Skip to content

Latest commit

 

History

History
200 lines (166 loc) · 14.4 KB

README.md

File metadata and controls

200 lines (166 loc) · 14.4 KB

Deep Exploration of Cross-Lingual Zero-Shot Generalization in Instruction Tuning

This is the official github repository for 'Deep Exploration of Cross-Lingual Zero-Shot Generalization in Instruction Tuning' [ACL 2024 Findings].

Citation:

@misc{han2024deep,
    title={Deep Exploration of Cross-Lingual Zero-Shot Generalization in Instruction Tuning},
    author={Janghoon Han and Changho Lee and Joongbo Shin and Stanley Jungkyu Choi and Honglak Lee and Kynghoon Bae},
    year={2024},
    eprint={2406.08796},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

KORANI datasets and task taxonomy. Green datasets are NLG datasets. Yellow datasets are NLU datasets.

korani_task_taxonomy

0. Install Dependencies

conda create -n korani python=3.10
conda activate korani
# install torch with the correct cuda version, check nvcc --version
pip install torch==2.1.1 torchvision==0.16.1 torchaudio==2.1.1 --index-url https://download.pytorch.org/whl/cu118
# install Hugging Face Libraries
pip install "transformers==4.37.0" "datasets==2.19.1" "accelerate==0.25.0" "evaluate==0.4.0" --upgrade
# install deepspeed and ninja for jit compilations of kernels
pip install "deepspeed==0.9.3" ninja --upgrade
# install additional dependencies needed for training
pip install rouge-score nltk py7zr tensorboard scikit-learn
pip install sentencepiece
pip install wandb
pip install gdown
pip install konlpy
pip install absl-py

If you want to get Korean rouge score for generation tasks, you have to download the Mecab.

cd KORANI-Instruction-Tuning
wget https://bitbucket.org/eunjeon/mecab-ko/downloads/mecab-0.996-ko-0.9.2.tar.gz
tar xvfz mecab-0.996-ko-0.9.2.tar.gz
cd mecab-0.996-ko-0.9.2
./configure
make
make check
sudo make install
wget https://bitbucket.org/eunjeon/mecab-ko-dic/downloads/mecab-ko-dic-2.1.1-20180720.tar.gz
tar xvfz mecab-ko-dic-2.1.1-20180720.tar.gz
cd mecab-ko-dic-2.1.1-20180720
./configure
./autogen.sh
make
sudo make install

If errors occurred in the command above, please follow the below.

cd ..
rm -r mecab-ko-dic-2.1.1-20180720

tar xvfz mecab-ko-dic-2.1.1-20180720.tar.gz
cd mecab-ko-dic-2.1.1-20180720
autoreconf
./configure
make
sudo make install

And finally,

pip install mecab-python

1. KORANI Datasets

Please download the KORANI datasets and unzip it to utilize it.

gdown https://drive.google.com/uc?id=1W8tXUZFK-J09kYV1-ZxDOE6QDOrYh_6x
jar xvf KORANI.zip

All of the train/eval data are saved in csv file format. For the datasets without license for releasing dataset as an open source, please visit the directory and follow the README.md in the directory. Please visit the following directory in KORANI folder.

coreference_resolution/nikl_coref
extractive_qa/korquad1
hatespeech/unsmile
paraphrasing/similar_corpus

Source and License of KORANI Datasets

  • Note that AIHub License allows distribution of derivative work, but not the original datasets.
Name Task License Link
NSMC Sentiment CC0 https://github.com/e9t/nsmc
Naver Shopping Sentiment Public Domain https://github.com/bab2min/corpus/tree/master/sentiment
AIHub Emo Sentiment AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=86
Sosang Sentiment Sentiment AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=102
Kobest Sentineg Sentiment CC-BY-SA https://huggingface.co/datasets/skt/kobest_v1
BEEP! HateSpeech CC-BY-SA https://github.com/kocohub/korean-hate-speech
Curse Detection HateSpeech Apache License 2.0 https://github.com/2runo/Curse-detection-v2
Unsmile HateSpeech CC-BY-NC-ND https://github.com/smilegate-ai/korean_unsmile_dataset
Apeach HateSpeech CC-BY-SA https://github.com/jason9693/APEACH
KLUE TC Topic Classification CC-BY-SA https://aistages-prod-server-public.s3.amazonaws.com/app/Competitions/000066/data/ynat-v1.1.tar.gz
Ko Conversation Topic Classification AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=116
Callcenter Topic Classification AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=98
Daily Chat Intent AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=&topMenu=&aihubDataSe=extrldata&dataSetSn=288
Sae4k Intent CC-BY-SA https://github.com/warnikchow/sae4k
StyleKQC Intent CC-BY-SA https://github.com/cynthia/stylekqc
ParaKQC Paraphrase Identification CC-BY-SA https://github.com/warnikchow/paraKQC
KLUE STS Paraphrase Identification CC-BY-SA https://aistages-prod-server-public.s3.amazonaws.com/app/Competitions/000067/data/klue-sts-v1.1.tar.gz
Question Pair Paraphrase Identification MIT https://github.com/songys/Question_pair
KorSTS Paraphrase Identification CC-BY-SA https://github.com/kakaobrain/KorNLUDatasets
KorSS Paraphrase Identification MIT https://github.com/yoongi0428/Kor-Sentence-Similarity
KLUE NLI Natural Language Inference CC-BY-SA https://aistages-prod-server-public.s3.amazonaws.com/app/Competitions/000068/data/klue-nli-v1.1.tar.gz
KorNLI Natural Language Inference CC-BY-SA https://github.com/kakaobrain/KorNLUDatasets
Similar Corpus Paraphrasing NOT FREE https://corpus.korean.go.kr/main.do
Kobest WiC Word Sense Disambiguation CC-BY-SA https://huggingface.co/datasets/skt/kobest_v1
Kobest Hellaswag Sentence Completion CC-BY-SA https://huggingface.co/datasets/skt/kobest_v1
Kobest COPA Sentence Completion CC-BY-SA https://huggingface.co/datasets/skt/kobest_v1
Com Gen Structure-to-Text AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=120&topMenu=100&aihubDataSe=extrldata&dataSetSn=459
Dacon News Summarization AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=97
Book Summarization AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=93
Document Editorial Summarization AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=97
Document News Summarization AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=97
Report Summarization AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=97
KorQuAD1 Extractive QA CC-BY-ND https://korquad.github.io/KorQuad%201.0/
NIA QA Extractive QA AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=106
KLUE MRC Extractive QA CC-BY-SA https://aistages-prod-server-public.s3.amazonaws.com/app/Competitions/000072/data/klue-mrc-v1.1.tar.gz
AIHub MRC Extractive QA AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=89
Book MRC Extractive QA AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=92
News QA Extractive QA AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=577
Kobest BoolQ Extractive QA CC-BY-SA https://huggingface.co/datasets/skt/kobest_v1
Document QA Multiple Choice QA AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=569
Twitter Dialogs AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&dataSetSn=269
AIHub Daily Dial Dialogs AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=543
AIHub Emo Dial Dialogs AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=86
AIHub Korean Dial Dialogs AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=116
AIHub Minwon Dialogs AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=98
AIHub TOD Dialogs AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=544
Ko-En Technology Translation AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=71266
Ko-En Social Translation AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=125
Ko-En Parallel Translation AIHub License https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=126
NIKL Coref Coreference Resolution NOT FREE https://corpus.korean.go.kr/main.do
ETRI QA Closed Book QA CC-BY-NC https://aiopen.etri.re.kr/corpusModel

2. Train any LMs in Huggingface

Before training, you have to gather instruction tuning datasets in accordance with your held-out setting. If you have decided which dataset to be included in the training dataset, please gather and save it as one csv file. Please refer to following directory as an example.

cd data/sample_ko

Before actual training, we will tokenize the train dataset before the actaul training. After filling out the configurations in 'run_configs' directory, run the following code for preprocessing the train data

bash preprocess_data.sh

Finally, run the run.sh file to train! ('localhost:{num1, num2}' to designate the CUDA_VISIBLE_DEVICES=num1, num2. Code will use all GPUs as default)

bash run.sh

3. Evaluate any LMs in Huggingface

Run the inference.sh file to evaluate! You can either choose task cluster(s) to evaluate or specific task(s) to evaluate.

bash inference.sh

Results on Korean LLMs

We report the ROUGE-L score for the generation task (Summarization and Coreference Resolution), and ACC for the others.

Model Sentiment Summarization Multiple Choice QA NLI Sent. Comp. WSD Coref. Resol. Average
PolyGlot 1.3b 47.723 3.277 34.133 33.56 48.183 2.17 48.81 31.122
PolyGlot 3.8b 47.783 3.138 34.133 33.56 48.167 0.435 48.81 30.861
PolyGlot 5.8b 49.076 3.995 37.333 33.75 52.817 2.497 48.837 32.615
PolyGlot 12.8b 45.963 4.026 37.7 33.817 54.734 2.497 49.79 32.647
Llama-3-Open-Ko-8B 44.303 2.256 49.267 34.355 52.983 1.28 48.81 33.322
T3Q-ko-solar-dpo-v7.0 76.753 10.138 67.4 61.925 74.475 6.185 78.295 53.596