What is FunGPT
FunGPT is an open-source project developed based on the InternLM2.5 series large models, designed specifically for emotional regulation. It has two core functions: Sweet Talk Mode and Sharp Roast Mode. Sweet Talk Mode can lift a user's mood with warm words and unique praise, while Sharp Roast Mode helps users release stress in a humorous and witty way. FunGPT uses the 1.8B series lightweight models, combined with AWQ quantization technology, which both saves GPU memory and improves inference speed.
Main features of FunGPT
- Sweet Talk Mode: When a user is feeling down, this mode can instantly lift the user's mood. Master Praiser will praise the user in the most suitable and unique way, making the user's confidence soar.
- Sharp Roast Mode: When a user feels overwhelmed by stress, this mode lets the user find an outlet while roasting others. Master Roaster's words are not only sharp but also humorous and witty, allowing users to experience wildly imaginative ways of roasting people in the process.
- Lightweight models: The 1.8B series lightweight models have been released, with a smaller size and outstanding performance. They use AWQ quantization technology to save GPU memory while improving inference speed.
Technical principles of FunGPT
- Model architecture: FunGPT uses the InternLM2.5 series models as its base architecture. The model is based on the Transformer architecture and has powerful language generation and understanding capabilities. The core advantage of the Transformer architecture lies in its multi-head attention mechanism, which can examine text from different perspectives and capture long-distance dependencies and contextual information.
- Fine-tuning technology: To meet personalized user needs, FunGPT uses Xtuner for instruction and full fine-tuning. In this way, the model can better adapt to specific task scenarios, such as Sweet Talk Mode and Sharp Roast Mode.
- Quantization technology: FunGPT adopts AWQ (Adaptive Weight Quantization) technology. By reducing the storage space of model parameters, it lowers the barrier to using the model, improves inference speed, and enables the model to run efficiently even on resource-constrained devices.
FunGPT project address
- Github repository：https://github.com/Alannikos/FunGPT
Application scenarios of FunGPT
- Creative inspiration: When new creative inspiration is needed, FunGPT can help users generate interesting ideas, such as providing inspiration for writing, painting, or design.
- Entertainment and pastime: When users feel bored, FunGPT can recommend entertainment content such as movies, music, and books, or have interesting conversations with users, making their leisure time richer.