParallel](https://nvidia.github.io/apex/parallel.html#apex.parallel.DistributedDataParallel) | Yes |\n\n#### Features\n[Apex AMP](https://nvidia.github.io/apex/amp.html) - a tool that enables Tensor\nCore-accelerated training. Refer to the [Enabling mixed\nprecision](#enabling-mixed-precision) section for more details.\n\n[Apex\nDistributedDataParallel](https://nvidia.github.io/apex/parallel.html#apex.parallel.DistributedDataParallel) -\na module wrapper that enables easy multiprocess distributed data parallel\ntraining, similar to\n[torch.nn.parallel.DistributedDataParallel](https://pytorch.org/docs/stable/nn.html#torch.nn.parallel.DistributedDataParallel).\n`DistributedDataParallel` is optimized for use with\n[NCCL](https://github.com/NVIDIA/nccl). It achieves high performance by\noverlapping communication with computation during `backward()` and bucketing\nsmaller gradient transfers to reduce the total number of transfers required.\n\n### Mixed precision training\nMixed precision is the combined use of different numerical precisions in a\ncomputational method.\n[Mixed precision](https://arxiv.org/abs/1710.03740) training offers significant\ncomputational speedup by performing operations in half-precision format, while\nstoring minimal information in single-precision to retain as much information\nas possible in critical parts of the network. Since the introduction of [Tensor\nCores](https://developer.nvidia.com/tensor-cores) in Volta, and following with\nboth the Turing and Ampere architectures, significant training speedups are\nexperienced by switching to mixed precision -- up to 3x overall speedup on the\nmost arithmetically intense model architectures. Using mixed precision training\npreviously required two steps:\n\n1. Porting the model to use the FP16 data type where appropriate.\n2. Manually adding loss scaling to preserve small gradient values.\n\nThe ability to train deep learning networks with lower precision was introduced\nin the Pascal architecture and first supported in [CUDA\n8](https://devblogs.nvidia.com/parallelforall/tag/fp16/) in the NVIDIA Deep\nLearning SDK.\n\nFor information about:\n* How to train using mixed precision, see the [Mixed Precision\n Training](https://arxiv.org/abs/1710.03740) paper and [Training With Mixed\n Precision](https://docs.nvidia.com/deeplearning/sdk/mixed-precision-training/index.html)\n documentation.\n* Techniques used for mixed precision training, see the [Mixed-Precision\n Training of Deep Neural\n Networks](https://devblogs.nvidia.com/mixed-precision-training-deep-neural-networks/)\n blog.\n* APEX tools for mixed precision training, see the [NVIDIA Apex: Tools for Easy\n Mixed-Precision Training in\n PyTorch](https://devblogs.nvidia.com/apex-pytorch-easy-mixed-precision-training/)\n .\n\n#### Enabling mixed precision\nMixed precision is enabled in PyTorch by using the Automatic Mixed Precision\n(AMP), library from [APEX](https://github.com/NVIDIA/apex) that casts variables\nto half-precision upon retrieval, while storing variables in single-precision\nformat. Furthermore, to preserve small gradient magnitudes in backpropagation,\na [loss\nscaling](https://docs.nvidia.com/deeplearning/sdk/mixed-precision-training/index.html#lossscaling)\nstep must be included when applying gradients. In PyTorch, loss scaling can be\neasily applied by using `scale_loss()` method provided by AMP. The scaling\nvalue to be used can be\n[dynamic](https://nvidia.github.io/apex/amp.html#apex.amp.initialize) or fixed.\n\nFor an in-depth walk through on AMP, check out sample usage\n[here](https://nvidia.github.io/apex/amp.html#).\n[APEX](https://github.com/NVIDIA/apex) is a PyTorch extension that contains\nutility libraries, such as AMP, which require minimal network code changes to\nleverage Tensor Cores performance.\n\nThe following steps were needed to enable mixed precision training in GNMT:\n\n* Import AMP from APEX (file: `seq2seq/train/trainer.py`):\n\n```\nfrom apex import amp\n```\n\n* Initialize AMP and wrap the model and the optimizer (file:\n `seq2seq/train/trainer.py`, class: `Seq2SeqTrainer`):\n\n```\nself.model, self.optimizer = amp.initialize(\n self.model,\n self.optimizer,\n cast_model_outputs=torch.float16,\n keep_batchnorm_fp32=False,\n opt_level='O2')\n```\n\n* Apply `scale_loss` context manager (file: `seq2seq/train/fp_optimizers.py`,\n class: `AMPOptimizer`):\n\n```\nwith amp.scale_loss(loss, optimizer) as scaled_loss:\n scaled_loss.backward()\n```\n\n* Apply gradient clipping on single precision master weights (file:\n `seq2seq/train/fp_optimizers.py`, class: `AMPOptimizer`):\n\n```\nif self.grad_clip != float('inf'):\n clip_grad_norm_(amp.master_params(optimizer), self.grad_clip)\n```\n\n#### Enabling TF32\nTensorFloat-32 (TF32) is the new math mode in [NVIDIA\nA100](https://www.nvidia.com/en-us/data-center/a100/) GPUs for handling the\nmatrix math also called tensor operations. TF32 running on Tensor Cores in A100\nGPUs can provide up to 10x speedups compared to single-precision floating-point\nmath (FP32) on Volta GPUs.\n\nTF32 Tensor Cores can speed up networks using FP32, typically with no loss of\naccuracy. It is more robust than FP16 for models which require high dynamic\nrange for weights or activations.\n\nFor more information, refer to the [TensorFloat-32 in the A100 GPU Accelerates\nAI Training, HPC up to\n20x](https://blogs.nvidia.com/blog/2020/05/14/tensorfloat-32-precision-format/)\nblog post.\n\nTF32 is supported in the NVIDIA Ampere GPU architecture and is enabled by\ndefault.","displayName","GNMT v2 for PyTorch",["D",1680631525404],"publisher","NVIDIA Deep Learning Examples","latestVersion","collectionCounts","imageUrl","/api/image/nvidia/dle/resources/gnmt_pyt","rawImageUrl","https://assets.nvidiagrid.net/ngc/logos/Deep-Learning-Examples.png","canGuestDownload","latestVersionSizeInBytes","licenseTerms","isMultinodeEnabled","isMultiArchitecture","bias","explainability","privacy","safetyAndSecurity","featureLabels",[],"displayLabels",[274],"policyLabels",[],"systemLabels",[],"products",[],"hasNimMetadata",{"_243":275,"_276":277,"_278":279,"_280":87},"Natural Language Understanding","value","uscs_natural_language_understanding","filter",{"_281":282,"_276":277},"category","field","application",{"_150":1261,"_152":1262},"ngcApps",[1214],"csaApiBaseUrl","https://www.nvidia.com/ai-agents/csa-agent/dcsm","datadogRum",{"_1195":1196,"_1197":1198,"_1199":1200,"_1201":1202,"_1203":294,"_109":320,"_1204":1205,"_1206":59,"_1207":90,"_1208":90,"_1209":1210,"_1211":90,"_1212":1213},"amplitude",{"_1192":1193,"_833":-7,"_1194":34},"disableJs","environment","prod","featureFlagsBlob","JTdCJTIydGVzdEZsYWclMjIlM0FmYWxzZSUyQyUyMnNob3dTZWFyY2hMYWJlbFNldEZpbHRlcnMlMjIlM0FmYWxzZSUyQyUyMmVuYWJsZUNzYVdpZGdldCUyMiUzQWZhbHNlJTJDJTIyZW5hYmxlUHJvZHVjdEdyb3VwaW5nQnlTdWJzY3JpcHRpb24lMjIlM0FmYWxzZSUyQyUyMmNvbnRhaW5lclNjYW5uaW5nUmVkZXNpZ25NMiUyMiUzQWZhbHNlJTdE","featureFlagsGuarded","isDevLikeEnv","isPlaywright","isMobileDevice","categorizedOrgs",{"_1186":1187,"_1188":1189,"_1190":1191},"currentOrgContext","currentContextRoles",[],"enterpriseOrgs",[],"manageableExternalUserGroups",[],"ngcApiOrigin","ngcHelmOrigin","origin","https://catalog.ngc.nvidia.com","usesMocks","theme","year",2026,"now",1787694353969,"1.12.0","defaultSearchSuggestions",{"_1073":400,"_1074":1075,"_1076":1077,"_1078":1079},"announcements",[],"consentState",{"_357":1063,"_1064":90,"_1065":34,"_1066":34,"_1067":1068},"consentConfig",{"_339":340,"_341":342,"_343":344,"_345":346,"_347":348,"_349":350,"_351":352,"_353":354,"_355":356,"_357":358},"showContactAdmin",1787694353973,[],[],[334,335,336,337,338],{"_48":63,"_50":73,"_52":69,"_54":-5,"_56":-5,"_58":59},{"_48":63,"_50":70,"_52":69,"_54":-5,"_56":-5,"_58":59},{"_48":63,"_50":64,"_52":69,"_54":-5,"_56":-5,"_58":59},{"_48":60,"_50":61,"_52":69,"_54":-5,"_56":-5,"_58":59},{"_48":49,"_50":51,"_52":69,"_54":-5,"_56":-5,"_58":59},"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",{"_1055":1056,"_1057":1058,"_1059":1060,"_1061":1062},"bannerAcceptLabel","Accept All","bannerRejectLabel","Reject Optional","bannerSettingsLabel","Manage Settings","preferenceSaveLabel","Save Choices","preferenceRejectLabel","Decline All","alwaysActiveText","Always Active","groups",[359,360,361,362,363],{"_364":745,"_112":746,"_211":747,"_368":90,"_369":16,"_371":748,"_373":749},{"_364":648,"_112":649,"_211":650,"_368":34,"_369":57,"_371":651,"_373":652},{"_364":555,"_112":556,"_211":557,"_368":34,"_369":558,"_371":559,"_373":560},{"_364":375,"_112":376,"_211":377,"_368":34,"_369":378,"_371":379,"_373":380},{"_364":365,"_112":366,"_211":367,"_368":34,"_369":370,"_371":372,"_373":374},"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,[497,498,499,500,501,502,503,504,505,506,507,508,509,510,511,512,513,514,515,516,517,518],[381,382,383,384,385,386,387,388,389,390,391,392,393,394,395,396,397],{"_398":492,"_243":492,"_211":400,"_371":493},{"_398":488,"_243":488,"_211":400,"_371":489},{"_398":483,"_243":483,"_211":400,"_371":484},{"_398":479,"_243":479,"_211":400,"_371":480},{"_398":474,"_243":474,"_211":400,"_371":475},{"_398":469,"_243":469,"_211":400,"_371":470},{"_398":461,"_243":461,"_211":400,"_371":462},{"_398":453,"_243":453,"_211":400,"_371":454},{"_398":448,"_243":448,"_211":400,"_371":449},{"_398":443,"_243":443,"_211":400,"_371":444},{"_398":438,"_243":438,"_211":400,"_371":439},{"_398":433,"_243":433,"_211":400,"_371":434},{"_398":428,"_243":428,"_211":400,"_371":429},{"_398":421,"_243":421,"_211":400,"_371":422},{"_398":416,"_243":416,"_211":400,"_371":417},{"_398":409,"_243":409,"_211":400,"_371":410},{"_398":399,"_243":399,"_211":400,"_371":401},"hostName","amazon-adsystem.com","",[402,403],{"_112":408,"_405":399,"_406":34,"_211":407},{"_112":404,"_405":399,"_406":34,"_211":407},"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",[411,412],{"_112":415,"_405":409,"_406":34,"_211":414},{"_112":413,"_405":409,"_406":34,"_211":414},"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",[418],{"_112":419,"_405":416,"_406":34,"_211":420},"MUID","This domain is owned by Mircosoft - it is the site for the search engine Bing.","assets.adobedtm.com",[423,424],{"_112":427,"_405":421,"_406":34,"_211":426},{"_112":425,"_405":421,"_406":34,"_211":426},"kndctr_F207D74D549850760A4C98C6_AdobeOrg_*","If you know anything about this host and its uses please add your contributions here. ","kndctr_F207D74D549850760A4C98C6_AdobeOrg_consent","tiktok.com",[430],{"_112":431,"_405":428,"_406":34,"_211":432},"_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",[435],{"_112":436,"_405":433,"_406":34,"_211":437},"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",[440],{"_112":441,"_405":438,"_406":34,"_211":442},"sp","Testing","google.com",[445],{"_112":446,"_405":443,"_406":34,"_211":447},"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",[450],{"_112":451,"_405":448,"_406":34,"_211":452},"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",[455,456],{"_112":459,"_405":453,"_406":34,"_211":460},{"_112":457,"_405":453,"_406":34,"_211":458},"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",[463,464],{"_112":467,"_405":461,"_406":34,"_211":468},{"_112":465,"_405":461,"_406":34,"_211":466},"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",[471],{"_112":472,"_405":469,"_406":34,"_211":473},"demdex","This cookie helps Adobe Audience Manger perform basic functions such as visitor identification, ID synchronization, segmentation, modeling, reporting, etc.","im-apps.net",[476],{"_112":477,"_405":474,"_406":34,"_211":478},"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",[481],{"_112":482,"_405":479,"_406":34,"_211":442},"R","6sc.co",[485],{"_112":486,"_405":483,"_406":34,"_211":487},"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",[490],{"_112":491,"_405":488,"_406":34,"_211":442},"_hly_gvid","t.contentsquare.net",[494],{"_112":495,"_405":492,"_406":90,"_211":496},"_cs_*","This is a session cookie. This is a pattern type cookie where a random 13-digit number is appended to the prefix _cs_.",{"_112":553,"_405":520,"_406":34,"_211":554},{"_112":551,"_405":520,"_406":34,"_211":552},{"_112":549,"_405":520,"_406":34,"_211":550},{"_112":431,"_405":520,"_406":34,"_211":548},{"_112":525,"_405":520,"_406":90,"_211":527},{"_112":547,"_405":520,"_406":34,"_211":400},{"_112":546,"_405":523,"_406":34,"_211":400},{"_112":544,"_405":520,"_406":34,"_211":545},{"_112":542,"_405":520,"_406":34,"_211":543},{"_112":541,"_405":520,"_406":34,"_211":400},{"_112":539,"_405":520,"_406":34,"_211":540},{"_112":525,"_405":538,"_406":90,"_211":527},{"_112":536,"_405":523,"_406":34,"_211":537},{"_112":536,"_405":526,"_406":34,"_211":537},{"_112":495,"_405":520,"_406":34,"_211":496},{"_112":534,"_405":520,"_406":34,"_211":535},{"_112":441,"_405":532,"_406":34,"_211":533},{"_112":530,"_405":520,"_406":34,"_211":531},{"_112":528,"_405":520,"_406":34,"_211":529},{"_112":525,"_405":526,"_406":90,"_211":527},{"_112":522,"_405":523,"_406":34,"_211":524},{"_112":519,"_405":520,"_406":90,"_211":521},"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,[590,591,592,593,594,595,596,597,598,599,600,601,602,603,604,605,606,607,608,609,610,611,612,613,614],[561,562,563,564,565],{"_398":474,"_243":474,"_211":400,"_371":585},{"_398":448,"_243":448,"_211":400,"_371":580},{"_398":575,"_243":575,"_211":400,"_371":576},{"_398":570,"_243":570,"_211":400,"_371":571},{"_398":566,"_243":566,"_211":400,"_371":567},"nvidia.partners",[568],{"_112":569,"_405":566,"_406":34,"_211":442},"edge","app.hushly.com",[572],{"_112":573,"_405":570,"_406":34,"_211":574},"*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",[577],{"_112":578,"_405":575,"_406":34,"_211":579},"_octo","This domain is owned by Github. The main business activity is: Source code repository and collaboration platform",[581,582],{"_112":584,"_405":448,"_406":34,"_211":452},{"_112":583,"_405":448,"_406":34,"_211":452},"lidc","li_gc",[586,587],{"_112":589,"_405":474,"_406":34,"_211":478},{"_112":588,"_405":474,"_406":34,"_211":478},"imid_*","imid",{"_112":615,"_405":523,"_406":34,"_211":617},{"_112":646,"_405":520,"_406":34,"_211":647},{"_112":615,"_405":645,"_406":34,"_211":617},{"_112":643,"_405":520,"_406":90,"_211":644},{"_112":641,"_405":520,"_406":34,"_211":642},{"_112":640,"_405":623,"_406":90,"_211":400},{"_112":615,"_405":619,"_406":34,"_211":617},{"_112":638,"_405":520,"_406":34,"_211":639},{"_112":615,"_405":526,"_406":34,"_211":617},{"_112":626,"_405":523,"_406":34,"_211":400},{"_112":615,"_405":637,"_406":34,"_211":617},{"_112":635,"_405":520,"_406":34,"_211":636},{"_112":632,"_405":633,"_406":34,"_211":634},{"_112":630,"_405":631,"_406":34,"_211":400},{"_112":615,"_405":627,"_406":34,"_211":617},{"_112":629,"_405":520,"_406":34,"_211":400},{"_112":615,"_405":628,"_406":34,"_211":617},{"_112":626,"_405":627,"_406":34,"_211":400},{"_112":625,"_405":623,"_406":90,"_211":400},{"_112":615,"_405":624,"_406":34,"_211":617},{"_112":615,"_405":623,"_406":34,"_211":617},{"_112":615,"_405":622,"_406":34,"_211":617},{"_112":621,"_405":520,"_406":90,"_211":617},{"_112":618,"_405":619,"_406":90,"_211":620},{"_112":615,"_405":616,"_406":34,"_211":617},"_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.",[673,674,675,676,677,678,679,680,681,682,683,684,685,686,687,688,689,690,691,692,693,694,695,696,697,698,699,700,701],[653,654,655,656],{"_398":668,"_243":668,"_211":400,"_371":669},{"_398":663,"_243":663,"_211":400,"_371":664},{"_398":461,"_243":461,"_211":400,"_371":660},{"_398":575,"_243":575,"_211":400,"_371":657},[658],{"_112":659,"_405":575,"_406":90,"_211":579},"_gh_sess",[661],{"_112":662,"_405":461,"_406":90,"_211":468},"YSC","bizible.com",[665],{"_112":666,"_405":663,"_406":34,"_211":667},"_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",[670],{"_112":671,"_405":668,"_406":34,"_211":672},"tuid","This domain is owned by Ookla which provides internet speed testing services through its Speedtest application.",{"_112":743,"_405":520,"_406":34,"_211":744},{"_112":728,"_405":742,"_406":90,"_211":729},{"_112":711,"_405":523,"_406":34,"_211":636},{"_112":711,"_405":624,"_406":34,"_211":636},{"_112":740,"_405":520,"_406":34,"_211":741},{"_112":728,"_405":526,"_406":90,"_211":729},{"_112":738,"_405":520,"_406":34,"_211":739},{"_112":728,"_405":523,"_406":90,"_211":729},{"_112":736,"_405":623,"_406":34,"_211":737},{"_112":728,"_405":645,"_406":90,"_211":729},{"_112":708,"_405":523,"_406":34,"_211":710},{"_112":734,"_405":520,"_406":90,"_211":735},{"_112":732,"_405":520,"_406":34,"_211":733},{"_112":730,"_405":520,"_406":90,"_211":731},{"_112":728,"_405":619,"_406":90,"_211":729},{"_112":726,"_405":520,"_406":34,"_211":727},{"_112":724,"_405":520,"_406":34,"_211":725},{"_112":722,"_405":520,"_406":90,"_211":723},{"_112":720,"_405":628,"_406":34,"_211":721},{"_112":708,"_405":619,"_406":34,"_211":710},{"_112":718,"_405":520,"_406":34,"_211":719},{"_112":716,"_405":520,"_406":90,"_211":717},{"_112":714,"_405":520,"_406":34,"_211":715},{"_112":712,"_405":520,"_406":34,"_211":713},{"_112":711,"_405":520,"_406":34,"_211":636},{"_112":708,"_405":709,"_406":34,"_211":710},{"_112":706,"_405":520,"_406":34,"_211":707},{"_112":704,"_405":520,"_406":90,"_211":705},{"_112":702,"_405":520,"_406":34,"_211":703},"_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.",[890,891,892,893,894,895,896,897,898,899,900,901,902,903,904,905,906,907,908,909,910,911,912,913,914,915,916,917,918,919,920,921,922,923,924,925,926,927,928,929,930,931,932,933,934,935,936,937,938,939,940,941,942,943,944,945,946,947,948,949,950,951,952,953,954,955,956,957,958,959,960,961,962],[750,751,752,753,754,755,756,757,758,759,760,761,762,763,764,765,766,767,768,769,770,771,772,773,774],{"_398":887,"_243":887,"_211":400,"_371":888},{"_398":881,"_243":881,"_211":400,"_371":882},{"_398":878,"_243":878,"_211":400,"_371":879},{"_398":866,"_243":866,"_211":400,"_371":867},{"_398":863,"_243":863,"_211":400,"_371":864},{"_398":860,"_243":860,"_211":400,"_371":861},{"_398":855,"_243":855,"_211":400,"_371":856},{"_398":461,"_243":461,"_211":400,"_371":852},{"_398":849,"_243":849,"_211":400,"_371":850},{"_398":453,"_243":453,"_211":400,"_371":847},{"_398":448,"_243":448,"_211":400,"_371":845},{"_398":840,"_243":840,"_211":400,"_371":841},{"_398":835,"_243":835,"_211":400,"_371":836},{"_398":830,"_243":830,"_211":400,"_371":831},{"_398":827,"_243":827,"_211":400,"_371":828},{"_398":575,"_243":575,"_211":400,"_371":824},{"_398":433,"_243":433,"_211":400,"_371":821},{"_398":570,"_243":570,"_211":400,"_371":814},{"_398":566,"_243":566,"_211":400,"_371":808},{"_398":802,"_243":802,"_211":400,"_371":803},{"_398":797,"_243":797,"_211":400,"_371":798},{"_398":792,"_243":792,"_211":400,"_371":793},{"_398":788,"_243":788,"_211":400,"_371":789},{"_398":783,"_243":783,"_211":400,"_371":784},{"_398":775,"_243":775,"_211":400,"_371":776},"polyfill.io",[777,778],{"_112":781,"_405":775,"_406":34,"_211":782},{"_112":779,"_405":775,"_406":34,"_211":780},"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",[785],{"_112":786,"_405":783,"_406":34,"_211":787},"_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",[790],{"_112":791,"_405":788,"_406":34,"_211":442},"__cflb","instagram.com",[794],{"_112":795,"_405":792,"_406":34,"_211":796},"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",[799],{"_112":800,"_405":797,"_406":34,"_211":801},"__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",[804,805],{"_112":807,"_405":802,"_406":34,"_211":442},{"_112":806,"_405":802,"_406":34,"_211":442},"AWSALBTG","AWSALBTGCORS",[809,810],{"_112":812,"_405":566,"_406":34,"_211":813},{"_112":811,"_405":566,"_406":34,"_211":442},"ak_bmsc","bm_mi","This domain is likely associated with NVIDIA's partner program, which could involve marketing, sales, and support services for partners.",[815,816,817],{"_112":820,"_405":570,"_406":34,"_211":442},{"_112":819,"_405":570,"_406":90,"_211":442},{"_112":818,"_405":570,"_406":34,"_211":442},"AWSALBCORS","JSESSIONID","AWSALB",[822],{"_112":800,"_405":433,"_406":34,"_211":823},"This is a CloudFoundry cookie",[825],{"_112":826,"_405":575,"_406":34,"_211":579},"logged_in","hcaptcha.com",[829],{"_112":800,"_405":827,"_406":34,"_211":823},"tracking.intentsify.io",[832],{"_112":833,"_405":830,"_406":34,"_211":834},"userId","Used to track usage of the embedded video player.","static.reo.dev",[837],{"_112":838,"_405":835,"_406":34,"_211":839},"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",[842],{"_112":843,"_405":840,"_406":34,"_211":844},"dd_cookie_test_","dd_cookie_test",[846],{"_112":800,"_405":448,"_406":34,"_211":823},[848],{"_112":800,"_405":453,"_406":34,"_211":823},"supabase.co",[851],{"_112":800,"_405":849,"_406":34,"_211":823},[853],{"_112":854,"_405":461,"_406":34,"_211":468},"VISITOR_PRIVACY_METADATA","dosa-prd.azurewebsites.net",[857],{"_112":858,"_405":855,"_406":34,"_211":859},"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",[862],{"_112":800,"_405":860,"_406":34,"_211":823},"cdn.readme.io",[865],{"_112":800,"_405":863,"_406":34,"_211":823},"queue-it.net",[868,869,870,871,872],{"_112":877,"_405":866,"_406":34,"_211":400},{"_112":876,"_405":866,"_406":34,"_211":400},{"_112":875,"_405":866,"_406":34,"_211":400},{"_112":874,"_405":866,"_406":34,"_211":400},{"_112":873,"_405":866,"_406":34,"_211":400},"Queue-it-{queueId}","Queue-it-{dmonvidia }________{eventId}","Queue-it-token","queueitsoftblock_{queueId}","Queue-it-{nvidia}________{eventId}","x.com",[880],{"_112":800,"_405":878,"_406":34,"_211":801},"api.hcaptcha.com",[883,884],{"_112":886,"_405":881,"_406":34,"_211":442},{"_112":791,"_405":881,"_406":34,"_211":885},"This domain is associated with hCaptcha, a service that provides CAPTCHA and reCAPTCHA alternatives for user verification and bot protection.","hmt_id","vimeo.com",[889],{"_112":800,"_405":887,"_406":34,"_211":823},{"_112":1023,"_405":520,"_406":34,"_211":1054},{"_112":1052,"_405":616,"_406":90,"_211":1053},{"_112":1045,"_405":1014,"_406":90,"_211":1046},{"_112":1051,"_405":709,"_406":90,"_211":400},{"_112":1049,"_405":520,"_406":34,"_211":1050},{"_112":807,"_405":1035,"_406":34,"_211":974},{"_112":982,"_405":1044,"_406":90,"_211":984},{"_112":806,"_405":973,"_406":34,"_211":974},{"_112":985,"_405":1044,"_406":34,"_211":986},{"_112":1047,"_405":520,"_406":90,"_211":1048},{"_112":1045,"_405":1021,"_406":90,"_211":1046},{"_112":979,"_405":1044,"_406":34,"_211":981},{"_112":1042,"_405":520,"_406":34,"_211":1043},{"_112":806,"_405":965,"_406":34,"_211":974},{"_112":970,"_405":1041,"_406":34,"_211":972},{"_112":818,"_405":971,"_406":34,"_211":987},{"_112":820,"_405":971,"_406":34,"_211":1038},{"_112":985,"_405":980,"_406":34,"_211":986},{"_112":1039,"_405":619,"_406":34,"_211":1040},{"_112":820,"_405":645,"_406":34,"_211":1038},{"_112":1036,"_405":520,"_406":34,"_211":1037},{"_112":812,"_405":645,"_406":34,"_211":1031},{"_112":1008,"_405":624,"_406":90,"_211":400},{"_112":806,"_405":1035,"_406":34,"_211":974},{"_112":1033,"_405":973,"_406":90,"_211":1034},{"_112":997,"_405":520,"_406":34,"_211":988},{"_112":970,"_405":520,"_406":34,"_211":972},{"_112":1032,"_405":532,"_406":34,"_211":400},{"_112":812,"_405":520,"_406":34,"_211":1031},{"_112":820,"_405":965,"_406":34,"_211":1030},{"_112":1028,"_405":628,"_406":90,"_211":1029},{"_112":1026,"_405":520,"_406":90,"_211":1027},{"_112":1025,"_405":520,"_406":34,"_211":400},{"_112":1023,"_405":971,"_406":34,"_211":1024},{"_112":1022,"_405":532,"_406":34,"_211":400},{"_112":1013,"_405":1021,"_406":90,"_211":1015},{"_112":1019,"_405":624,"_406":34,"_211":1020},{"_112":1017,"_405":709,"_406":34,"_211":1018},{"_112":1016,"_405":526,"_406":34,"_211":400},{"_112":1013,"_405":1014,"_406":90,"_211":1015},{"_112":1011,"_405":980,"_406":90,"_211":1012},{"_112":1009,"_405":520,"_406":34,"_211":1010},{"_112":1008,"_405":523,"_406":90,"_211":400},{"_112":979,"_405":983,"_406":34,"_211":981},{"_112":811,"_405":998,"_406":34,"_211":988},{"_112":1006,"_405":523,"_406":34,"_211":1007},{"_112":1004,"_405":520,"_406":34,"_211":1005},{"_112":1002,"_405":520,"_406":34,"_211":1003},{"_112":1001,"_405":520,"_406":34,"_211":400},{"_112":843,"_405":523,"_406":34,"_211":1000},{"_112":997,"_405":645,"_406":34,"_211":988},{"_112":800,"_405":616,"_406":34,"_211":999},{"_112":997,"_405":998,"_406":34,"_211":988},{"_112":996,"_405":532,"_406":34,"_211":400},{"_112":995,"_405":520,"_406":34,"_211":968},{"_112":993,"_405":980,"_406":34,"_211":994},{"_112":806,"_405":991,"_406":34,"_211":992},{"_112":990,"_405":532,"_406":34,"_211":400},{"_112":989,"_405":520,"_406":34,"_211":400},{"_112":982,"_405":980,"_406":90,"_211":984},{"_112":811,"_405":645,"_406":34,"_211":988},{"_112":818,"_405":645,"_406":34,"_211":987},{"_112":985,"_405":983,"_406":34,"_211":986},{"_112":982,"_405":983,"_406":90,"_211":984},{"_112":979,"_405":980,"_406":34,"_211":981},{"_112":977,"_405":973,"_406":90,"_211":978},{"_112":975,"_405":631,"_406":34,"_211":976},{"_112":807,"_405":973,"_406":34,"_211":974},{"_112":970,"_405":971,"_406":34,"_211":972},{"_112":800,"_405":623,"_406":34,"_211":969},{"_112":967,"_405":520,"_406":34,"_211":968},{"_112":818,"_405":965,"_406":34,"_211":966},{"_112":963,"_405":520,"_406":34,"_211":964},"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.

",{"_745":16,"_648":59,"_555":59,"_375":59,"_365":59},"hasConsented","gpcApplied","showBanner","geolocation",{"_1069":1070,"_1071":1072},"country","US","state","OR","query","categories",[1092,1093,1094,1095,1096],"optionUrls",[1080,1081,1082,1083,1084,1085,1086,1087,1088,1089,1090,1091],"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/-",{"_20":1170,"_1098":1171,"_1100":59},{"_20":1135,"_1098":1136,"_1100":57},{"_20":1118,"_1098":1119,"_1100":1120},{"_20":21,"_1098":1111,"_1100":1112},{"_20":1097,"_1098":1099,"_1100":19},"helm-charts","items",[1101,1102],"startIndex",{"_20":1097,"_243":1108,"_205":206,"_207":-7,"_238":1109,"_246":1105,"_271":1110,"_1107":1090},{"_20":1097,"_243":1103,"_205":206,"_207":-7,"_238":1104,"_246":1105,"_271":1106,"_1107":1091},"NVIDIA NIM Operator","k8s-nim-operator","NVIDIA",[],"url","GPU Operator","gpu-operator",[],[1113],9,{"_20":21,"_243":1114,"_205":206,"_207":1115,"_238":1116,"_246":1105,"_271":1117,"_1107":1089},"Riva Skills Quick Start","riva","riva_quickstart",[],"models",[1121,1122,1123],6,{"_20":1118,"_243":1131,"_205":206,"_207":1132,"_238":1133,"_246":1105,"_271":1134,"_1107":1086},{"_20":1118,"_243":1128,"_205":206,"_207":1125,"_238":1129,"_246":1105,"_271":1130,"_1107":1087},{"_20":1118,"_243":1124,"_205":206,"_207":1125,"_238":1126,"_246":1105,"_271":1127,"_1107":1088},"TrafficCamNet","tao","trafficcamnet",[],"PeopleNet","peoplenet",[],"StyleGAN3 pretrained models","research","stylegan3",[],"collections",[1137,1138,1139],{"_20":1135,"_243":1167,"_205":206,"_207":1141,"_238":1168,"_246":1105,"_271":1169,"_1107":1083},{"_20":1135,"_243":1164,"_205":206,"_207":-7,"_238":1165,"_246":1105,"_271":1166,"_1107":1084},{"_20":1135,"_243":1140,"_205":206,"_207":1141,"_238":1142,"_246":1105,"_271":1143,"_1107":1085},"Omniverse Kit App Streaming","omniverse","kit-appstreaming-collection",[1144,1145,1146],{"_112":1158,"_243":1159,"_240":1160,"_250":1161,"_1151":1162},{"_112":1141,"_243":1154,"_240":1155,"_250":1150,"_1151":1156},{"_112":1147,"_243":1148,"_240":1149,"_250":1150,"_1151":1152},"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",[1153],"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.",[1153,1157],"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",[1153,1157,1163],"NGC_CONSOLE","DeepStream SDK","deepstream_sdk",[],"Omniverse Kit (FB)","kit",[1144,1145,1146],"containers",[1172,1173,1174],{"_20":1170,"_243":1182,"_205":1179,"_207":1183,"_238":1184,"_246":1105,"_271":1185,"_1107":1080},{"_20":1170,"_243":1178,"_205":1179,"_207":206,"_238":1180,"_246":1105,"_271":1181,"_1107":1081},{"_20":1170,"_243":1175,"_205":206,"_207":-7,"_238":1176,"_246":1105,"_271":1177,"_1107":1082},"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",{"_1215":1216,"_1217":313,"_1218":34,"_1219":1220,"_1098":1221,"_1222":1223,"_1224":1216,"_1225":34,"_1226":34,"_1227":34,"_1228":1229},"appName","ngc-catalog","appOrigin","hideOnPrivateOrg","icon","ObjectsModules2",[1230,1231,1232,1233,1234,1235,1236],"menuIcon","editor-swatches","productName","requiresAllScopes","requiresLogin","requiresOrgLevelScopesOnly","title","Catalog",{"_1228":1258,"_50":1259,"_1222":1260},{"_1251":1252,"_1228":1253,"_50":1254,"_1224":1158},{"_1228":1249,"_50":1250,"_1222":1223},{"_1228":1246,"_50":1247,"_1222":1248},{"_1228":1243,"_50":1244,"_1222":1245},{"_1228":1240,"_50":1241,"_1222":1242},{"_1228":1237,"_50":1238,"_1222":1239},"Resources","/resources","misc-library","Models","/models","misc-neural-network","Helm Charts","/helm-charts","misc-ship-wheel","Containers","/containers","3d-cube","Collections","/collections","access",[1255],"Enterprise Catalog","/enterprise",[1256,1257],"NVAIE","READ","Explore Catalog","/","common-home",{"_155":156,"_157":158,"_159":160,"_161":162,"_163":164,"_165":166},{"_154":34}]