Supported models

Cloud TPU provides the following set of supported models that are optimized for fast and accurate training.

.
TensorFlow version Support start Support end Model category Supported models
TF 2.1 January 8, 2020 (End date not yet set) Image classification ResNet-2.x, MNIST-2.x
Language modeling Transformer-2.x, BERT-2.x
Object detection RetinaNet-2.x
TF 1.15 October 22, 2019 (End date not yet set) Image classification ResNet, AmoebaNet,
EfficientNet,
MNasNet, MNIST
Language modeling Transformer, BERT
Object detection RetinaNet,
Image segmentation Mask R-CNN, ShapeMask
TF 1.14 July 19, 2019 (End date not yet set) Image classification ResNet, AmoebaNet,
EfficientNet,
MNasNet, MNIST
Language modeling
Object detection RetinaNet, Mask R-CNN
TF 1.13 March 11, 2019 (End date not yet set) Image classification ResNet, AmoebaNet,
MNasNet, MNIST
Language modeling
Object detection RetinaNet
PyTorch version Support start Support end Model category Supported models
PyTorch 1.x Oct 17, 2019 (End date not yet set) Image classification ResNet
Language modeling FairSeq Transformer, FairSeq RoBERTa