GPT4DST: Interact with AI Pro-Gamer in Don't Starve Together

🔥 GPT4DST

English | 中文

Introduction

GPT4DST is a secondary development application based on the langchain library calling OpenAI's large language model. The project aims to provide an intelligent interactive experience for the Don't Starve Together (DST) game by developing specialized mods to interact with players.

Quick Start

---

Data

Further Pretrain Data

Instruction Tuning: DSTSignal

Benchmark: DSTBench

Code

Further Pretrain Configurations

Fine-Tuning

Cases

Evaluation

Contributors

Acknowledgements

License

Citation

---

Developer Notes

DATA COLLECTION AND CURATION

• Collect Raw DST Knowledge from Wiki Fandom using API
• Clean Raw DST Knowledge into Formatted DST Knowledge/Structured DST Knowledge
• Construct pipeline for creating DSTSignal/DSTBench

LLM DOMAIN SPECIFIED USING RAG

• Consolidate DST Knowledge into vectorstore using Chroma/Milvus/...
• Implement RAG to enhance LLM capabilities using langchain/llamaindex/...
• Implement adavanced RAG techniques (Create Q-A Pairs with SOTA Languange Model)
• Real-Time Updates (Ref to ChatOllama Project) (Scripts Update Timely)

DEPLOYMENT WITH WEBPAGES

• Deployment project using Vue/React/Streamlit

DEPLOYMENT WITH API

• Deployment project using FASTAPI

INTERACTION WITH ENVIRONMENT USING AGENT

• Import llm modules into the DST mod
• Achieve simple QA in DST

---

Outline of Scientific Writing

Paper Title: DST-Agent: Intelligent Game Playing Guidance by Empowering Large Language Models with Environment Interaction
Abstract: ...
Main Contributions: DSTBench & DSTSignal ; Deep Integration in Game as Workshop Mods

INTRODUCTION & RELATED WORK

• Introduction to LLMs
  • A Survey of Large Language Models
• Knowledge Injection
  • Fine-Tuning or Retrieval? Comparing Knowledge Injection in LLMs
  • Retrieval-Augmented Generation for Large Language Models: A Survey
• Game Playing Agent
  • A Survey on Game Playing Agents
  • PokeLLMon: A Human-Parity Agent for Pokemon Battles with Large Language
  • SPRING: Studying the Paper and Reasoning to Play Games

DATA COLLECTION AND CURATION

• Pipeline of DSTSignal & DSTBench

KNOWLEDGE INJECTION

• Retrival-Augmented Generation(RAG)
• Chain of Thought (COT)/Tree of Thought(TOT)/Chain of Abstraction(COA)

REINFORCEMENT LEARNING FOR ENVIRONMENT INTERACTION

AGENT

• Reasoning and Act(ReACT)

CASES STUDY

EVALUATION AND RESULTS

CONCLUSION