(原理)Prompt Engineering

Basic Prompting [2] #

Zero-Shot Prompting [3] #

Few-Shot Prompting [3] #

CoT [2] #

Chain-of-Thought Prompting(CoT) [3] #

  • Few-shot CoT
  • Zero-shot COT
    “Let’s think step by step”

Self-Consistency(CoT-SC) [3] #

The idea is to sample multiple, diverse reasoning paths through few-shot CoT, and use the generations to select the most consistent answer.

Tree of Thoughts (ToT) #

CoT vs. CoT-SC vs. ToT [3] #

TOT

Tips and Extensions [2] #

Self-Ask

Automatic Prompt Design [2] #

  • Automatic Chain-of-Thought (Auto-CoT) [3]

Six strategies for getting better results[1] #

Write clear instructions #

清晰的指令

Provide reference text #

Split complex tasks into simpler subtasks # 

复杂任务简单化

Give the model time to “think” #

给模型时间去思考

Use external tools #

使用外部工具

Test changes systematically #

优点vs 缺点 #

优点 #

简单 容易上手

缺点 #

  • 上限有限
  • 模型适配 prompt要适配每个模型

参考 #

  1. Prompt engineering openai
  2. Prompt Engineering paper
  3. Prompt Engineering Guide guide Prompt-Engineering-Guide *** git

1xx. 【社区第十三讲】 老刘说NLP线上交流 *** 很全

1xx. [论文阅读] Prompt Engineering综述

1xx. The Prompt Landscape langchain

1xx. CometLLM - suite of LLMOps tools - track and visualize LLM prompts and chains

1xx. 大模型 PUA 指南：来自 Google Meta Microsoft 等大厂

1xx. NLP（十三）：Prompt Engineering 面面观

1xx. prompt-engineering git

1xx. Chain-of-Thought Prompting 简读

1xx. ChatGPT应用端的Prompt解析：从概念、基本构成、常见任务、构造策略到开源工具与数据集

1xx. AutoPrompt Repo git

案例 #

  1. 运维大模型探索之 Text2PromQL 问答机器人 架构图， 最后两个重点总结 未