xwild / OlaGPT

Official implementation for "OlaGPT: Empowering LLMs With Human-like Problem-Solving Abilities" (keep updating)

OlaGPT: Empowering LLMs With Human-like Problem-Solving Abilities

OlaGPT carefully studied a cognitive architecture framework, and propose to simulate certain aspects of human cognition. Check out our paper for more information.

Requirements

python3.8
openai
langchain
datasets
faiss-cpu

Usage

Several important parameters:

• is_eval: whether it is evaluation mode
• question: valid when is_eval is False, suitable for testing a single question
• eval_full: whether to evaluate all data
• eval_num: valid when eval_full is False, can specify how many data to evaluate
• is_random: valid when eval_full is False, controls whether to randomly select eval_num data
• n_split: parallelism during evaluation, the higher the value, the higher the parallelism, but be aware of API concurrency limits
• model_name: controls the output path in prediction.

Quick Start

OlaGPT

python agents/multi_actions_agent.py --is_eval=True --dataset=aqua --model_name=cos --eval_full=True --n_split=30

Single Template

python agents/single_action_agent.py --is_eval=True --dataset=aqua --model_name=st --eval_full=True --n_split=30

More details

• definitions：utils/parser.py
• examples：
  • run OlaGPT
    • run-cos.sh
  • run Single Template
    • run-zeroshot.sh
    • run-random.sh
    • run-retrieval.sh
    • run-combine.sh

Citing OlaGPT

@article{xie2023olagpt,
  title={OlaGPT: Empowering LLMs With Human-like Problem-Solving Abilities}, 
  author={Yuanzhen Xie and Tao Xie and Mingxiong Lin and WenTao Wei and Chenglin Li and Beibei Kong and Lei Chen and Chengxiang Zhuo and Bo Hu and Zang Li},
  journal={arXiv preprint arXiv:2305.16334},
  year={2023}
}

License

This project is licensed under the Apache-2.0 License.