Yougen/Qwen3Fangwusha14B
Yougen/Qwen3Fangwusha14B is a 14 billion parameter auto-regressive language model developed by Yougen Yuan, fine-tuned from Qwen3-14B. This model specializes in enhancing Chinese dialogue capabilities, instruction following, and general task performance. Optimized with BF16 precision on high-quality Chinese datasets, it excels in Chinese semantic understanding, logical reasoning, and multi-turn conversations. It is designed to provide a high-quality, reliable AI assistant service for Chinese users across various NLP tasks.
Loading preview...
Overview
Yougen/Qwen3Fangwusha14B is a 14 billion parameter auto-regressive language model, fine-tuned by Yougen Yuan from the Qwen3-14B architecture. It leverages BF16 precision training on high-quality Chinese datasets to significantly improve Chinese dialogue, instruction following, and overall task performance. The model is part of the Fangwusha series, focusing on delivering a robust and reliable AI assistant experience for Chinese users.
Key Capabilities
- Enhanced Chinese NLP: Optimized for Chinese semantic understanding, logical reasoning, and multi-turn dialogue.
- Versatile Applications: Capable of Chinese conversation and Q&A, text generation, information extraction, summarization, translation, code assistance, and creative writing.
- Fine-tuning Potential: Can be further fine-tuned for specialized applications like domain-specific knowledge Q&A, customer service bots, educational systems, and enterprise intelligent assistants.
Training Details
The model was fine-tuned using diverse, high-quality Chinese datasets, including general dialogue, instruction-following, knowledge Q&A, and logical reasoning datasets. These datasets underwent rigorous quality filtering and deduplication. Training utilized an AdamW optimizer with BF16 mixed precision.
Limitations and Recommendations
Users should be aware that the model may generate inaccurate or biased information, especially in specialized domains, and its English capabilities are weaker than its Chinese. It is recommended to fact-check outputs, use cautiously in high-risk scenarios, and report any inappropriate content.