An extendable neural machine translation toolkit built around the Transformer model and implemented in TensorFlow. Supports multi-GPU training and gradient aggregation for large-scale experimentation. Transformer implementation now part of [Nematus](https://github.com/EdinburghNLP/nematus).