syntheticbot/ocr-qwen

VISIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:32kPublished:Feb 17, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

syntheticbot/ocr-qwen is a 7 billion parameter vision-language model fine-tuned from Qwen/Qwen2.5-VL-7B-Instruct, specifically optimized for Optical Character Recognition (OCR) tasks. It excels at extracting text from diverse images, including documents and scene text, with enhanced accuracy across various fonts, styles, and orientations. The model provides robust handling of document variations, offers structured output generation (JSON, CSV) for recognized text and layout, and includes text localization capabilities, making it ideal for document processing and data extraction from visual content.

Loading preview...

syntheticbot/ocr-qwen: Specialized OCR Vision-Language Model

syntheticbot/ocr-qwen is a 7 billion parameter vision-language model, fine-tuned from the robust Qwen/Qwen2.5-VL-7B-Instruct base, with a 32K context length. This model is specifically engineered for high-accuracy Optical Character Recognition (OCR) across a wide range of visual inputs, from structured documents to complex scene text.

Key Capabilities

  • Enhanced Text Recognition: Achieves superior accuracy in extracting text, adapting to diverse fonts, styles, sizes, and orientations.
  • Robust Document Handling: Designed to manage complexities like varied layouts, noise, and distortions commonly found in documents.
  • Structured Output: Capable of generating recognized text and layout information in structured formats such as JSON or CSV, particularly useful for invoices and tables.
  • Text Localization: Provides precise bounding box information for text elements within images.
  • Improved Visual Text Analysis: Maintains proficiency in analyzing charts and graphics, with enhanced recognition of embedded text.

Good for

  • Document Processing: Automating data extraction from scanned documents, PDFs, and images.
  • Invoice and Table Extraction: Converting visual tables and invoices into structured data formats.
  • Scene Text Recognition: Identifying and extracting text from real-world images and environments.
  • Automated Data Entry: Reducing manual effort in transcribing text from visual sources.
  • Content Analysis: Extracting textual information from charts, graphs, and other visual media for analytical purposes.

Finetunes
1 models