README.md

This repo is for source code of paper "Beyound Language Processing: LLMs Rules-Injected Instruction Tuning for Traffic Prediction"

Run

The traffic rules are saved in data/rule_samples. Training and testing data are saved in data/prompts. Unzip the experimental data in data/prompts for tuning and testing the model.

Step 1. Create a folder named 'pretrained_model' in the current directory, then download and store the follwing language models

-gte-base https://huggingface.co/thenlper/gte-base

-vicuna https://huggingface.co/lmsys/vicuna-7b-v1.5-16k

step 1. Create enviroment and install package

conda create -n rulegpt rulegpt python==3.10.14
conda activate rulegpt
pip install -r requirement.txt
pip install -e .

step 2. Finetune Model

bash scripts/finetune_lora.sh

step 3. Test Model

bash scripts/eval_model.sh ## the result output will be saved into answers-file

Acknowledgement

This implementation is mainly based on LLaVA. We thank the authors for their great jobs!