FunGPT – An open-source AI emotional regulation project based on the InternLM2.5 series models

What is FunGPT?

FunGPT is an open-source project built on the InternLM2.5 series of large language models, specifically designed for emotional regulation. It offers two core modes: the “Sweet Talk Mode” and the “Sharp Roast Mode.” The Sweet Talk Mode uplifts users’ moods with warm, unique compliments, while the Sharp Roast Mode uses witty and humorous jabs to help users release stress. FunGPT is powered by a lightweight 1.8B model series and leverages AWQ quantization technology, enabling both reduced GPU memory usage and faster inference.

Key Features of FunGPT

Sweet Talk Mode: When users are feeling down, this mode lifts their spirits by delivering personalized, unique compliments. The “Praise Master” makes users feel confident and appreciated.
Sharp Roast Mode: For users under pressure, this mode provides a humorous outlet for emotional release. The “Roast Master” offers sharp yet funny remarks that entertain while helping users vent in a lighthearted way.
Lightweight Model: The 1.8B lightweight model series provides outstanding performance with a smaller footprint. The use of AWQ quantization reduces GPU memory consumption while accelerating inference.

Technical Overview of FunGPT

Model Architecture: FunGPT is built on the InternLM2.5 series, which is based on the Transformer architecture. The model excels in language generation and understanding, leveraging multi-head attention to capture long-range dependencies and contextual information from multiple perspectives.
Fine-Tuning Techniques: To meet personalized user needs, FunGPT uses Xtuner for instruction and full fine-tuning. This enables the model to adapt effectively to specific tasks like Sweet Talk Mode and Sharp Roast Mode.
Quantization Technology: FunGPT employs AWQ (Adaptive Weight Quantization), which reduces parameter storage size, lowers the model’s usage threshold, and improves inference speed—making it efficient even on resource-constrained devices.

FunGPT Project Repository

GitHub Repository: https://github.com/Alannikos/FunGPT

Application Scenarios for FunGPT

Creative Inspiration: When users need fresh ideas, FunGPT can help spark creativity in writing, art, design, and other fields.
Entertainment and Leisure: For moments of boredom, FunGPT can recommend movies, music, books, or engage users in fun conversations, enriching their leisure time.