RecurvAI/Recurv-Medical-Deepseek-R1

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jan 29, 2025License:mitArchitecture:Transformer0.0K Open Weights Cold

RecurvAI/Recurv-Medical-Deepseek-R1 is an 8 billion parameter instruction-tuned language model based on DeepSeek R1 Distill Llama, specifically optimized for medical applications. Developed by RecurvAI, it excels at answering medical questions, aiding in patient history gathering, and generating comprehensive, context-specific explanations for healthcare professionals and researchers. The model was fine-tuned using 67,299 high-quality medical Q&A pairs from sources like PubMed and clinical guidelines, with a knowledge cut-off date of January 22, 2025.

Loading preview...

Recurv-Medical-Deepseek-R1: Medical Language Model

RecurvAI's Recurv-Medical-Deepseek-R1 is an 8 billion parameter language model, an enhanced version of DeepSeek R1 Distill Llama, specifically engineered for the healthcare domain. It is designed to provide accurate and context-specific support for medical professionals and researchers.

Key Capabilities

  • Medical Query Optimization: Optimized for medical-specific queries across various specialties.
  • Clinical Workflow Support: Fine-tuned to assist with clinical and research-oriented workflows.
  • Multi-turn Conversations: Supports multi-turn interactions for context-rich dialogues.
  • Evidence-Based Answers: Generates comprehensive answers and evidence-based suggestions.
  • Patient History Aid: Effective in aiding patient history gathering, including anamnesis workflows using EHR-simulated data.

Training and Architecture

This model was fine-tuned using a lightweight parameter-efficient approach with safetensors. The training dataset comprised 67,299 high-quality Q&A pairs curated from diverse medical sources, including:

  • PubMed: Insights from open-access medical research.
  • Clinical Guidelines: Data from WHO, CDC, and specialty-specific guidelines.
  • EHR-Simulated Data: Synthetic datasets for patient record modeling.

The model has a context length of 4,096 tokens and was trained for 100,000 steps. Its knowledge cut-off date is January 22, 2025.

Licensing

Recurv-Medical-Deepseek-R1 is released under the MIT License.