<div align="center"> <img src="Assets/svg/FunGPT-logo.svg" width="100"/>

📘Documentation | 🛠️Quick Start | 🤔Reporting Issues

English | 简体中文

</div> <div align="center"> <!-- PROJECT SHIELDS -->

</div>

---

Latest News 🎉

<details open> <summary><b>2024</b></summary>

• [2025/05/30] 🤗📊 Open-sourced the compliment fine-tuning dataset for BoostBot. ComplimentCorpus
• [2025/05/29] 🛠️🎯 Fixed some path-related bugs FIX
• [2024/11/30] 📢📺 We've released a project introduction video! You can find the video on Bilibili. If you like it, don't forget to give it a thumbs up 👍 and follow us!
• [2024/11/19] 🔒🛡️ Incorporate a sensitive word processing module to prevent the model from outputting harmful viewpoints.
• [2024/11/18] 🔧🌟 Completed Quick Start section documentation updates and resolved environment setup bugs.
• [2024/11/14] 🎉✨ Successfully quantized four models using the LMDeploy tool. The FunGPT family welcomes new members! Models are now available on HuggingFace.
• [2024/11/13] 🎉✨ Released two new 1.8B models, expanding the FunGPT family! The models are BanterBot_1_8b-chat and BoostBot_1_8b-chat.
• [2024/11/10] 🎉✨ Launched two brand-new 7B models, adding new members to the FunGPT family! The models are BanterBot-7b-chat and BoostBot-7b-chat.
• [2024/11/10] 🛠️🎯 Major project updates completed, including fixing known absolute path bugs.
• [2024/10/28] 🎈🥳 Achieved fine-tuning large language models using xtuner and released the first version of BoostBot_v1!
• [2024/10/19] 🎉💬 Developed a toolchain to generate fine-tuned dialogue data using the Chat-GLM4 series.
• [2024/10/03] 🎨🐞 Beautified the system interface and fixed some known bugs.
• [2024/10/02] 🚀💻 Added model loading and unloading mechanisms to optimize GPU memory usage.
• [2024/10/01] 😄🐍 Integrated an exception handling module to enhance application stability.
• [2024/09/28] 👋👋 Completed initial testing and evaluation of individual functionalities for LLM (InternLM2.5_1.8b), ASR (Sensevoice), and TTS (ChatTTS).

</details>

---

🌟 Project Introduction

$\quad$ In this fast-paced world 🌍, we all need a little spice to balance our lives. Whether you’re looking for some sweet compliments 🍭 to lift your spirits ✨ or a dose of sharp retorts 💥 to blow off steam 😤, FunGPT, developed based on the InternLM2.5 series models 🤖, is here to meet your needs 🎯!

🍬 Sweet Compliment Mode:

• Mood Booster 🌟✨: When you’re feeling down, our Sweet Compliment Mode will instantly lift your spirits, just like tasting an incredibly sweet candy.
• Confidence Fuel Station 💪🌈: Meanwhile, our Praise Master will compliment you in the most suitable and unique ways, making your confidence soar.

🔪 Sharp Retort Mode:

• Stress Release Valve 💥😤: When you’re feeling overwhelmed, our Retort Mode provides an outlet to blow off steam while delivering sharp remarks.
• Humorous Roasting Machine 😂👅: The words of the Roast Master are not only sharp but also humorous and imaginative, letting you experience brain-twisting comebacks while having fun.

💬 Join Our WeChat Group

Scan the QR code below to join our WeChat discussion group and communicate with more developers and users!

<div align="center"> <img src="Assets/image/wechat_group.jpg" width="200"/> </div>

🤔 Project Highlights

$\quad$FunGPT is built on the cutting-edge InternLM2.5 series models. Using Xtuner, we performed both instruction and full fine-tuning, enabling the models to meet personalized user needs. To enhance accessibility, we released the 1.8B series of lightweight models, which deliver exceptional performance despite their reduced size. Additionally, we employed AWQ quantization using LMDeploy on multiple models, saving GPU memory while boosting inference speed! ⚡

$\quad$ This project caters to every users by focusing on the two scenarios of praise and roasting, allowing users to chat with our two "Masters" to regulate their moods. For developers just entering the world of LLMs, we provide extensive documentation and video tutorials covering LLM, ASR, and TTS, making this project an ideal entry point for mastering large model development skills.

$\quad$ Our advantages include, but are not limited to:

1. 🤗 Master of Compliments: Generate sweet words to brighten your daily life.
2. 🗯️ Roasting Expert: Tailored responses with sharp wit, engaging in a battle of wits with "me."
3. 📊 Data Collection Guide: Fully open-source, helping you quickly grasp the creation of fine-tuning datasets.
4. 📖 Complete LLM Workflow Guide: Comprehensive code and documentation, open-source, making it easy to get started.
5. 🔊 Complete ASR Workflow Guide: Open everything to help you realize your dream of speech recognition.
6. 🎙️ Complete TTS Workflow Guide: From basics to advanced, fully open-source with no reservations!
7. 📂 Clear Structure: Detailed annotations and documentation ensure seamless onboarding.
8. ⚡ Model Quantization: Lower the usage barrier and experience the magic of AI anytime, anywhere.
9. 🎥 Video Tutorials: Stay tuned for our complete project introduction videos!

🖼️ Screenshots

|Original_7b_BoostBot|BoostBot-7b| |:---:|:---:| |<img src="./Assets/gif/Original_7b_BoostBot.gif" width="400">|<img src="./Assets/gif/BoostBot-7b.gif" width="400">|

|Original_7b_BanterBot|BanterBot-7b| |:---:|:---:| |<img src="./Assets/gif/Original_7b_BanterBot.gif" width="400">|<img src="./Assets/gif/BanterBot-7b.gif" width="400">|

🏗️ Project Architecture Diagram

🧳 Model Collection

| Model | Base | Type | Link | | ------------------------------ | --------------------- | -------------------------- | ------------------------------------------------------------ | | BanterBot-7b-chat | internlm2_5_chat_7b | Pre-trained + QLoRA fine-tuning | HuggingFace<br/>OpenXLab | | BoostBot-7b-chat | internlm2_5_chat_7b | Pre-trained + QLoRA fine-tuning | HuggingFace<br/>OpenXLab | | BanterBot_1_8b-chat | internlm2_5_chat_1_8b | Pre-trained + QLoRA fine-tuning | HuggingFace<br>OpenXLab | | BoostBot_1_8b-chat | internlm2_5_chat_1_8b | Pre-trained + QLoRA fine-tuning | HuggingFace<br>OpenXLab | | BanterBot-7b-chat-w4a16-4bit | internlm2_5_chat_7b | Pre-trained + QLoRA fine-tuning + w4a16 quantization | HuggingFace<br>OpenXLab | | BoostBot-7b-chat-w4a16-4bit | internlm2_5_chat_7b | Pre-trained + QLoRA fine-tuning + w4a16 quantization | HuggingFace<br>OpenXLab | | BanterBot_1_8b-chat-w4a16-4bit | internlm2_5_chat_1_8b | Pre-trained + QLoRA fine-tuning + w4a16 quantization | [H