ThaiLLM/ThaiLLM-8B-SFT-IQ

ThaiLLM-8B-SFT-IQ (Medical) Overview

ThaiLLM-8B-SFT-IQ (Medical) is an 8 billion parameter, Thai-language large language model developed by ThaiLLM, specifically fine-tuned for medical information querying and citation-grounded question answering. Built upon the ThaiLLM-8B-SFT base model, it is optimized for Retrieval-Augmented Generation (RAG) systems, requiring answers to be generated exclusively from provided medical contexts with explicit citations.

Key Capabilities & Features

• Medical Specialization: Fine-tuned with medical-domain data for enhanced relevance and accuracy in Thai medical contexts.
• Citation-Grounded QA: Designed to provide answers strictly based on supplied medical contexts, including fact IDs for citations.
• RAG Optimization: Ideal for integration into RAG workflows where external medical documents are used to ground responses.
• Performance: Achieves a Jaccard score of 0.5485 for citations and a BLEU score of 0.4363 in medical information query settings, outperforming its base model and Qwen3-8B-Bas.
• Strict JSON Output: Optimized for structured JSON responses, facilitating programmatic integration.

Intended Use Cases

• Medical RAG Systems (Thai): Building systems that retrieve and generate answers from Thai medical documents.
• Medical Document Question Answering: Answering specific questions based on provided medical texts.
• Citation-Grounded Medical QA: Ensuring generated answers are verifiable and traceable to source documents.
• Medical Education and Evaluation: Assisting in educational tools or evaluating understanding of medical information.

Note: This model is not intended for medical diagnosis, treatment decisions, patient-facing clinical advice, or emergency use.