Tutorials

Migration

• Migrating to TPUEstimator API

  This tutorial describes how to convert a model program using the Estimator API to one using the TPUEstimator API.

Image Classification

• Running MNIST on Cloud TPU (TF 2.x)

  An MNIST image classification model using TensorFlow, optimized to run on Cloud TPU.

• Running MNIST on Cloud TPU (TF 1.x)

  An MNIST image classification model using TensorFlow, optimized to run on Cloud TPU.

• Training ResNet on Cloud TPU (TF 2.x)

  A ResNet image classification model using TensorFlow, optimized to run on Cloud TPU.

• Training ResNet on Cloud TPU (TF 1.x)

  A ResNet image classification model using TensorFlow, optimized to run on Cloud TPU.

• (PyTorch) Training ResNet on Cloud TPU

  A ResNet image classification model using PyTorch, optimized to run on Cloud TPU.

• Training MnasNet on Cloud TPU (TF 1.x)

  An MnasNet image classification model using TensorFlow, optimized to run on Cloud TPU.

• Training EfficientNet on Cloud TPU (TF 2.x)

  An EfficientNet image classification model using TensorFlow, optimized to run on Cloud TPU.

• Training EfficientNet on Cloud TPU (TF 1.x)

  An EfficientNet image classification model using TensorFlow, optimized to run on Cloud TPU.

• Training AmoebaNet-D on Cloud TPU (TF 1.x)

  An AmoebaNet image classification model using TensorFlow, optimized to run on Cloud TPU.

• Running Inception on Cloud TPU (TF 1.x)

  A guide to training the Inception models on Cloud TPU.

• Running Deeplab v3 on Cloud TPU (TF 1.x)

  A guide to training the Deeplab v3 model on Cloud TPU.

Object Detection

• Training RetinaNet on Cloud TPU (TF 1.x)

  A RetinaNet object detection model using TensorFlow, optimized to run on Cloud TPU.

• Training RetinaNet on Cloud TPU (TF 2.x)

  A RetinaNet object detection model using TensorFlow, optimized to run on Cloud TPU.

• Training ShapeMask on Cloud TPU (TF 1.x)

  A ShapeMask object detection model using TensorFlow, optimized to run on Cloud TPU.

• Training Mask RCNN on Cloud TPU (TF 1.x)

  A Mask RCNN model using TensorFlow, optimized to run on Cloud TPU.

Recommendation Systems

• (PyTorch) Training DLRM model on Cloud TPU

  A guide to training the PyTorch DLRM model on Cloud TPU.

• Training a Neural Collaboration Filtering model on Cloud TPU (TF 2.x)

  An implementation of the Neural Collaborative Filtering (NCF) framework with the Neural Matrix Factorization (NeuMF) model

Distributed Processing on a Pod

• (PyTorch) Training models on Cloud TPU Pods

  A ResNet-50 image classification model using PyTorch, optimized to run on a Cloud TPU Pod.

Natural Language Processing

• BERT FineTuning with Cloud TPU: Sentence and Sentence-Pair Classification Tasks (TF 2.x)

  Discover how to use Bidirectional Encoder Representations from Transformers (BERT) with Cloud TPU.

• BERT FineTuning with Cloud TPU: Sentence and Sentence-Pair Classification Tasks (TF 1.x)

  Discover how to use Bidirectional Encoder Representations from Transformers (BERT) with Cloud TPU.

• (PyTorch) Pre-training FairSeq RoBERTa on Cloud TPU

  A guide to pre-training the FairSeq version of the RoBERTa model on Cloud TPU using the public wikitext-103 dataset.

T2T Transformer

• Training Transformer on Cloud TPU (TF 2.x)

  A guide to training the Tensor2Tensor Transformer model on Cloud TPU, for English-German translation.

• Running the Transformer with Tensor2Tensor on Cloud TPU (TF 1.x)

  A guide to training the Tensor2Tensor Transformer model on Cloud TPU, for translation, language modeling, and sentiment analysis.

• (PyTorch) Training FairSeq Transformer on Cloud TPU

  A guide to training the FairSeq version of the Transformer model on Cloud TPU and running the WMT 18 translation task translating English to German.

• Running Automatic Speech Recognition on Cloud TPU (TF 1.x)

  A voice-text model using TensorFlow, optimized to run on Cloud TPU.

Google Cloud Service Integrations

• Training ResNet with Cloud TPU and GKE

  Using GKE to manage your Cloud TPU resources when training a ResNet model.

More Models on GitHub

• More Models

  A summary of additional models available on GitHub that are optimized for running on Cloud TPU.