MRU-ASTP

Acknowledgments

Our code utilizes the framework from the LLaMA-Factory. We extend our gratitude to the contributors of this project. Additionally, we declare that the code we use adheres to the original project's copyright and licensing agreements.

Getting Started

Installation

git clone --depth 1 https://github.com/hiyouga/LLaMA-Factory.git
cd LLaMA-Factory
pip install -e ".[torch,metrics]"

Using llama3-8b-instruct

Download the model file to the same directory as the llama factory. https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct

About Dataset

We provide all data in our experiments in the directory "data/" about ASQP, ASTE, ACOS, TASD. You can just copy them into the "LLaMA-Factory/data"

Quickstart

To run MRU-ASTP with the LLaMA-Factory framework, follow these steps:

1. Update LLaMA-Factory/data/dataset_info.json: Add dataset information for MRU to the file. Ensure the format and details are consistent with existing entries . You can use our file(data_config/dataset_info.json)

2. Add YAML Files: Add the YAML files to this directory (LLaMA-Factory/examples/lora_multi_gpu/) . Ensure the paths and dataset names in the YAML files match the entries in LLaMA-Factory/data/dataset_info.json. Here we provide some yaml files about multi-task, single-task, low-resource setting in the directory (scripts/)

Fine-tuning

First step into LLaMA-Factory dictory.

llamafactory-cli train examples/train_lora/qwen7b_lora_sft.yaml

Prediction

llamafactory-cli train examples/train_lora/qwen7b_lora_predict.yaml

Evaluation

First set your prediction file(.jsonl) address in the evaluate/evaluate_group_plus.py Than use the evaluate/evaluate_group_plus.py script.

python evaluate/evaluate_group_plus.py