LLM Riddles

:thinking: What's This

Welcome to LLM Riddles! This is a game of wits and courage with language models. In the game, you need to construct questions that interact with the language model to get answers that meet the requirements. In this process, you can use your brain and use all the methods you can think of to get the model to output the results required by the answer.

:space_invader: How to Play

We provide an online version for players to directly access and try out. Local deployment can be done in the following ways:

ChatGPT + Chinese

QUESTION_LANG=cn QUESTION_LLM='chatgpt' QUESTION_LLM_KEY=<your API key> python3 -u app.py

ChatGPT + English

QUESTION_LANG=en QUESTION_LLM='chatgpt' QUESTION_LLM_KEY=<your API key> python3 -u app.py

Mistral-7B-Instruct-v0.1 + English

QUESTION_LANG=en QUESTION_LLM='mistral-7b' python3 -u app.py

:technologist: Why Doing This

Our goal is to use this game to give participants a deeper understanding of the fascinating aspects of prompt engineering and natural language processing. This process will show players how to cleverly construct prompts and how to use them to trigger surprising responses from artificial intelligence systems, while also helping them better understand the incredible power of deep learning and natural language processing technologies. .

:raising_hand: How to Submit a Custom Level

If you have interesting questions or ideas, players are welcome to submit their own ideas. You can Initiate a Pull Request and submit it to us. We will include it in the level after approval. The question format should include the following points:

Pull Request title, example: feature(username): Chapter X-Level Design
The ID you want to be mentioned
Modify the corresponding chapter question files
Modification of init.py

For a complete example, please refer to: Submit your own level design

:writing_hand: Roadmap

Support custom levels
Online trial link
Hugging Face Space link
Support Mistral-7B（English version）
Support Baichuan2-7B（Chinese version）
Support LLaMA2-7B（English version）
LLM inference speed optimization

:speech_balloon: Feedback and Contribution

Start an Issue on GitHub
Contact us by email ([email protected])
Discuss on OpenDILab's WeChat group (i.e. add us on WeChat: ding314assist)

:star2: Special Thanks

Thanks to Haoqiang Fan for his original idea and title, which provided inspiration and motivation for the development and expansion of this project.
Thanks to HuggingFace for supporting and assisting the game.
Thanks to LLM Riddles contributors for their implementation and support.

:label: License

All code within this repository is under Apache License 2.0.

(back to top)