LeakDistill

This is the repo for Incorporating Graph Information in Transformer-based AMR Parsing. The paper introduces a novel way to incorporate structural information at training time using Structural Adapters. This repo is an extension of SPRING repo. If you use our code, please reference this work in your paper:

@inproceedings{vasylenko-etal-2023-leakdistill,
    title = {Incorporating Graph Information in Transformer-based AMR Parsing},
    author = {Vasylenko, Pavlo and Huguet Cabot, Pere-Lluís and Martínez Lorenzo, Abelardo Carlos and Navigli, Roberto},
    booktitle = {Findings of ACL},
    year = {2023}
}

Installation

cd LeakDistill
pip install -r requirements.txt
pip install -e .

Training

LeakDistill

python bin/train.py --config configs/config_leak_distill.yaml

Graph Leakage Model

python bin/train.py --config configs/config_leak.yaml

Vanilla Knowledge Distillation

python bin/train_kd.py --config configs/config_kd.yaml --teacher <path_to_checkpoint>

SPRING

python bin/train.py --config configs/config_spring.yaml

Pretrained Checkpoints

For any questions or inquiries, please contact Pavlo Vasylenko at [email protected] or Pere-Lluís Huguet Cabot at [email protected]

Evaluation

python bin/predict_amrs.py \
    --config configs/config_leak_distill.yaml \
    --datasets '<path_to_datasets>' \
    --gold-path data/tmp/amr2.0/gold.amr.txt \
    --pred-path data/tmp/amr2.0/pred.amr.txt \
    --beamsize 10 \
    --checkpoint <path_to_checkpoint> \
    --device cuda

if datasets is not specified the path is taken from the config.test.

gold.amr.txt and pred.amr.txt will contain, respectively, the concatenated gold and the predictions.

To reproduce our paper's results, you will also need need to run BLINK entity linking system on the prediction file (data/tmp/amr2.0/pred.amr.txt in the previous code snippet). To do so, you will need to install BLINK, and download their models:

git clone https://github.com/facebookresearch/BLINK.git
cd BLINK
pip install -r requirements.txt
sh download_blink_models.sh
cd models
wget http://dl.fbaipublicfiles.com/BLINK//faiss_flat_index.pkl
cd ../..

Then, you will be able to launch the blinkify.py script:

python bin/blinkify.py \
    --datasets data/tmp/amr2.0/pred.amr.txt \
    --out data/tmp/amr2.0/pred.amr.blinkified.txt \
    --device cuda \
    --blink-models-dir BLINK/models

To have comparable Smatch scores you will also need to use the scripts available at https://github.com/mdtux89/amr-evaluation, which provide results that are around ~0.3 Smatch points lower than those returned by bin/predict_amrs.py.

License

This project is released under the CC-BY-NC-SA 4.0 license (see LICENSE). If you use LeakDistill, please reference the paper and put a link to this repo.