This repository serves as a comprehensive, curated list of research papers focused on reasoning with language model prompting, targeting researchers and practitioners in NLP and AI. It aims to systematically organize and present the rapidly evolving landscape of LLM reasoning capabilities, providing a valuable resource for understanding methods, benchmarks, and tools in this domain.

How It Works

The project categorizes research papers based on reasoning methods (e.g., Chain of Thought, Self-Optimization, Tool Learning) and knowledge enhancement strategies (Implicit, Explicit). It also lists relevant benchmarks and tools, offering a structured overview of the field. The repository is maintained as a living document, updated with new publications and research directions.

Quick Start & Requirements

• Access: The primary interaction is through browsing the GitHub repository and its linked survey paper.
• Requirements: No specific software installation is required to access the paper list and associated resources. Access to academic papers (e.g., via arXiv) is needed for full content review.
• Resources: The survey paper is available at https://github.com/zjunlp/Prompt4ReasoningPapers.

Highlighted Details

• Extensive categorization of reasoning methods, including single-stage, multi-stage, self-optimization, and tool learning.
• Comprehensive lists of benchmarks for various reasoning skills (Arithmetic, Commonsense, Symbolic, Logical, Multimodal).
• Curated list of tools and libraries relevant to LLM prompting and reasoning (e.g., LangChain, LogiTorch, λprompt).
• Includes a detailed survey paper published at ACL 2023, providing a structured overview and analysis of the field.

Maintenance & Community

The repository is actively maintained by the zjunlp team, with recent updates in early 2024. The project encourages community contributions for adding new papers and updates.

Licensing & Compatibility

The repository content is typically licensed under permissive terms (e.g., MIT for the code, though the survey paper itself is governed by ACL proceedings). Compatibility for commercial use is generally high, as it primarily lists research papers and resources.

Limitations & Caveats

The repository is a curated list and does not provide executable code for the described methods. The rapid pace of research means the list may not always be exhaustive or perfectly up-to-date between updates.