Code for "Vocabulary Expansion for Low-resource Cross-lingual Transfer"
Vocabulary Expansion for Low-resource Cross-lingual Transfer
===
This is the official code for the paper titled "Vocabulary Expansion for Low-resource Cross-lingual Transfer." For reproduction, please refer to Reproduction.
## Requirements
* Python 3.11.7 or later
* CUDA 11.8
* torch==2.2.1
* transformers==4.39.0.dev0
* jupyterlab==4.1.2
* peft==0.8.2
* datasets==2.17.1
* evaluate==0.4.1
* bitsandbytes==0.42.0
* scikit-learn==1.4.1.post1
* seaborn==0.13.2
* sumeval==0.2.2
* janome==0.5.0
* protobuf==4.25.1
* entmax==1.3
* fastdist==1.1.6
* rouge-score==0.1.2
* numba==0.59.0
* tensorboardX==2.6.2.2
* tensorboard==2.16.2
* torch_tb_profiler==0.4.3
* pyarabic==0.6.15
* rouge==1.0.1
* huggingface-hub==0.21.1
* zstandard==0.22.0
* lm_eval==0.4.2
* lighteval==0.4.0
* openai==1.25.0
* tiktoken==0.6.0
* fasttext==0.9.2 (See below)
## Installation
After manually installing `PyTorch` and `transformers`, please run the following.
```bash
pip install -r requirements.txt
git clone
github.com
cd fastText
pip install .
```
## Reproduction
### 1. Preprocessing
See Preprocessing.
### 2. Target Model Initialization
See Instantiation
### 3. Language Adaptive Pre-training
See LAPT
### 4. Evaluation
See Evaluation
## Adapted Models
All models will be available on the Hugging Face Model Hub.
## License
MIT License
## Citation
If you find this work useful, please cite the following:
```
@article{yamaguchi2024vocabulary,
title={Vocabulary Expansion for Low-resource Cross-lingual Transfer},
author={Atsuki Yamaguchi and Aline Villavicencio and Nikolaos Aletras},
year={2024},
journal={arXiv},
volume={abs/2406.11477},
url={
arxiv.org
}
```