Grounded Language Agent for Product Search via Intelligent Web Interactions

In this repository, the code for our paper "Grounded Language Agent for Product Search via Intelligent Web Interactions" is provided

Installation steps

1. Clone the project

git clone https://github.com/MultifacetedNLP/Web-Agents-Unsupervised.git Web-Agents-Unsupervised; cd Web-Agents-Unsupervised

2. Create conda env

conda create -n WebAgent python=3.10.8; conda activate WebAgent

3. Install PyTorch

conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.3 -c pytorch

4. Install required packages

pip install -r requirements.txt

5. Install Wevshop Environment:

cd web_agent_site; pip install -e .; cd ..

6. Set up $SCRATCH (path for storing datsets and LLM models), and$PROJECT (path to the project directory) if they are empty

export SCRATCH="/path/to/scratch_folder"
export PROJECT="/path/to/project_folder"

7. Download the Datasets and Prepare the Search Engine for Environment

chmod +x ./setup.sh; ./setup.sh -d all

8. Install Accelerate

cd v0.13.2/accelerate-0.13.2; pip install -e .; cd ../..

9. Install Lamorel

cd lamorel/lamorel; pip install -e .; cd ../..

Train

The following code will trian the Flan-T5 large model in the webshop environment using Proximal Policy Optimization (local machine)

chmod +x ./experiments/bash_files/train_ppo/local_train_ppo_run.sh;
./experiments/bash_files/train_ppo/local_train_ppo_run.sh

The following code will trian the Flan-T5 large model in the webshop environment using Proximal Policy Optimization (slurm work manager)

chmod +x ./experiments/bash_files/train_ppo/slurm_train_ppo_run.sh;
./experiments/bash_files/train_ppo/slurm_train_ppo_run.sh

Inference

The following code will download the model that was only trained with PPO, and put it in the $SCATCH path

cd $SCRATCH;
mkdir -p storage/models;
cd storage/models;
gdown "https://drive.google.com/uc?id=1GYumAWzrIyo-fby5wT5JsXjkto-8bbzq&confirm=t";
unzip flan_t5_large_2_observations_only_ppo_1000000_steps.zip

Run the following code to perform inference on the only-ppo model (local machine)

chmod +x ./experiments/bash_files/test_only_ppo/local_test_run.sh;
./experiments/bash_files/test_only_ppo/local_test_run.sh

Run the following code to perform inference on the only-ppo model (slurm work manager)

chmod +x ./experiments/bash_files/test_only_ppo/slurm_test_run.sh;
./experiments/bash_files/test_only_ppo/slurm_test_run.sh

Citation

If you use this code or reference our study in your work, please cite:

@inproceedings{@inproceedings{fereidouni-etal-2024-grounded,
    title = "Grounded Language Agent for Product Search via Intelligent Web Interactions",
    author = "Fereidouni, Moghis  and Mosharrof, Adib  and Siddique, A.b.",
    editor = "Kumar, Sachin  and Balachandran, Vidhisha  and Park, Chan Young  and Shi, Weijia  and Hayati, Shirley Anugrah  and Tsvetkov, Yulia  and Smith, Noah  and Hajishirzi, Hannaneh  and Kang, Dongyeop  and Jurgens, David",
    booktitle = "Proceedings of the 1st Workshop on Customizable NLP: Progress and Challenges in Customizing NLP for a Domain, Application, Group, or Individual (CustomNLP4U)",
    month = nov,
    year = "2024",
    address = "Miami, Florida, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.customnlp4u-1.7/",
    doi = "10.18653/v1/2024.customnlp4u-1.7",
    pages = "63--75"
}