CongJ-Pan/XiaoHong-v1
CongJ-Pan/XiaoHong-v1 is a Qwen3-8B based large language model, fine-tuned using QLoRA with the RA-DIT LM-FT method, specifically designed for Retrieval-Augmented Generation (RAG) scenarios. This model specializes in classical Chinese literature, poetry, historical allusions, and traditional culture, offering deep reasoning capabilities activated via an internal tag. It is optimized for traditional Chinese contexts and classical literary tone, providing both factual answers and in-depth academic analysis. The model's unique feature is its forced reasoning mechanism, breaking down complex problems into logical steps before generating a refined answer.
Loading preview...
XiaoHong-v1: Classical Chinese Literature RAG Model
XiaoHong-v1 (小紅) is a specialized large language model developed by CongJ-Pan, built upon the Qwen3-8B architecture. It leverages the RA-DIT (Retrieval-Augmented Dual Instruction Tuning) LM-FT method, fine-tuned with QLoRA, making it ideal for Retrieval-Augmented Generation (RAG) applications.
Key Capabilities & Features
- Classical Literature Expertise: Highly proficient in topics related to Dream of the Red Chamber, classical Chinese literature, poetry, historical allusions, and traditional culture.
- Forced Reasoning: Incorporates a unique
<think>tag mechanism for deep logical deduction and knowledge organization, ensuring more accurate and profound answers to complex queries. This requires specific client-side handling for deployment with vLLM or HF Endpoints. - Traditional Chinese Optimization: Deeply optimized for traditional Chinese contexts and classical literary tone.
- Persona-driven: Embodies a friendly and professional classical literature assistant persona named "XiaoHong" (小紅).
Use Cases & Considerations
XiaoHong-v1 excels at providing both concise factual responses and in-depth academic analysis within its specialized domain. It is particularly suited for applications requiring detailed understanding and generation of content related to classical Chinese studies. Users should note that the model primarily supports traditional Chinese and may hallucinate on modern or out-of-domain topics. Proper implementation of the recommended System Prompt and specific deployment strategies (especially for vLLM/HF Endpoints) are crucial to leverage its full reasoning capabilities.