Efficient Model Fine-Tuning with LoRA for Biomedical Named Entity Recognition

This paper provides an in-depth discussion of the biomedical named entity recognition (NER) task and focuses on a task-specific fine-tuning method using large language models (LLMs) through parameter-efficient fine-tuning (PEFT) technology, particularly low-rank adaptive technology (LoRA). The BioCreative II Gene Recognition (BC2GM) dataset was used in the experiment, with Llama 3.2 and Qwen 2.5 chosen as the base models. The model was fine-tuned by carefully configuring the LoRA parameters. The results demonstrate that this method can significantly enhance the model's performance in biomedical NER tasks, verifying the effectiveness and practicality of LoRA technology in such tasks and highlighting the vast potential of large-scale language models in complex, specialized domains. Our experiment source code is available on GitHub at: https://github.com/daojiaxu/LLM_NER_LORA.