This repository contains the official code for the paper Is Fine-tuning Needed? Pre-trained Language Models Are Near Perfect for Out-of-Domain Detection (ACL 2023).

First, create a virtual environment for the project (we use Conda to create a python 3.9 environment) and install all the requirments using requirements.txt.

1. conda create -n ood_det python==3.9
2. conda activate ood_det
3. pip install -r requirements.txt

Now, create two directories:

1. Data directory
2. Model directory - create two subdirectories: pretrained_models and finetuned_models

Both directories can be located anywhere but should be specified in config.py (See more below).

At the start of each session, run the following: . bin/start.sh

Directory paths and training arguments can be specified in config.py. Some important arguments are:

• DATA_DIR: Path to the data directory
• MODEL_DIR: Path to the model directory
• task_name: Name of the dataset to use as in-distribution (ID) data.
• ood_datasets: List of datasets to use as out-of-distribution (OOD) data.
• model_class: Type of model to use. Options are roberta, gpt2and t5 (base versions of all models used).
• do_train: If true, trains the model on the ID data before performing OOD detection.

After specifying the arguments, run the following command: python run_ood_detection.py

Running the command below will extend the pretraining process on a specified dataset. Once again, the values in config.json can be used to specify the dataset and model to be used. python tapt_training/pretrain_roberta.py

After the new model has been saved to the pretrained_models directory within the MODELS_DIR directory (specified in config.py), it can be used for OOD detection by running run_ood_detection.py.

If you find this repo helpful, you are welcome to cite our work: <--@inproceedings{uppaal2023fine, title={Is Fine-tuning Needed? Pre-trained Language Models Are Near Perfect for Out-of-Domain Detection}, author={Rheeya Uppaal and Junjie Hu and Yixuan Li }, booktitle = {Annual Meeting of the Association for Computational Linguistics}, year = {2023} }-->

@inproceedings{uppaal-etal-2023-fine,
    title = "Is Fine-tuning Needed? Pre-trained Language Models Are Near Perfect for Out-of-Domain Detection",
    author = "Uppaal, Rheeya  and Hu, Junjie  and Li, Yixuan",
    editor = "Rogers, Anna  and Boyd-Graber, Jordan  and Okazaki, Naoaki",
    booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2023",
    address = "Toronto, Canada",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2023.acl-long.717",
    doi = "10.18653/v1/2023.acl-long.717",
    pages = "12813--12832",
}

Our codebase borrows from the following:

@inproceedings{zhou2021contrastive,
  title={Contrastive Out-of-Distribution Detection for Pretrained Transformers},
  author={Zhou, Wenxuan and Liu, Fangyu and Chen, Muhao},
  booktitle={Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing},
  pages={1100--1111},
  year={2021}
}

@article{liu2020tfew,
  title={Few-Shot Parameter-Efficient Fine-Tuning is Better and Cheaper than In-Context Learning},
  author={Liu, Haokun and Tam, Derek and Muqeeth, Mohammed and Mohta, Jay and Huang, Tenghao and Bansal, Mohit and Raffel, Colin},
  journal={arXiv preprint arXiv:2205.05638},
  year={2022}
}