Taiyi-LLM - A bilingual biomedical large language model based on Qwen-7B-base.

## Base Model of Taiyi-LLM Taiyi-LLM is fine-tuned from **Qwen-7B-base**, a large language model from Alibaba Cloud's Tongyi Qianwen series. It has **7 billion parameters** and was pretrained on over **2 trillion tokens** of multilingual text, code, and mathematical content. ## Biomedical Applications of Taiyi-LLM Taiyi-LLM supports: - **Medical Q&A** (single/multi-turn dialogue) - **Literature summarization** - **Disease prediction** - **Biomedical information extraction** (NER, relation/event extraction) - **Medical report generation** - **Bilingual translation** of biomedical texts - **Text classification** and semantic similarity analysis ## Hosting Platforms for Taiyi-LLM Key platforms include: - **GitHub**: [Primary repository](https://github.com/DUTIR-BioNLP/Taiyi-LLM) - **Hugging Face**: [Model card](https://huggingface.co/DUTIR-BioNLP/Taiyi-LLM) - **ModelScope**: [Demo](https://modelscope.cn/studios/DUTIRbionlp/Taiyi2_Demo) - **Academic paper**: [DOI:10.1093/jamia/ocae037](https://doi.org/10.1093/jamia/ocae037) ## Technical Specifications for Taiyi-LLM **Environment requirements**: - Python dependencies: `torch==1.13.0`, `transformers==4.30.2`, `accelerate==0.21.0` - Recommended hardware: **NVIDIA 4090 GPU** - Inference parameters: `max_new_tokens=500`, `top_p=0.9`, `temperature=0.3` ## Key Differentiators of Taiyi-LLM **Distinctive features**: 1. **Bilingual capability**: Optimized for Chinese-English biomedical tasks 2. **Extensive training data**: 38 Chinese datasets (10 BioNLP tasks) + 102 English datasets (12 tasks) 3. **Instruction fine-tuning**: 1M+ instruction samples covering diverse biomedical tasks 4. **Open-source resources**: Releases model weights, datasets (CC BY-NC-SA 4.0), and deployment scripts ### Citation sources: - [Taiyi-LLM](https://github.com/DUTIR-BioNLP/Taiyi-LLM) - Official URL Updated: 2025-04-01