Hugging Face Overview
Overview
The microsoft/MediPhi-MedCode model is a 3.8 billion parameter small language model (SLM) developed by Microsoft Healthcare & Life Sciences. It is part of the MediPhi Model Collection, which specializes in medical and clinical domains. This particular model is derived from Phi-3.5-mini-instruct and has been fine-tuned specifically for medical coding using the SLERP merging method, integrating an expert trained on various medical coding datasets (ICD10CM, ICD10PROC, ICD9CM, ICD9PROC, and ATC).
Key Capabilities
- Medical Domain Specialization: Optimized for clinical natural language processing tasks, particularly medical coding.
- Modular Design: Part of a collection of expert models, allowing for focused domain adaptation while retaining general abilities.
- Performance: Demonstrates strong performance on medical benchmarks, achieving 68.7% on ICD10CM in the CLUE+ benchmark, significantly outperforming its base model.
- Efficiency: Designed for research in memory/compute constrained and latency-bound environments.
- Safety: Retains the safety capabilities of its base model, Phi-3.5-mini-instruct, with demonstrated conservation of safety behaviors against jailbreaking and harmfulness.
Good For
- Clinical NLP Research: Ideal for researchers working on language models in medical and clinical scenarios.
- Medical Coding: Specifically strong in tasks related to medical coding, as indicated by its training data and benchmark results.
- Resource-Constrained Deployments: Suitable for environments with limited computational resources or strict latency requirements.
- Benchmarking: Intended for use in benchmarking contexts or with expert user verification of outputs due to its research-oriented nature.