Technoculture/MT7Bi-sft: A Medical Domain LLM

Technoculture/MT7Bi-sft is a 7 billion parameter instruction-tuned language model, built by merging the Technoculture/MT7Bi-alpha adapter with its Meditron 7B base model. This model is specifically designed and fine-tuned for applications within the medical and health domains.

Key Capabilities & Performance

The model demonstrates enhanced performance on various medical benchmarks compared to its Meditron 7B base. While its general reasoning scores on the Open LLM Leaderboard (ARC: 54.1, HellaSwag: 75.11, TruthfulQA: 43.08, Winogrande: 72.14) are competitive for its size, its strength lies in specialized medical tasks. For instance, on MMLU-Medical, it scores 46.9, and on PubMedQA, it achieves 65.2. These scores indicate a targeted improvement in understanding and generating content relevant to healthcare.

When to Use This Model

• Medical Question Answering: Ideal for applications requiring accurate responses to health-related queries.
• Healthcare Information Retrieval: Suitable for tasks involving the extraction or summarization of medical data.
• Specialized Domain Applications: Best utilized in scenarios where domain-specific knowledge in health and medicine is crucial, rather than general-purpose tasks.