olmOCR-7B-0225-preview
| Property | Value |
|---|---|
| Base Model | Qwen2-VL-7B-Instruct |
| License | Apache 2.0 |
| Author | AllenAI |
| Use Case | Document Image Analysis & OCR |
What is olmOCR-7B-0225-preview?
olmOCR-7B-0225-preview is a specialized document analysis model that combines advanced OCR capabilities with visual language understanding. Fine-tuned from Qwen2-VL-7B-Instruct using the olmOCR-mix-0225 dataset, it's designed to process document images efficiently at scale.
Implementation Details
The model processes document images with the longest dimension scaled to 1024 pixels and requires specific metadata formatting for optimal performance. It leverages the olmOCR toolkit for efficient inference via sglang, enabling processing of millions of documents.
- Built on Qwen2-VL-7B-Instruct architecture
- Supports bfloat16 precision for efficient processing
- Includes comprehensive document metadata extraction
- Integrates with the olmOCR toolkit for streamlined deployment
Core Capabilities
- Document image analysis and text extraction
- Language identification and rotation validation
- Table and diagram detection
- Efficient batch processing of large document collections
Frequently Asked Questions
Q: What makes this model unique?
The model combines state-of-the-art visual language understanding with specialized document processing capabilities, all while maintaining efficient processing through the olmOCR toolkit integration.
Q: What are the recommended use cases?
The model is primarily intended for research and educational use in document analysis, particularly when processing large volumes of documents requiring text extraction, layout analysis, and metadata generation.
