Lifelong Language Knowledge Distillation 🙋‍♂️ → 🏫 → 👨‍🎓

Code for the paper "Lifelong Language Knowledge Distillation"
In The 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP 2020)
by Yung-Sung Chuang, Shang-Yu Su, Yun-Nung Chen

Our code is based on the released code from LAnguage-MOdeling-for-Lifelong-Language-Learning. Most of the settings are identical to theirs.

📚 Dataset

Task	Dataset (Original Data Link)
Summarization	CNN/DM
Goal-Oriented Dialogue	WOZ
Semantic Parsing	WikiSQL
Natural Language Generation	E2ENLG
Natural Language Generation	RNNLG
Text Classification	AGNews, Yelp, Amazon, DBPedia, Yahoo

We use the released data from LAMOL's authors here, except for E2ENLG and RNNLG datasets.

We also release our processed data in here.

💻 Dependencies (same as LAMOL)

Ubuntu >= 16.04
This code only supports the following GPUs:
- NVIDIA Geforce RTX 2080TI
- NVIDIA TESLA V100
python3
cuda 10.1
python packages are listed in requirements.txt

🔧 Setup (same as LAMOL)

Create the following two directories in wherever you want. (you can name the directories arbitrarily):
- data directory: Where the dataset will be load by the model.
- model directory: The place for the model to dump its outputs.
Download the dataset: Download here and decompress it. After decompression, move all the files in the decompressed directory into data directory.
Make a copy of env.example and save it as env. In env, set the value of DATA_DIR as data directory and set the value of MODEL_ROOT_DIR as model directory.

👨‍🏫 Training and Testing (same as LAMOL)

train.sh and test.sh are the entrance for training and testing. Main options for them include:

Options	Description
seq_train_type	The mode to deal with a sequence of tasks. Mode include: lll\|finetune\|multitask\|mas\|ewc\|gem. "lll" is the default value corresponding our proposed method. The others are the methods for comparing with our proposal.
tasks	A sequence of tasks we want to train by seq_train_type. Leave a space between tasks after the `--tasks` tag. Tasks are the keys in TASK_DICT variable in `settings.py`
model_name	The language model we want to use. The default is `gpt2`. Options include gpt2\|openai-gpt,
gen_lm_sample_percentage	This tag only works with `--seq_train_type lll`. The percentage of the size of the dataset will be generated as pseudo samples for our proposed method.
lm_lambda	Lambda value for the loss function.
max_n_epochs	Maximum epoch value for all tasks.
min_batch_size	Minimum batch size for all tasks.
min_n_steps	Minimum step for optimizing the model for all tasks.
n_train_epochs	Epochs for training for all tasks.
n_gpu	Number of gpu to be used.
reg_lambda	Lambda value for mas and ewc.
top_k_lm	Top k sampling for the language model.
top_k_qa	Top k sampling for the qa model.
train_batch_size	Batch size for all tasks. The default is 0. Once the value equals to 0, The batch size will be decided dynamically based on the memory usage of the gpu.

🚨 New Arguments

Options	Description
distil	Use `--distil` to conduct Word-KD (the teacher model under `models/gpt2/lll/[TASK]_0.2/` is needed if `[TASK]` is in your LLL tasks.)
seq_distil	Use `--seq_distil` to conduct Seq-KD (distilled data need to be put in `data/[TASK]_to_squad-distil-v2.0.json`, which can be found in Supplementary Materials.)

Examples

See examples in run_seqsoftkd-WCS.sh/run_seqsoftkd-NLG.sh/run_seqsoftkd-TC.sh, which conduct Seq-KD(soft) on all the experiments in our paper.

In the examples, both --seq_distil and --distil are add to the arguments.

If you want to conduct Word-KD, skip --seq_distil in the arguments.

If you want to conduct Seq-KD, skip --distil in the arguments.

Outputs:

We add $SEED suffix to the model dir
If assigning multitask to --seq_train_type tag, the model will be dumped in $MODEL_ROOT_DIR / model_name / seq_train_type /TASK1_TASK2_... directory. Otherwise, it will be in $MODEL_ROOT_DIR / model_name / seq_train_type / TASK1_TASK2_... / TASK1, $MODEL_ROOT_DIR / model_name / seq_train_type / TASK1_TASK2_... / TASK2, ... directories.

📝 Acknowledgements:

We adapted the open source code of LAMOL provided by Cheng-Hao Ho and Fan-Keng Sun.
We use the language model offered by transformers, a state-of-the-art natural language processing models library by Thomas Wolf et al.
The implementation of MAS follows MAS-Memory-Aware-Synapses, the Memory Aware Synapses method implementation code by Aljundi R. et al.
The implementation of GEM follows GradientEpisodicMemory, the Gradient Episodic Memory method implementation code by Lopez-Paz, David et al.
The implementation of fp16 (fp16.py, fp16util.py) is from Megatron-LM, the ongoing research training transformer language models at scale by NVIDIA.
Data format conversion refer to decaNLP, the Natural Language Decathlon: Multitask Learning as Question Answering implementation code by Bryan McCann et al.

📕 Citation

@article{chuang2020lifelong,
  title={Lifelong Language Knowledge Distillation},
  author={Chuang, Yung-Sung and Su, Shang-Yu and Chen, Yun-Nung},
  journal={arXiv preprint arXiv:2010.02123},
  year={2020}
}

@inproceedings{sun2019lamol,
  title={LAMOL: LAnguage MOdeling for Lifelong Language Learning},
  author={Sun, Fan-Keng and Ho, Cheng-Hao and Lee, Hung-Yi},
  booktitle={International Conference on Learning Representations},
  year={2019}
}

wanyao1992 / L2KD