View Code? Open in Web Editor NEW

Official code and data release for Efficiently Aligned Cross-Lingual Transfer Learning for Conversational Tasks using Prompt-Tuning, accepted by findings of EACL 2024.

License: Apache License 2.0

Python 99.34% Shell 0.66%

fewxc's Introduction

FewXC (Few-shot Cross-lingual Conversational Research)

Official code and data release for Efficiently Aligned Cross-Lingual Transfer Learning for Conversational Tasks using Prompt-Tuning, accepted by findings of EACL 2024.

XSGD Data

https://console.cloud.google.com/storage/browser/multilingual-sgd-data-research

MASSIVE Data

Please refer to https://github.com/alexa/massive for data and pre-processing pipeline.

Backbone Model

XLM-RoBERTa series, https://huggingface.co/docs/transformers/model_doc/xlm-roberta

Examples

# NLI style intent classificaiton with MASSIVE dataset
CUDA_VISIBLE_DEVICES=0  python xlmr_massive_finetuning_nli.py \
--training_strategy 5 \
--learning_rate 1e-5 \
--training_data /export/home/xclm/1.1/parsed_data.train \
--intent_schema /export/home/xclm/1.1/parsed_data.intents \
--slot_schema /export/home/xclm/1.1/parsed_data.slots \
--dev_data /export/home/xclm/1.1/parsed_data.dev \
--prefix # turn on prefix tuning

# Regular intent classfication with MASSIVE dataset
CUDA_VISIBLE_DEVICES=0  python xlmr_massive_finetuning_vanilla.py \
--training_strategy 5 \
--learning_rate 1e-5 \
--training_data /export/home/xclm/1.1/parsed_data.train \
--intent_schema /export/home/xclm/1.1/parsed_data.intents \
--slot_schema /export/home/xclm/1.1/parsed_data.slots \
--dev_data /export/home/xclm/1.1/parsed_data.dev \
--prefix \
--task intent,slot

Citation

@inproceedings{
anonymous2024efficiently,
title={Efficiently Aligned Cross-Lingual Transfer Learning for Conversational Tasks using Prompt-Tuning},
author={Lifu Tu, Jin Qu, Semih Yavuz, Shafiq Joty, Wenhao Liu, Caiming Xiong, Yingbo Zhou},
booktitle={18th Conference of the European Chapter of the Association for Computational Linguistics},
year={2024},
url={https://openreview.net/forum?id=Lb4qW0NiIb}
}

Recommend Projects