Skip to content

LARC-CMU-SMU/Enconter

Repository files navigation

Enconter

This is the official implementation of 2021 EACL paper

ENCONTER: Entity Constrained Progressive Sequence Generation viaInsertion-based Transformer

File description

train_CoNLL.sh: training args
train_enconter.py: main training script
preprocess_CoNLL.ipynb: preprocessing CoNLL dataset for training
dataset_utils.py, utils.py: some utilities

How to use

  1. Clone the repo
  2. Run preprocess_CoNLL.ipynb to preprocess training and testing data
  3. Run train_CoNLL.sh to train POINTER-E, Greedy Enconter, BBT Enconter
  4. Run predict_CoNLL.sh to conduct generation

Dependicies

tqdm==4.49.0
transformers==3.0.0
torch==1.5.0
numpy==1.19.2
matplotlib==3.2.1
yake==0.4.3

Reference