https://github.com/madaan/self-refine

LLMs can generate feedback on their work, use it to improve the output, and repeat this process iteratively.
https://github.com/madaan/self-refine

chatgpt few-shot-learning gpt-35 gpt-4 language-generation large-language-models llms prompting prompts reasoning

Last synced: 2 months ago
JSON representation

LLMs can generate feedback on their work, use it to improve the output, and repeat this process iteratively.

• Host: GitHub
• URL: https://github.com/madaan/self-refine
• Owner: madaan
• License: apache-2.0
• Created: 2023-03-28T13:32:20.000Z (about 2 years ago)
• Default Branch: main
• Last Pushed: 2024-01-13T18:34:23.000Z (over 1 year ago)
• Last Synced: 2024-08-03T01:20:56.057Z (11 months ago)
• Topics: chatgpt, few-shot-learning, gpt-35, gpt-4, language-generation, large-language-models, llms, prompting, prompts, reasoning
• Language: Python
• Homepage: https://selfrefine.info
• Size: 53.7 MB
• Stars: 534
• Watchers: 13
• Forks: 43
• Open Issues: 8
• Metadata Files:
  • Readme: README.md
  • License: LICENSE
  • Citation: CITATION.bib

Awesome Lists containing this project

• awesome-ChatGPT-repositories - self-refine - LLMs can generate feedback on their work, use it to improve the output, and repeat this process iteratively. (Prompts)
• StarryDivineSky - madaan/self-refine - refine展示了大语言模型（LLM）如何通过自我反思和迭代改进其输出。核心思想是让LLM生成对其自身工作的反馈，并利用这些反馈来优化结果，重复此过程以达到更好的效果。这种自我完善的方法可以显著提升LLM的性能。该项目探索了LLM在没有人工干预的情况下，独立提升自身能力的可能性。通过迭代地生成、评估和改进，LLM能够逐步完善其输出质量。这种自我反思机制模拟了人类的学习过程，使LLM能够更有效地解决复杂问题。项目重点在于展示和研究这种自我迭代改进的潜力，为未来LLM的自主学习和发展提供思路。 (A01_文本生成_文本对话 / 大语言对话模型及数据)