{"title":"Base on ChatGLM extraction of medication events in aquaculture with few samples","authors":"Zhenglin Li, Sijia Zhang, Zongshi An","doi":"10.1007/s10499-024-01686-y","DOIUrl":null,"url":null,"abstract":"<div><p>This study proposes a fine-tuning method based on a large language model (LLM), aiming to improve the model’s generalization and accuracy in the task of extracting aquaculture medication. Model’s ability to recognize disease prevention and treatment medication events has been enhanced by constructing an instruction tuning (IT) dataset. ChatGLM4-9B model was further optimized using the rank-stable scaling factor with low-rank fine-tuning (rsLoRA) technique and hyper-parameters were adjusted in real-time using a dynamic parameter fine-tuning strategy. Experimental results show that the RSI-Tune method reaches more than 95% on several evaluation metrics, significantly outperforming traditional LoRA and DoRA methods. This research not only provides technical support for intelligent management of aquaculture but also offers new insights for the development of smart agriculture.</p></div>","PeriodicalId":8122,"journal":{"name":"Aquaculture International","volume":"33 1","pages":""},"PeriodicalIF":2.2000,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Aquaculture International","FirstCategoryId":"97","ListUrlMain":"https://link.springer.com/article/10.1007/s10499-024-01686-y","RegionNum":3,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"FISHERIES","Score":null,"Total":0}
引用次数: 0
Abstract
This study proposes a fine-tuning method based on a large language model (LLM), aiming to improve the model’s generalization and accuracy in the task of extracting aquaculture medication. Model’s ability to recognize disease prevention and treatment medication events has been enhanced by constructing an instruction tuning (IT) dataset. ChatGLM4-9B model was further optimized using the rank-stable scaling factor with low-rank fine-tuning (rsLoRA) technique and hyper-parameters were adjusted in real-time using a dynamic parameter fine-tuning strategy. Experimental results show that the RSI-Tune method reaches more than 95% on several evaluation metrics, significantly outperforming traditional LoRA and DoRA methods. This research not only provides technical support for intelligent management of aquaculture but also offers new insights for the development of smart agriculture.
期刊介绍:
Aquaculture International is an international journal publishing original research papers, short communications, technical notes and review papers on all aspects of aquaculture.
The Journal covers topics such as the biology, physiology, pathology and genetics of cultured fish, crustaceans, molluscs and plants, especially new species; water quality of supply systems, fluctuations in water quality within farms and the environmental impacts of aquacultural operations; nutrition, feeding and stocking practices, especially as they affect the health and growth rates of cultured species; sustainable production techniques; bioengineering studies on the design and management of offshore and land-based systems; the improvement of quality and marketing of farmed products; sociological and societal impacts of aquaculture, and more.
This is the official Journal of the European Aquaculture Society.