Library of deep learning models and datasets designed to make deep learning more accessible and accelerate ML research.
Python Jupyter Notebook JavaScript Other
Fetching latest commit…
Cannot retrieve the latest commit at this time.


PyPI version GitHub Issues Contributions welcome Gitter License Travis Run on FH

Tensor2Tensor, or T2T for short, is a library of deep learning models and datasets designed to make deep learning more accessible and accelerate ML research. T2T is actively used and maintained by researchers and engineers within the Google Brain team and a community of users. We're eager to collaborate with you too, so feel free to open an issue on GitHub or send along a pull request (see our contribution doc). You can chat with us on Gitter and join the T2T Google Group.

Quick Start

This iPython notebook explains T2T and runs in your browser using a free VM from Google, no installation needed. Alternatively, here is a one-command version that installs T2T, downloads MNIST, trains a model and evaluates it:

pip install tensor2tensor && t2t-trainer \
  --generate_data \
  --data_dir=~/t2t_data \
  --output_dir=~/t2t_train/mnist \
  --problem=image_mnist \
  --model=shake_shake \
  --hparams_set=shake_shake_quick \
  --train_steps=1000 \


Suggested Datasets and Models

Below we list a number of tasks that can be solved with T2T when you train the appropriate model on the appropriate problem. We give the problem and model below and we suggest a setting of hyperparameters that we know works well in our setup. We usually run either on Cloud TPUs or on 8-GPU machines; you might need to modify the hyperparameters if you run on a different setup.

Image Classification

For image classification, we have a number of standard data-sets:

  • ImageNet (a large data-set): --problem=image_imagenet, or one of the re-scaled versions (image_imagenet224, image_imagenet64, image_imagenet32)
  • CIFAR-10: --problem=image_cifar10 (or --problem=image_cifar10_plain to turn off data augmentation)
  • CIFAR-100: --problem=image_cifar100
  • MNIST: --problem=image_mnist

For ImageNet, we suggest to use the ResNet or Xception, i.e., use --model=resnet --hparams_set=resnet_50 or --model=xception --hparams_set=xception_base. Resnet should get to above 76% top-1 accuracy on ImageNet.

For CIFAR and MNIST, we suggest to try the shake-shake model: --model=shake_shake --hparams_set=shakeshake_big. This setting trained for --train_steps=700000 should yield close to 97% accuracy on CIFAR-10.

Image Generation

For (un)conditional image generation, we have a number of standard data-sets:

  • CelebA: --problem=img2img_celeba for image-to-image translation, namely, superresolution from 8x8 to 32x32.
  • CelebA-HQ: --problem=image_celeba256_rev for a downsampled 256x256.
  • CIFAR-10: --problem=image_cifar10_plain_gen_rev for class-conditional 32x32 generation.
  • LSUN Bedrooms: --problem=image_lsun_bedrooms_rev
  • MS-COCO: --problem=image_text_ms_coco_rev for text-to-image generation.
  • Small ImageNet (a large data-set): --problem=image_imagenet32_gen_rev for 32x32 or --problem=image_imagenet64_gen_rev for 64x64.

We suggest to use the Image Transformer, i.e., --model=imagetransformer, or the Image Transformer Plus, i.e., --model=imagetransformerpp that uses discretized mixture of logistics, or variational auto-encoder, i.e., --model=transformer_ae. For CIFAR-10, using --hparams_set=imagetransformer_cifar10_base or --hparams_set=imagetransformer_cifar10_base_dmol yields 2.90 bits per dimension. For Imagenet-32, using --hparams_set=imagetransformer_imagenet32_base yields 3.77 bits per dimension.

Language Modeling

For language modeling, we have these data-sets in T2T:

  • PTB (a small data-set): --problem=languagemodel_ptb10k for word-level modeling and --problem=languagemodel_ptb_characters for character-level modeling.
  • LM1B (a billion-word corpus): --problem=languagemodel_lm1b32k for subword-level modeling and --problem=languagemodel_lm1b_characters for character-level modeling.

We suggest to start with --model=transformer on this task and use --hparams_set=transformer_small for PTB and --hparams_set=transformer_base for LM1B.

Sentiment Analysis

For the task of recognizing the sentiment of a sentence, use

  • the IMDB data-set: --problem=sentiment_imdb

We suggest to use --model=transformer_encoder here and since it is a small data-set, try --hparams_set=transformer_tiny and train for few steps (e.g., --train_steps=2000).

Speech Recognition

For speech-to-text, we have these data-sets in T2T:

  • Librispeech (US English): --problem=librispeech for the whole set and --problem=librispeech_clean for a smaller but nicely filtered part.

  • Mozilla Common Voice (US English): --problem=common_voice for the whole set --problem=common_voice_clean for a quality-checked subset.


For summarizing longer text into shorter one we have these data-sets:

  • CNN/DailyMail articles summarized into a few sentences: --problem=summarize_cnn_dailymail32k

We suggest to use --model=transformer and --hparams_set=transformer_prepend for this task. This yields good ROUGE scores.


There are a number of translation data-sets in T2T:

  • English-German: --problem=translate_ende_wmt32k
  • English-French: --problem=translate_enfr_wmt32k
  • English-Czech: --problem=translate_encs_wmt32k
  • English-Chinese: --problem=translate_enzh_wmt32k
  • English-Vietnamese: --problem=translate_envi_iwslt32k

You can get translations in the other direction by appending _rev to the problem name, e.g., for German-English use --problem=translate_ende_wmt32k_rev (note that you still need to download the original data with t2t-datagen --problem=translate_ende_wmt32k).

For all translation problems, we suggest to try the Transformer model: --model=transformer. At first it is best to try the base setting, --hparams_set=transformer_base. When trained on 8 GPUs for 300K steps this should reach a BLEU score of about 28 on the English-German data-set, which is close to state-of-the art. If training on a single GPU, try the --hparams_set=transformer_base_single_gpu setting. For very good results or larger data-sets (e.g., for English-French), try the big model with --hparams_set=transformer_big.



Here's a walkthrough training a good English-to-German translation model using the Transformer model from Attention Is All You Need on WMT data.

pip install tensor2tensor

# See what problems, models, and hyperparameter sets are available.
# You can easily swap between them (and add new ones).
t2t-trainer --registry_help




# Generate data
t2t-datagen \
  --data_dir=$DATA_DIR \
  --tmp_dir=$TMP_DIR \

# Train
# *  If you run out of memory, add --hparams='batch_size=1024'.
t2t-trainer \
  --data_dir=$DATA_DIR \
  --problem=$PROBLEM \
  --model=$MODEL \
  --hparams_set=$HPARAMS \

# Decode

echo "Hello world" >> $DECODE_FILE
echo "Goodbye world" >> $DECODE_FILE
echo -e 'Hallo Welt\nAuf Wiedersehen Welt' >


t2t-decoder \
  --data_dir=$DATA_DIR \
  --problem=$PROBLEM \
  --model=$MODEL \
  --hparams_set=$HPARAMS \
  --output_dir=$TRAIN_DIR \
  --decode_hparams="beam_size=$BEAM_SIZE,alpha=$ALPHA" \
  --decode_from_file=$DECODE_FILE \

# See the translations
cat translation.en

# Evaluate the BLEU score
# Note: Report this BLEU score in papers, not the internal approx_bleu metric.
t2t-bleu --translation=translation.en


# Assumes tensorflow or tensorflow-gpu installed
pip install tensor2tensor

# Installs with tensorflow-gpu requirement
pip install tensor2tensor[tensorflow_gpu]

# Installs with tensorflow (cpu) requirement
pip install tensor2tensor[tensorflow]


# Data generator

# Trainer
t2t-trainer --registry_help

Library usage:

python -c "from tensor2tensor.models.transformer import Transformer"


  • Many state of the art and baseline models are built-in and new models can be added easily (open an issue or pull request!).
  • Many datasets across modalities - text, audio, image - available for generation and use, and new ones can be added easily (open an issue or pull request for public datasets!).
  • Models can be used with any dataset and input mode (or even multiple); all modality-specific processing (e.g. embedding lookups for text tokens) is done with Modality objects, which are specified per-feature in the dataset/task specification.
  • Support for multi-GPU machines and synchronous (1 master, many workers) and asynchronous (independent workers synchronizing through a parameter server) distributed training.
  • Easily swap amongst datasets and models by command-line flag with the data generation script t2t-datagen and the training script t2t-trainer.
  • Train on Google Cloud ML and Cloud TPUs.

T2T overview


Datasets are all standardized on TFRecord files with tensorflow.Example protocol buffers. All datasets are registered and generated with the data generator and many common sequence datasets are already available for generation and use.

Problems and Modalities

Problems define training-time hyperparameters for the dataset and task, mainly by setting input and output modalities (e.g. symbol, image, audio, label) and vocabularies, if applicable. All problems are defined either in or are registered with @registry.register_problem (run t2t-datagen to see the list of all available problems). Modalities, defined in, abstract away the input and output data types so that models may deal with modality-independent tensors.


T2TModels define the core tensor-to-tensor transformation, independent of input/output modality or task. Models take dense tensors in and produce dense tensors that may then be transformed in a final step by a modality depending on the task (e.g. fed through a final linear transform to produce logits for a softmax over classes). All models are imported in the models subpackage, inherit from T2TModel - defined in - and are registered with @registry.register_model.

Hyperparameter Sets

Hyperparameter sets are defined and registered in code with @registry.register_hparams and are encoded in objects. The HParams are available to both the problem specification and the model. A basic set of hyperparameters are defined in and hyperparameter set functions can compose other hyperparameter set functions.


The trainer binary is the main entrypoint for training, evaluation, and inference. Users can easily switch between problems, models, and hyperparameter sets by using the --model, --problem, and --hparams_set flags. Specific hyperparameters can be overridden with the --hparams flag. --schedule and related flags control local and distributed training/evaluation (distributed training documentation).

Adding your own components

T2T's components are registered using a central registration mechanism that enables easily adding new ones and easily swapping amongst them by command-line flag. You can add your own components without editing the T2T codebase by specifying the --t2t_usr_dir flag in t2t-trainer.

You can do so for models, hyperparameter sets, modalities, and problems. Please do submit a pull request if your component might be useful to others.

See the example_usr_dir for an example user directory.

Adding a dataset

To add a new dataset, subclass Problem and register it with @registry.register_problem. See TranslateEndeWmt8k for an example.

Also see the data generators README.

Run on FloydHub

Run on FloydHub

Click this button to open a Workspace on FloydHub. You can use the workspace to develop and test your code on a fully configured cloud GPU machine.

Tensor2Tensor comes preinstalled in the environment, you can simply open a Terminal and run your code.

# Test the quick-start on a Workspace's Terminal with this command
t2t-trainer \
  --generate_data \
  --data_dir=./t2t_data \
  --output_dir=./t2t_train/mnist \
  --problem=image_mnist \
  --model=shake_shake \
  --hparams_set=shake_shake_quick \
  --train_steps=1000 \

Note: Ensure compliance with the FloydHub Terms of Service.


When referencing Tensor2Tensor, please cite this paper.

  author    = {Ashish Vaswani and Samy Bengio and Eugene Brevdo and
    Francois Chollet and Aidan N. Gomez and Stephan Gouws and Llion Jones and
    \L{}ukasz Kaiser and Nal Kalchbrenner and Niki Parmar and Ryan Sepassi and
    Noam Shazeer and Jakob Uszkoreit},
  title     = {Tensor2Tensor for Neural Machine Translation},
  journal   = {CoRR},
  volume    = {abs/1803.07416},
  year      = {2018},
  url       = {},

Tensor2Tensor was used to develop a number of state-of-the-art models and deep learning methods. Here we list some papers that were based on T2T from the start and benefited from its features and architecture in ways described in the Google Research Blog post introducing T2T.

Note: This is not an official Google product.