kenyano/Llama3-ELAINE-medLLM-instruct-8B_v0.1
The kenyano/Llama3-ELAINE-medLLM-instruct-8B_v0.1 is an 8 billion parameter Llama 3-based trilingual (English, Japanese, Chinese) large language model adapted for the biomedical domain, with an 8192 token context length. Developed by Ken Yano et al., it is fine-tuned for conversational QA in medical contexts and exhibits superior trilingual capabilities compared to existing bilingual or multilingual medical LLMs. This model is optimized for biomedical question answering across these three languages, balancing trilingual performance with the base model's general knowledge.
Loading preview...
ELAINE-medLLM-instruct-8B_v0.1: Trilingual Biomedical LLM
ELAINE (EngLish-jApanese-chINesE)-medLLM-instruct-8B_v0.1 is an 8 billion parameter model built upon Llama 3, specifically adapted for the biomedical domain. Its primary differentiator is its trilingual capability across English, Japanese, and Chinese, making it suitable for medical applications requiring multilingual support.
Key Capabilities & Training
- Trilingual Biomedical Expertise: Trained on a diverse and voluminous dataset of biomedical text in English, Japanese, and Chinese, including scientific papers, medical guidelines, web text, textbooks, and PubMed abstracts.
- Two-Stage Training: Underwent continued pre-training on domain-specific data followed by supervised fine-tuning (SFT) using conversational QA datasets like Medical Meadow, HealthCareMagic, iClilic, and augmented HuatuoGPT-2.
- Improved Japanese Performance: This
v0.1version is a bug-fixed iteration that specifically enhances Japanese language performance, though with slight trade-offs in English and Chinese benchmarks compared to its predecessor.
Performance Highlights
The model demonstrates competitive performance on various biomedical benchmarks:
- English: Evaluated on MedQA, MedMCQA, MMLU, and PubMedQA.
- Japanese: Evaluated on DenQA, IgakuQA, and JJSIMQA, showing strong results in this language.
- Chinese: Evaluated on CMExam, MedQA, and MedQA-4op.
Use Cases
This model is ideal for applications requiring multilingual biomedical question answering and conversational AI in healthcare settings, particularly where English, Japanese, and Chinese language support is critical. It is designed to act as an AI Health Assistant capable of responding to medical queries in these languages.