Kasugan0/Nyarin-4B
Kasugan0/Nyarin-4B is a 4 billion parameter Chinese dialogue model, fine-tuned from Qwen3/Qwen3-4B-Instruct-2507. It specializes in generating "cat-girl" persona responses, trained on the NekoQA-10K dataset. This model is designed for interactive conversational applications where a specific, endearing character persona is desired.
Loading preview...
Nyarin-4B: Cat-Girl Persona Chinese Dialogue Model
Nyarin-4B is a 4 billion parameter instruction-tuned model developed by Kasugan0, based on the Qwen3/Qwen3-4B-Instruct-2507 architecture. Its primary distinction is its specialization in generating responses with a "cat-girl" (猫娘) persona, aiming to provide users with an engaging and unique conversational experience.
Key Capabilities
- Specialized Persona Generation: Fine-tuned specifically on the NekoQA-10K dataset for 3 epochs, enabling it to adopt a cat-girl character in dialogues.
- Chinese Dialogue: Optimized for conversational interactions in Chinese.
- Efficient Deployment: Available in various formats including BF16 (main branch for highest quality), AWQ (4-bit) for balanced precision and speed, and GGUF (Q5_K_M, Q4_K_M) for broad compatibility across different hardware configurations, including CPU-only setups.
Training Details
The model was fine-tuned using DeepSpeed ZeRO-3 with BF16 precision on 3 NVIDIA A800 80GB GPUs. The training process involved 3 epochs on the NekoQA-10K dataset.
Usage Considerations
Users should be aware that the model inherits potential limitations and biases from its base model. Responsible use is encouraged, particularly given its specialized persona generation capabilities. Example Python code is provided for quick setup and interaction, demonstrating how to load the model with quantization and apply a system prompt to guide its persona.