MASS

MASS is a novel pre-training method for sequence to sequence based language generation tasks. It randomly masks a sentence fragment in the encoder, and then predicts it in the decoder.

The current codebase is for unsupersied neural machine translation. We will release our implementation for supervised machine translation, and other language generation tasks in the future.

Unsupervised NMT

Unsupervised Neural Machine Translation just uses monolingual data to train the models. For this task, we implement MASS on XLM.

We also provide pre-trained and fine-tuned models:

Languages	Pre-trained Model	Fine-tuned Model	BPE codes	Vocabulary
EN-FR	release soon	release soon	BPE codes	Vocabulary

We are also preparing larger models on more language pairs, and will release them in the future.

Data Ready

We use the same BPE codes and vocabulary with XLM. Here we take English-French as an example.

cd Unsupervised

wget https://dl.fbaipublicfiles.com/XLM/codes_enfr
wget https://dl.fbaipublicfiles.com/XLM/vocab_enfr

./get-data-nmt.sh --src en --tgt fr --reload_codes codes_enfr --reload_vocab vocab_enfr

Pre-training:

python train.py                                      \
--exp_name unsupMT_enfr                              \
--data_path ./data/processed/en-fr/                  \
--lgs 'en-fr'                                        \
--bt_steps 'en-fr-en,fr-en-fr'                       \
--mass_steps 'en,fr'                                 \
--lambda_bt '0:0,10:0'                               \
--encoder_only false                                 \
--emb_dim 1024                                       \
--n_layers 6                                         \
--n_heads 8                                          \
--dropout 0.1                                        \
--attention_dropout 0.1                              \
--gelu_activation true                               \
--tokens_per_batch 3000                              \
--optimizer adam_inverse_sqrt,beta1=0.9,beta2=0.98,lr=0.0001 \
--epoch_size 200000                                  \
--max_epoch 100                                      \
--eval_bleu true                                     \
--word_mass 0.5                                      \
--min_len 5                                          \

During the pre-training prcess, even without any back-translation, you can observe the model can achieve some intial BLEU scores:

epoch -> 4
valid_fr-en_mt_bleu -> 10.55
valid_en-fr_mt_bleu ->  7.81
test_fr-en_mt_bleu  -> 11.72
test_en-fr_mt_bleu  ->  8.80

Fine-tuning

After pre-training, we use back-translation to fine-tune the pre-trained model on unsupervised machine translation:

MODEL=mass_enfr_1024.pth

python train.py \
  --exp_name unsupMT_enfr                              \
  --data_path ./data/processed/en-fr/                  \
  --lgs 'en-fr'                                        \
  --bt_steps 'en-fr-en,fr-en-fr'                       \
  --encoder_only false                                 \
  --emb_dim 1024                                       \
  --n_layers 6                                         \
  --n_heads 8                                          \
  --dropout 0.1                                        \
  --attention_dropout 0.1                              \
  --gelu_activation true                               \
  --tokens_per_batch 2000                              \
  --batch_size 32	                                     \
  --bptt 256                                           \
  --optimizer adam_inverse_sqrt,beta1=0.9,beta2=0.98,lr=0.0001 \
  --epoch_size 200000                                  \
  --max_epoch 30                                       \
  --eval_bleu true                                     \
  --reload_model "$MODEL,$MODEL"                       \

Reference

If you find MASS useful in your work, you can cite the paper as below:

@inproceedings{song2019mass,
    title={MASS: Masked Sequence to Sequence Pre-training for Language Generation},
    author={Song, Kaitao and Tan, Xu and Qin, Tao and Lu, Jianfeng and Liu, Tie-Yan},
    booktitle={International Conference on Machine Learning},
    pages={5926--5936},
    year={2019}
}