Overview

IL-7B (Immuno-LLM 7 Billion) is a 7 billion parameter language model, a merge of two models derived from Mistral-7B and BioMistral-7B. Developed by cmcmaster, this model is uniquely specialized for the domain of clinical rheumatology and immunology. The merging process utilized mergekit with the SLERP method, combining two unreleased models: rheumistral-sft (trained from original Mistral-7B) and biorheumistral-sft (trained from BioMistral-7B).

Key Capabilities

• Specialized Domain Knowledge: Highly proficient in generating text related to clinical rheumatology and immunology.
• Fine-tuned for Medical Q&A: Trained on curated datasets of rheumatology and immunology texts, including supervised fine-tuning on synthetic and human-generated QA pairs and chat logs.
• Mistral-7B Architecture: Benefits from the robust base architecture of Mistral-7B, providing a strong foundation for its specialized knowledge.

Intended Use

IL-7B is designed as an AI tool for research and general interest in rheumatology and autoimmune diseases. It is important to note that it has not been validated for and should not be used for direct clinical decision-making. The model uses the same prompt format as Zephyr from Hugging Face, facilitating easy integration and use with existing pipelines.