nvidia/deeplearningexamples

★ 14,806Jupyter NotebookAudience · researcherComplexity · 5/5StaleSetup · hard

Mindmap

mindmap
  root((DeepLearningExamples))
    Inputs
      Training datasets
      NGC containers
      GPU hardware
      Config files
    Outputs
      Trained models
      ONNX exports
      TensorRT engines
      Triton deployments
    Use Cases
      Train ResNet variants
      Reproduce benchmarks
      Deploy with Triton
      Fine tune NLP
    Tech Stack
      PyTorch
      TensorFlow
      CUDA
      TensorRT
      Triton

mindmap root((DeepLearningExamples)) Inputs Training datasets NGC containers GPU hardware Config files Outputs Trained models ONNX exports TensorRT engines Triton deployments Use Cases Train ResNet variants Reproduce benchmarks Deploy with Triton Fine tune NLP Tech Stack PyTorch TensorFlow CUDA TensorRT Triton

Click or tap to explore — scroll the page freely

Things people build with this

USE CASE 1

Train an EfficientNet or Mask R-CNN model on your own dataset with NVIDIA-tuned hyperparameters

USE CASE 2

Reproduce published accuracy benchmarks on a Tensor Core GPU before customizing the model

USE CASE 3

Export a trained model to ONNX or TensorRT for fast inference deployment

USE CASE 4

Serve a model with Triton using the example deployment scripts as a template

Tech stack

PyTorchTensorFlowCUDATensorRTTriton

Getting it running

Difficulty · hard Time to first run · 1day+

You need NVIDIA GPUs with recent CUDA drivers, Docker with nvidia-container-runtime, and the matching monthly NGC container before any example will run.

In plain English

This repository is a collection of deep learning example projects published by NVIDIA. The README frames it as a set of ready-made training and deployment scripts for well-known model architectures, tuned so that they reach the published accuracy numbers and good performance on NVIDIA's Volta, Turing, and Ampere generation GPUs. The phrase Tensor Cores in the title refers to special units inside those GPUs that accelerate the mixed-precision math used in modern training. The code is meant to be run inside Docker containers that NVIDIA publishes monthly on its NGC container registry. Those containers include the examples from this repository, NVIDIA's contributions to the upstream framework projects, and supporting libraries such as cuDNN, NCCL, and cuBLAS. The README points out that everything in the monthly image goes through a quality assurance pass first, and links to per-release notes for the different framework images. Most of the README is a large table that lists every model in the collection. The Computer Vision section alone covers image classification networks such as several variants of EfficientNet and GPUNet, segmentation models including two Mask R-CNN ports and nnUNet, and other vision tasks. Each row notes which framework the example is written in (PyTorch, TensorFlow2, MXNet, or PaddlePaddle), and whether the example supports automatic mixed precision, multi-GPU training, multi-node training, export to TensorRT for fast inference, export to ONNX, deployment with the Triton inference server, an entry on NVIDIA's Deep Learning Containers index, and a companion Jupyter notebook. Beyond computer vision, the wider repository follows the same table format for natural language processing models, recommendation models, forecasting models, drug discovery work, and large language model training. The topics tags on the GitHub page list those areas, which is why one repository covers such a broad range. Each row in the tables links to the subfolder where the actual training scripts, configuration files, and documentation for that single model live. For a non-technical reader, the practical takeaway is that the project is a reference library aimed at engineers who have NVIDIA GPU hardware and want a known-good starting point for training or running a particular model, rather than a single product you install and use directly. The full README is longer than what was shown.

Copy-paste prompts

Prompt 1

Run the NVIDIA EfficientNet PyTorch example inside the matching NGC container on an A100

Prompt 2

Adapt the BERT pretraining script in DeepLearningExamples to my custom corpus

Prompt 3

Export a trained Mask R-CNN model from this repo to TensorRT and benchmark it

Prompt 4

Set up multi-GPU training with the nnUNet example and the recommended NGC image

Prompt 5

Compare two image classification examples in this repo by accuracy and throughput

Open on GitHub → Explain another repo

← nvidia on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.