Thai Medical Large Language Model: Llama-3.1-EIRAI-8B-instruct
EIRTHAIMED/Llama-3.1-EIRAI-8B-instruct is an 8-billion parameter model developed by EIRAI-Thaimedical, specifically tailored for Thai medical applications. It possesses expertise in both Thai medical language and English medical terminology, making it a unique resource for medical NLP in the region.
Key Capabilities
- Bilingual Medical Expertise: Proficient in both Thai and English medical terminology.
- Strong Benchmark Performance: Demonstrated capabilities across key medical benchmarks including MMLU, MedQA, PubMedQA, and MedMCQA.
- Thai Language Assessment: Evaluated on Thai-specific assessments such as ThaiExam, M3Exam, XNLI, and XCOPA.
- Clinically Adapted Thai Testing: Features a custom "Clinically Adapted Model Enhanced test" in Thai, designed to support clinical use in hospitals and improve Thai medical Retrieval-Augmented Generation (RAG).
- Superior Translation: Achieves a BLEU score of 61.10 for translation, significantly outperforming other models like Meta Llama 3.1-8B Instruct (35.74) and Typhoon-v1.5x-8B-Instruct (34.42).
- High Clinical Task Performance: Outperforms GPT-3.5, Typhoon-v1.5x-8B-instruct, and often GPT-4o on various Clinically Adapted Thai Medical Tasks, including Named Entity Recognition, Temporal Information Extraction, and Question Answering, with an average score of 7.11.
Good for
- Medical Research: Exploring the potential of LLMs in medical contexts, particularly for Thai healthcare.
- Improving Thai Medical RAG: Enhancing Retrieval-Augmented Generation systems for Thai medical information.
- Bilingual Medical Text Processing: Tasks requiring understanding and generation of medical text in both Thai and English.
- Benchmarking and Evaluation: Researchers interested in evaluating medical LLMs on diverse datasets, including specialized Thai medical tasks.
Important Notice: While designed with high-quality medical knowledge, this model is currently in the research phase and not optimized for safe, practical use in real-world medical settings. It should not be used for clinical decision-making without further validation.