baichuan-inc/BaichuanMed-OCR-72B

VISIONPricing:Input $3.2 / Output $4Concurrent Unit Cost:4Model Size:72BQuant:FP8Context Size:32kPublished:Apr 2, 2025License:otherArchitecture:Transformer0.0K Featherless Exclusive Cold

BaichuanMed-OCR-72B is a 72 billion parameter vision-language model developed by baichuan-inc, fine-tuned from Qwen2.5-VL-72B-Instruct. It specializes in Optical Character Recognition (OCR) and question answering on medical report images, supporting structured Markdown output. This model is designed for accurate extraction and comprehension of textual content from complex medical reports, achieving 88.6% accuracy on a private medical report dataset.

Loading preview...

BaichuanMed-OCR-72B Overview

BaichuanMed-OCR-72B is a 72 billion parameter vision-language model developed by baichuan-inc, specifically fine-tuned from Qwen2.5-VL-72B-Instruct. This model is engineered for advanced Optical Character Recognition (OCR) and intelligent question answering based on medical report images. It leverages a curated dataset of medical report images with associated questions and answers to enhance its domain-specific performance.

Key Capabilities

  • Robust Medical OCR: Accurately recognizes and extracts complex textual content, including tables and specific symbols, from diverse medical report images.
  • Structured Output: Capable of generating extracted information in a structured Markdown format, facilitating easier data integration and analysis.
  • Accurate Comprehension: Understands user queries and generates logically consistent answers directly derived from the OCR-extracted text, avoiding repetition of watermarks or redundant information.

Performance and Use Cases

Evaluations on a private medical report dataset demonstrate BaichuanMed-OCR-72B's strong performance, achieving an accuracy of 88.6%. This surpasses baseline models like Qwen2.5-VL-7B-Instruct (71.3%) and Qwen2.5-VL-72B-Instruct (83.3%) on similar tasks. This model is particularly well-suited for applications requiring precise information extraction and summarization from medical reports, such as automating data entry, supporting clinical decision-making, or enhancing medical research by digitizing and analyzing large volumes of medical documents.

Finetunes
1 models