https://github.com/ysymyth/ReAct

[ICLR 2023] ReAct: Synergizing Reasoning and Acting in Language Models
https://github.com/ysymyth/ReAct

decision-making large-language-models llm prompting reasoning

Last synced: 3 months ago
JSON representation

[ICLR 2023] ReAct: Synergizing Reasoning and Acting in Language Models

• Host: GitHub
• URL: https://github.com/ysymyth/ReAct
• Owner: ysymyth
• License: mit
• Created: 2022-11-13T19:51:09.000Z (almost 2 years ago)
• Default Branch: master
• Last Pushed: 2024-02-06T02:34:32.000Z (9 months ago)
• Last Synced: 2024-06-14T01:59:30.859Z (5 months ago)
• Topics: decision-making, large-language-models, llm, prompting, reasoning
• Language: Jupyter Notebook
• Homepage:
• Size: 6.16 MB
• Stars: 1,687
• Watchers: 16
• Forks: 172
• Open Issues: 3
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• StarryDivineSky - ysymyth/ReAct
• awesome-and-novel-works-in-slam - [Code

README

        
# ReAct Prompting

GPT-3 prompting code for ICLR 2023 paper [ReAct: Synergizing Reasoning and Acting in Language Models](https://arxiv.org/abs/2210.03629).

To use ReAct for more tasks, consider trying [LangChain's zero-shot ReAct Agent](https://python.langchain.com/docs/modules/agents/agent_types/react.html).

## Setup

You need to first have an OpenAI API key and store it in the environment variable ``OPENAI_API_KEY`` (see [here](https://help.openai.com/en/articles/5112595-best-practices-for-api-key-safety)).

Package requirement: ``openai``, and install ``alfworld`` following instructions [here](https://github.com/alfworld/alfworld).

## Experiments

Run ``{hotpotqa,fever,alfworld,webshop}.ipynb``. As HotpotQA and FEVER have large validation sets, we only run 500 random examples (see notebooks). We find PaLM and GPT-3 are better at different tasks.

|                    | HotpotQA (500 random dev, EM) | FEVER (500 random dev, EM) | AlfWorld (success rate) | WebShop  (success rate) |

|--------------------|-------------------------------|----------------------------|-------------------------|-------------------------|

| PaLM-540B (paper)  | 29.4                          | 62.2                       | 70.9                    | 40                      |

| GPT-3 (davinci-002) | 30.4                          | 54                         | 78.4                    | 35.8                    |

## Citation

```bibtex

@inproceedings{yao2023react,

  title = {{ReAct}: Synergizing Reasoning and Acting in Language Models},

  author = {Yao, Shunyu and Zhao, Jeffrey and Yu, Dian and Du, Nan and Shafran, Izhak and Narasimhan, Karthik and Cao, Yuan},

  booktitle = {International Conference on Learning Representations (ICLR) },

  year = {2023},

  html = {https://arxiv.org/abs/2210.03629},

}

```