https://github.com/chihaya-yuka/multiplex-cot

[arXiv 2501.13117]The Multiplex CoT makes AI more thoughtful.
https://github.com/chihaya-yuka/multiplex-cot

api cot deep-learning llm lrm orange-ai prompt-engineering

Last synced: 4 months ago
JSON representation

[arXiv 2501.13117]The Multiplex CoT makes AI more thoughtful.

• Host: GitHub
• URL: https://github.com/chihaya-yuka/multiplex-cot
• Owner: Chihaya-Yuka
• License: apache-2.0
• Created: 2025-01-20T01:26:24.000Z (9 months ago)
• Default Branch: main
• Last Pushed: 2025-02-09T12:13:51.000Z (9 months ago)
• Last Synced: 2025-04-07T22:23:59.330Z (7 months ago)
• Topics: api, cot, deep-learning, llm, lrm, orange-ai, prompt-engineering
• Language: Jupyter Notebook
• Homepage: https://arxiv.org/abs/2501.13117
• Size: 187 KB
• Stars: 18
• Watchers: 0
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Multiplex CoT: A way for the LLM to review its own thinking while reasoning by initiating double CoT thinking

[![Open Auto-CoT in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1rB3Re3D7alu28JgChFUy6BKmvmNADsdk?usp=sharing) [![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/mygo-multiplex-cot-a-method-for-self/gsm8k-on-gsm8k)](https://paperswithcode.com/sota/gsm8k-on-gsm8k?p=mygo-multiplex-cot-a-method-for-self) [![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/mygo-multiplex-cot-a-method-for-self/humaneval-on-humaneval-1)](https://paperswithcode.com/sota/humaneval-on-humaneval-1?p=mygo-multiplex-cot-a-method-for-self) [![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/mygo-multiplex-cot-a-method-for-self/llm-real-life-tasks-on-llm-real-life-tasks)](https://paperswithcode.com/sota/llm-real-life-tasks-on-llm-real-life-tasks?p=mygo-multiplex-cot-a-method-for-self) [![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/mygo-multiplex-cot-a-method-for-self/mmlu-on-mmlu-pro)](https://paperswithcode.com/sota/mmlu-on-mmlu-pro?p=mygo-multiplex-cot-a-method-for-self)

By employing the Prompt method, the LLM can attain an effect that closely resembles that of the LRM without necessitating additional training.

In the context of reasoning and decision-making, Multiplex CoT (Chain of Thought) enables the model to simulate a form of self-reflection, improving its ability to generate coherent, logical answers. This method works by prompting the LLM to first generate a chain of reasoning (CoT), then iteratively reviewing and refining it by initiating a second round of reasoning, which acts as a critique or review of the first.

![Figure 1](Figure_1.png)

## Quickly start

Run `Multiplex_CoT.ipynb`.

## How to use

See `example.py`.