316 lines (244 sloc) 11.5 KB


PyPI version GitHub Issues Contributions welcome Gitter License Travis

Tensor2Tensor, or T2T for short, is a library of deep learning models and datasets. It has binaries to train the models and to download and prepare the data for you. T2T is modular and extensible and can be used in notebooks for prototyping your own models or running existing ones on your data. It is actively used and maintained by researchers and engineers within the Google Brain team and was used to develop state-of-the-art models for translation (see Attention Is All You Need), summarization, image generation and other tasks. You can read more about T2T in the Google Research Blog post introducing it.

We're eager to collaborate with you on extending T2T, so please feel free to open an issue on GitHub or send along a pull request to add your dataset or model. See our contribution doc for details and our open issues. You can chat with us and other users on Gitter and please join our Google Group to keep up with T2T announcements.

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 data, trains an English-German translation model, and evaluates it:

pip install tensor2tensor && t2t-trainer \
  --generate_data \
  --data_dir=~/t2t_data \
  --problems=translate_ende_wmt32k \
  --model=transformer \
  --hparams_set=transformer_base_single_gpu \

You can decode from the model interactively:

t2t-decoder \
  --data_dir=~/t2t_data \
  --problems=translate_ende_wmt32k \
  --model=transformer \
  --hparams_set=transformer_base_single_gpu \
  --output_dir=~/t2t_train/base \

See the Walkthrough below for more details on each step and Suggested Models for well performing models on common tasks.



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=$DATA_DIR \
  --problems=$PROBLEM \
  --model=$MODEL \
  --hparams_set=$HPARAMS \

# Decode

echo "Hello world" >> $DECODE_FILE
echo "Goodbye world" >> $DECODE_FILE


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


Suggested Models

Here are some combinations of models, hparams and problems that we found work well, so we suggest to use them if you're interested in that problem.


For translation, esp. English-German and English-French, we suggest to use the Transformer model in base or big configurations, i.e. for --problems=translate_ende_wmt32k use --model=transformer and --hparams_set=transformer_base. When trained on 8 GPUs for 300K steps this should reach a BLEU score of about 28.


For summarization suggest to use the Transformer model in prepend mode, i.e. for --problems=summarize_cnn_dailymail32k use --model=transformer and --hparams_set=transformer_prepend.

Image Classification

For image classification suggest to use the ResNet or Xception, i.e. for --problems=image_imagenet use --model=resnet50 and --hparams_set=resnet_base or --model=xception and --hparams_set=xception_base.


# 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.

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, --problems, 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.

Note: This is not an official Google product.