Skip to content

Formal-LLM: Integrating Formal Language and Natural Language for Controllable LLM-based Agents

Notifications You must be signed in to change notification settings

agiresearch/Formal-LLM

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

23 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Formal-LLM: Integrating Formal Language and Natural Language for Controllable LLM-based Agents

Recent advancements on Large Language Models (LLMs) enable AI Agents to automatically generate and execute multi-step plans to solve complex tasks. However, since LLM's content generation process is hardly controllable, current LLM-based agents frequently generate invalid or non-executable plans, which jeopardizes the performance of the generated plans and corrupts users' trust in LLM-based agents. In response, this work proposes a novel "Formal-LLM" framework for LLM-based agents by integrating the expressiveness of natural language and the precision of formal language. Specifically, the framework allows human users to express their requirements or constraints for the planning process as an automaton. A stack-based LLM plan generation process is then conducted under the supervision of the automaton to ensure that the generated plan satisfies the constraints, making the planning process controllable. We conduct experiments on both benchmark tasks and practical real-life tasks, and our framework achieves over 50% overall performance increase, which validates the feasibility and effectiveness of employing Formal-LLM to guide the plan generation of agents, preventing the agents from generating invalid and unsuccessful plans. Further, more controllable LLM-based agents can facilitate the broader utilization of LLM in application scenarios where high validity of planning is essential.

Citation

@article{li2024formalllm,
      title={Formal-LLM: Integrating Formal Language and Natural Language for Controllable LLM-based Agents}, 
      author={Zelong Li and Wenyue Hua and Hao Wang and He Zhu and Yongfeng Zhang},
      journal={arXiv:2402.00798},
      year={2024}
}

Requirements

  • Python==3.9
  • PyTorch==1.13.1
  • transformers==4.28.0
  • langchain==0.0.153
  • peft==0.7.1

Usage

  1. Clone this repo.

  2. Create a conda virtual environment and install the Pytorch matching your CUDA version. For example, for CUDA version 11.7:

conda create -n your_env_name python=3.9
conda activate your_env_name

conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.7 -c pytorch -c nvidia
  1. Install necessary packages:
pip install -r requirements.txt
  1. Download the preprocessed data from this Google Drive link, put it into the FormalLLM/src/ folder, then unzip it.

  2. The codes for the real-world tasks are in the Formal/src/realworld_tasks folder, and the codes for benchmark tasks are in the Formal/src/open_source and Formal/src/closed_source folders. Specifically, the open_source LLMs include Flan-T5-large, Vicuna-7B, and LLaMA-2-13B. The closed_source LLMs include GPT-3.5-turbo, Claude-2, and GPT-4.

  3. Before running any codes, you need replace "YOUR OPENAI TOKEN", "HUGGINGFACE TOKEN", and "YOUR CLAUDE TOKEN", with your OpenAI, Huggingface, and Claude API keys, respectively.

  4. Make sure you are in the Formal/src folder before running the codes. Otherwise,

cd src
  1. To evaluate the performance of closed_source LLM-based agent with our Formal-LLM framework, you need to run Python with these arguments: huggingface_cache and log_name. For example,
python closed_source/FLLM_ChatGPT.py --log_name ./0.txt --huggingface_cache ./
  1. To evaluate the performance of open_source LLM-based agent with our Formal-LLM framework, you need to run Python with these arguments: cache_dir, log_name, and ckp_dir (directory for saving checkpoints). For example,
python open_source/FLLM_schema-flan-t5.py --log_name ./0.txt --cache_dir ./ --ckp_dir ./
  1. To evaluate the performance on real-life tasks (use daily planning as an example):
python realworld_tasks/daily.py

Reference

  • We leveraged the dataset of OpenAGI projects to implement our experiment.

About

Formal-LLM: Integrating Formal Language and Natural Language for Controllable LLM-based Agents

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 4

  •  
  •  
  •  
  •