OpenNMT-tf is a general purpose sequence modeling tool in TensorFlow with production in mind. While neural machine translation is the main target task, it has been designed to more generally support:
- sequence to sequence mapping
- sequence tagging
- sequence classification
OpenNMT-tf focuses on modularity to support advanced modeling and training capabilities:
- arbitrarily complex encoder architectures
e.g. mixing RNNs, CNNs, self-attention, etc. in parallel or in sequence. - hybrid encoder-decoder models
e.g. self-attention encoder and RNN decoder or vice versa. - multi-source training
e.g. source text and Moses translation as inputs for machine translation. - multiple input format
text with support of mixed word/character embeddings or real vectors serialized in TFRecord files. - on-the-fly tokenization
apply advanced tokenization dynamically during the training and detokenize the predictions during inference or evaluation. - automatic evaluation
support for saving evaluation predictions and running external evaluators (e.g. BLEU).
and all of the above can be used simultaneously to train novel and complex architectures. See the predefined models to discover how they are defined and the API documentation to customize them.
OpenNMT-tf is also compatible with some of the best TensorFlow features:
- replicated and distributed training
- monitoring with TensorBoard
- inference with TensorFlow Serving and the TensorFlow C++ API
tensorflow
(1.6
)pyyaml
A minimal OpenNMT-tf run consists of 3 elements:
- a run type:
train_and_eval
,train
,eval
,infer
, orexport
- a Python file describing the model
- a YAML file describing the parameters
that are passed to the main script:
python -m bin.main <run_type> --model <model_file.py> --config <config_file.yml>
- For more information about configuration files, see the documentation.
- For more information about command line options, see the help flag
python -m bin.main -h
.
Here is a minimal workflow to get you started in using OpenNMT-tf. This example uses a toy English-German dataset for machine translation.
1. Build the word vocabularies:
python -m bin.build_vocab --size 50000 --save_vocab data/toy-ende/src-vocab.txt data/toy-ende/src-train.txt
python -m bin.build_vocab --size 50000 --save_vocab data/toy-ende/tgt-vocab.txt data/toy-ende/tgt-train.txt
2. Train with preset parameters:
python -m bin.main train_and_eval --model config/models/nmt_small.py --config config/opennmt-defaults.yml config/data/toy-ende.yml
3. Translate a test file with the latest checkpoint:
python -m bin.main infer --config config/opennmt-defaults.yml config/data/toy-ende.yml --features_file data/toy-ende/src-test.txt
Note: do not expect any good translation results with this toy example. Consider training on larger parallel datasets instead.
For more advanced usages, see the documentation or the WMT training scripts.
OpenNMT-tf also exposes well-defined and documented functions and classes. For example, OpenNMT-tf has been used to implement unsupervised NMT in TensorFlow.
OpenNMT-tf has been designed from scratch and compatibility with the {Lua,Py}Torch implementations in terms of usage, design, and features is not a priority. Please submit a feature request for any missing feature or behavior that you found useful in the {Lua,Py}Torch implementations.
The implementation is inspired by the following: