yoyo-yun / PLoRA

[AAAI‘24] The official PyTorch implimentation of our AAAI 2024 paper: Personalized LoRA for Human-Centered Text Understanding

Personalized LoRA for Human-Centered Text Understanding

Installation

1. download datasets from here:

   dataset	IMDB	YELP	GDRD	PPR
   original link	Link	Link	Link	Link

   and organize data as follows:

   |--$corpus
   |  |-- imdb2
   |  |  | a
   |  |  | b
   |  |-- yelp2
   |  |  | a
   |  |  | b
   |  |-- gdrd
   |  |  | a
   |  |  | b
   |  |-- ppr
   |  |  | a
   |  |  | b
   

2. install required packages:

   pip install -r requirement.txt

Training and Evaluating

1. config with configuration files:

   |--$cfgs
   |  |-- {Model}_model.yml # configurations for each base model
   |  |-- config.py # configurations for running settings
   |  |-- constants # constant configurations
   

2. training and evaluating on full-shot scenario (e.g., IMDB datasets)

   python run.py --run train --version test_fullshot --gpu 0 --dataset imdb2a --model bert

3. training and evaluating on few-shot scenario (e.g., IMDB datasets)

   python run.py --run fewshot --version test_fewshot --gpu 0 --dataset imdb2b --model bert

4. evaluating on zero-shot scenario (e.g., IMDB datasets)

   python run.py --run zeroshot --version test_zeroshot --gpu 0 --dataset imdb2b --model bert

Citation

awaiting update.