Overview
LEADS-Mistral-7B-v1 is a 7.25 billion parameter model developed by zifeng-ai, fine-tuned from Mistral-7B-Instruct-v0.3. Its primary purpose is to automate and enhance systematic review literature mining, assisting researchers and clinicians in processing biomedical literature more efficiently. The model was trained on the extensive LEADSInstruct dataset, comprising 633,759 instruction data points specifically tailored for systematic review tasks.
Key Capabilities
- Systematic Review Automation: Specialized in tasks like literature search query generation, study eligibility prediction, and detailed data extraction.
- Biomedical Text Optimization: Leverages PubMed and ClinicalTrials.gov data for enhanced performance in domain-specific contexts.
- Multi-Step Support: Handles various stages of systematic review methodology, from initial search to extracting specific study characteristics, trial results, participant statistics, and arm designs.
- Instruction-Tuned Precision: Utilizes structured input-output instructions for high accuracy in task-specific adaptations.
Evaluation & Performance
LEADS-Mistral-7B-v1 was benchmarked against a range of proprietary (GPT-4o, GPT-3.5, Haiku-3) and open-source LLMs (Mistral-7B, Llama-3, BioMistral, MedAlpaca). It demonstrated state-of-the-art performance in literature mining, significantly improving data extraction accuracy and study screening recall, thereby reducing manual effort and time costs in systematic reviews.
Good for
- Automating systematic review processes.
- Extracting structured data from biomedical literature.
- Generating search queries for medical databases.
- Screening citations for study eligibility.
- Researchers and clinicians involved in evidence synthesis.