Reproduce “FoodSky: A Food-oriented Large Language Model that Passes the Chef and Dietetic Examination” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Pengfei Zhou, Weiqing Min, Chaoran Fu, Ying Jin, Mingyu Huang, Xiangyang Li, Shuhuan Mei, Shuqiang Jiang
FoodSky introduces the first Chinese food-oriented Large Language Model, achieving expert-level performance in culinary and dietetic knowledge. The model passed the Chinese National Chef Exam with 67.2% accuracy and the National Dietetic Exam with 66.4% accuracy, leveraging a comprehensive Chinese food corpus and novel architectural components for fine-grained understanding and knowledge retrieval.
Key insights
- Domain-specific LLMs can achieve expert-level performance in complex, knowledge-intensive fields when trained on carefully curated data and with tailored architectures.
- High-quality, culturally specific instruction data is crucial for specialized LLM performance, with quality outweighing quantity in the food domain.
- Integrating architectural innovations like the Topic-based Selective State Space Model (TS3M) and Hierarchical Topic Retrieval Augmented Generation (HTRAG) significantly enhances a base LLM's understanding and knowledge integration in a specialized domain.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.