This is an implementation of the Transformer translation model as described in the Attention is All You Need paper. Based on the code provided by the authors: Transformer code from Tensor2Tensor.
Transformer is a neural network architecture that solves sequence to sequence problems using attention mechanisms. Unlike traditional neural seq2seq models, Transformer does not involve recurrent connections. The attention mechanism learns dependencies between tokens in two sequences. Since attention weights apply to all tokens in the sequences, the Tranformer model is able to easily capture long-distance depedencies.
Transformer's overall structure follows the standard encoder-decoder pattern. The encoder uses self-attention to compute a representation of the input sequence. The decoder generates the output sequence one token at a time, taking the encoder output and previous decoder-outputted tokens as inputs.
The model also applies embeddings on the input and output tokens, and adds a constant positional encoding. The positional encoding adds information about the position of each token.
- Contents
- Walkthrough
- Benchmarks
- Detailed instructions
- Export trained model
- Implementation overview
- Term definitions
Below are the commands for running the Transformer model. See the Detailed instrutions for more details on running the model.
cd /path/to/models/official/transformer
# Ensure that PYTHONPATH is correctly defined as described in
# https://github.com/tensorflow/models/tree/master/official#requirements
# export PYTHONPATH="$PYTHONPATH:/path/to/models"
# Export variables
PARAM_SET=big
DATA_DIR=$HOME/transformer/data
MODEL_DIR=$HOME/transformer/model_$PARAM_SET
VOCAB_FILE=$DATA_DIR/vocab.ende.32768
# Download training/evaluation datasets
python data_download.py --data_dir=$DATA_DIR
# Train the model for 10 epochs, and evaluate after every epoch.
python transformer_main.py --data_dir=$DATA_DIR --model_dir=$MODEL_DIR \
--vocab_file=$VOCAB_FILE --param_set=$PARAM_SET \
--bleu_source=test_data/newstest2014.en --bleu_ref=test_data/newstest2014.de
# Run during training in a separate process to get continuous updates,
# or after training is complete.
tensorboard --logdir=$MODEL_DIR
# Translate some text using the trained model
python translate.py --model_dir=$MODEL_DIR --vocab_file=$VOCAB_FILE \
--param_set=$PARAM_SET --text="hello world"
# Compute model's BLEU score using the newstest2014 dataset.
python translate.py --model_dir=$MODEL_DIR --vocab_file=$VOCAB_FILE \
--param_set=$PARAM_SET --file=test_data/newstest2014.en --file_out=translation.en
python compute_bleu.py --translation=translation.en --reference=test_data/newstest2014.de
Currently, both big and base parameter sets run on a single GPU. The measurements below are reported from running the model on a P100 GPU.
| Param Set | batches/sec | batches per epoch | time per epoch |
|---|---|---|---|
| base | 4.8 | 83244 | 4 hr |
| big | 1.1 | 41365 | 10 hr |
Below are the case-insensitive BLEU scores after 10 epochs.
| Param Set | Score |
|---|---|
| base | 27.7 |
| big | 28.9 |
-
Follow the instructions described in the Requirements section to add the models folder to the python path.
Export the following variables, or modify the values in each of the snippets below:
PARAM_SET=big DATA_DIR=$HOME/transformer/data MODEL_DIR=$HOME/transformer/model_$PARAM_SET VOCAB_FILE=$DATA_DIR/vocab.ende.32768 -
data_download.py downloads and preprocesses the training and evaluation WMT datasets. After the data is downloaded and extracted, the training data is used to generate a vocabulary of subtokens. The evaluation and training strings are tokenized, and the resulting data is sharded, shuffled, and saved as TFRecords.
1.75GB of compressed data will be downloaded. In total, the raw files (compressed, extracted, and combined files) take up 8.4GB of disk space. The resulting TFRecord and vocabulary files are 722MB. The script takes around 40 minutes to run, with the bulk of the time spent downloading and ~15 minutes spent on preprocessing.
Command to run:
python data_download.py --data_dir=$DATA_DIRArguments:
--data_dir: Path where the preprocessed TFRecord data, and vocab file will be saved.- Use the
--helpor-hflag to get a full list of possible arguments.
-
transformer_main.py creates a Transformer model, and trains it using Tensorflow Estimator.
Command to run:
python transformer_main.py --data_dir=$DATA_DIR --model_dir=$MODEL_DIR \ --vocab_file=$VOCAB_FILE --param_set=$PARAM_SETArguments:
--data_dir: This should be set to the same directory given to thedata_download'sdata_dirargument.--model_dir: Directory to save Transformer model training checkpoints.--vocab_file: Path to subtoken vacbulary file. If data_download was used, you may find the file indata_dir.--param_set: Parameter set to use when creating and training the model. Options arebaseandbig(default).- Use the
--helpor-hflag to get a full list of possible arguments.
By default, the model will train for 10 epochs, and evaluate after every epoch. The training schedule may be defined through the flags:
- Training with epochs (default):
--train_epochs: The total number of complete passes to make through the dataset--epochs_between_evals: The number of epochs to train between evaluations.
- Training with steps:
--train_steps: sets the total number of training steps to run.--steps_between_evals: Number of training steps to run between evaluations.
Only one of
train_epochsortrain_stepsmay be set. Since the default option is to evaluate the model after training for an epoch, it may take 4 or more hours between model evaluations. To get more frequent evaluations, use the flags--train_steps=250000 --steps_between_evals=1000.Note: At the beginning of each training session, the training dataset is reloaded and shuffled. Stopping the training before completing an epoch may result in worse model quality, due to the chance that some examples may be seen more than others. Therefore, it is recommended to use epochs when the model quality is important.
Use these flags to compute the BLEU when the model evaluates:
--bleu_source: Path to file containing text to translate.--bleu_ref: Path to file containing the reference translation.--stop_threshold: Train until the BLEU score reaches this lower bound. This setting overrides the--train_stepsand--train_epochsflags.
The test source and reference files located in the
test_datadirectory are extracted from the preprocessed dataset from the NMT Seq2Seq tutorial.When running
transformer_main.py, use the flags:--bleu_source=test_data/newstest2014.en --bleu_ref=test_data/newstest2014.deTraining and evaluation metrics (loss, accuracy, approximate BLEU score, etc.) are logged, and can be displayed in the browser using Tensorboard.
tensorboard --logdir=$MODEL_DIRThe values are displayed at localhost:6006.
-
translate.py contains the script to use the trained model to translate input text or file. Each line in the file is translated separately.
Command to run:
python translate.py --model_dir=$MODEL_DIR --vocab_file=$VOCAB_FILE \ --param_set=$PARAM_SET --text="hello world"Arguments for initializing the Subtokenizer and trained model:
--model_dirand--param_set: These parameters are used to rebuild the trained model--vocab_file: Path to subtoken vacbulary file. If data_download was used, you may find the file indata_dir.
Arguments for specifying what to translate:
--text: Text to translate--file: Path to file containing text to translate--file_out: If--fileis set, then this file will store the input file's translations.
To translate the newstest2014 data, run:
python translate.py --model_dir=$MODEL_DIR --vocab_file=$VOCAB_FILE \ --param_set=$PARAM_SET --file=test_data/newstest2014.en --file_out=translation.enTranslating the file takes around 15 minutes on a GTX1080, or 5 minutes on a P100.
-
Use compute_bleu.py to compute the BLEU by comparing generated translations to the reference translation.
Command to run:
python compute_bleu.py --translation=translation.en --reference=test_data/newstest2014.deArguments:
--translation: Path to file containing generated translations.--reference: Path to file containing reference translations.- Use the
--helpor-hflag to get a full list of possible arguments.
-
TPU support for this version of Transformer is experimental. Currently it is present for demonstration purposes only, but will be optimized in the coming weeks.
To export the model as a Tensorflow SavedModel format, use the argument --export_dir when running transformer_main.py. A folder will be created in the directory with the name as the timestamp (e.g. $EXPORT_DIR/1526427396).
EXPORT_DIR=$HOME/transformer/saved_model
python transformer_main.py --data_dir=$DATA_DIR --model_dir=$MODEL_DIR \
--vocab_file=$VOCAB_FILE --param_set=$PARAM_SET --export_model=$EXPORT_DIR
To inspect the SavedModel, use saved_model_cli:
SAVED_MODEL_DIR=$EXPORT_DIR/{TIMESTAMP} # replace {TIMESTAMP} with the name of the folder created
saved_model_cli show --dir=$SAVED_MODEL_DIR --all
Let's translate "hello world!", "goodbye world.", and "Would you like some pie?".
The SignatureDef for "translate" is:
signature_def['translate']:
The given SavedModel SignatureDef contains the following input(s):
inputs['input'] tensor_info:
dtype: DT_INT64
shape: (-1, -1)
name: Placeholder:0
The given SavedModel SignatureDef contains the following output(s):
outputs['outputs'] tensor_info:
dtype: DT_INT32
shape: (-1, -1)
name: model/Transformer/strided_slice_19:0
outputs['scores'] tensor_info:
dtype: DT_FLOAT
shape: (-1)
name: model/Transformer/strided_slice_20:0
Follow the steps below to use the translate signature def:
-
This can be done using
utils.tokenizer.Subtokenizer, and the vocab file in the SavedModel assets ($SAVED_MODEL_DIR/assets.extra/vocab.txt).from official.transformer.utils.tokenizer import Subtokenizer s = Subtokenizer(PATH_TO_VOCAB_FILE) print(s.encode("hello world!", add_eos=True))The encoded inputs are:
"hello world!" = [6170, 3731, 178, 207, 1]"goodbye world." = [15431, 13966, 36, 178, 3, 1]"Would you like some pie?" = [9092, 72, 155, 202, 19851, 102, 1]
-
The encoded inputs should be padded so that they are the same length. The padding token is
0.ENCODED_INPUTS="[[26228, 145, 178, 1, 0, 0, 0], \ [15431, 13966, 36, 178, 3, 1, 0], \ [9092, 72, 155, 202, 19851, 102, 1]]"Now, use the
runcommand withsaved_model_clito get the outputs.saved_model_cli run --dir=$SAVED_MODEL_DIR --tag_set=serve --signature_def=translate \ --input_expr="input=$ENCODED_INPUTS"The outputs will look similar to:
Result for output key outputs: [[18744 145 297 1 0 0 0 0 0 0 0 0 0 0] [ 5450 4642 21 11 297 3 1 0 0 0 0 0 0 0] [25940 22 66 103 21713 31 102 1 0 0 0 0 0 0]] Result for output key scores: [-1.5493642 -1.4032784 -3.252089 ] -
Use the
Subtokenizerand vocab file as described in step 1 to decode the output integer arrays.from official.transformer.utils.tokenizer import Subtokenizer s = Subtokenizer(PATH_TO_VOCAB_FILE) print(s.decode([18744, 145, 297, 1]))The decoded outputs from above are:
[18744, 145, 297, 1] = "Hallo Welt<EOS>"[5450, 4642, 21, 11, 297, 3, 1] = "Abschied von der Welt.<EOS>"[25940, 22, 66, 103, 21713, 31, 102, 1] = "Möchten Sie einen Kuchen?<EOS>"
A brief look at each component in the code:
The model subdirectory contains the implementation of the Transformer model. The following files define the Transformer model and its layers:
- transformer.py: Defines the transformer model and its encoder/decoder layer stacks.
- embedding_layer.py: Contains the layer that calculates the embeddings. The embedding weights are also used to calculate the pre-softmax probabilities from the decoder output.
- attention_layer.py: Defines the multi-headed and self attention layers that are used in the encoder/decoder stacks.
- ffn_layer.py: Defines the feedforward network that is used in the encoder/decoder stacks. The network is composed of 2 fully connected layers.
Other files:
- beam_search.py contains the beam search implementation, which is used during model inference to find high scoring translations.
- model_params.py contains the parameters used for the big and base models.
- model_utils.py defines some helper functions used in the model (calculating padding, bias, etc.).
transformer_main.py creates an Estimator to train and evaluate the model.
Helper functions:
- utils/dataset.py: contains functions for creating a
datasetthat is passed to theEstimator. - utils/metrics.py: defines metrics functions used by the
Estimatorto evaluate the
Aside from the main file to train the Transformer model, we provide other scripts for using the model or downloading the data:
data_download.py downloads and extracts data, then uses Subtokenizer to tokenize strings into arrays of int IDs. The int arrays are converted to tf.Examples and saved in the tf.RecordDataset format.
The data is downloaded from the Workshop of Machine Transtion (WMT) news translation task. The following datasets are used:
- Europarl v7
- Common Crawl corpus
- News Commentary v12
See the download section to explore the raw datasets. The parameters in this model are tuned to fit the English-German translation data, so the EN-DE texts are extracted from the downloaded compressed files.
The text is transformed into arrays of integer IDs using the Subtokenizer defined in utils/tokenizer.py. During initialization of the Subtokenizer, the raw training data is used to generate a vocabulary list containing common subtokens.
The target vocabulary size of the WMT dataset is 32,768. The set of subtokens is found through binary search on the minimum number of times a subtoken appears in the data. The actual vocabulary size is 33,708, and is stored in a 324kB file.
Translation is defined in translate.py. First, Subtokenizer tokenizes the input. The vocabulary file is the same used to tokenize the training/eval files. Next, beam search is used to find the combination of tokens that maximizes the probability outputted by the model decoder. The tokens are then converted back to strings with Subtokenizer.
compute_bleu.py: Implementation from https://github.com/tensorflow/tensor2tensor/blob/master/tensor2tensor/utils/bleu_hook.py.
The newstest2014 files are extracted from the NMT Seq2Seq tutorial. The raw text files are converted from the SGM format of the WMT 2016 test sets.
Steps / Epochs:
- Step: unit for processing a single batch of data
- Epoch: a complete run through the dataset
Example: Consider a training a dataset with 100 examples that is divided into 20 batches with 5 examples per batch. A single training step trains the model on one batch. After 20 training steps, the model will have trained on every batch in the dataset, or one epoch.
Subtoken: Words are referred to as tokens, and parts of words are referred to as 'subtokens'. For example, the word 'inclined' may be split into ['incline', 'd_']. The '_' indicates the end of the token. The subtoken vocabulary list is guaranteed to contain the alphabet (including numbers and special characters), so all words can be tokenized.