PaddleOCR-VL-1.6 is a compact vision-language model specialized for document parsing and visual document understanding. It is designed to recognize and interpret text, tables, formulas, charts, seals, and other structured elements contained within document images.
This repository contains GGUF quantized variants of PaddleOCR-VL-1.6 intended for efficient local inference with compatible GGUF runtimes. The compact 0.9B-parameter design makes the model suitable for resource-conscious document-processing deployments while retaining strong multimodal document understanding capabilities.
PaddleOCR-VL-1.6 introduces region-aware data optimization and progressive post-training improvements over PaddleOCR-VL-1.5. The model is particularly focused on improving difficult document elements such as complex tables, ancient Chinese documents, rare Chinese characters, seals, text spotting, and charts.
This repository provides GGUF quantized versions of PaddleOCR-VL-1.6 for memory-efficient local inference. The available formats provide different trade-offs between model size, computational requirements, and retention of the original model's visual-document understanding behavior.
PaddleOCR-VL-1.6 builds upon the PaddleOCR-VL-1.5 architecture while concentrating development effort on improving data quality, difficult document regions, and post-training optimization.
Optical Character Recognition Extracts textual content from document images and other visual inputs, including difficult characters and complex text regions.
Document Parsing Converts visual document content into structured representations while considering the relationships between different document elements.
Table Recognition Identifies and interprets structured tables, including their textual content and organization.
Formula Recognition Processes mathematical formulas appearing within document images.
Chart Understanding Recognizes and interprets information contained in charts and other graphical document elements.
Text Spotting Supports recognition of text together with its spatial location within visual documents.
Seal and Stamp Recognition Provides specialized recognition capabilities for seals and stamps appearing in documents.
Complex Document Understanding Handles challenging document categories including ancient Chinese documents, rare Chinese characters, and visually complex pages.
Multilingual Processing Supports multilingual document understanding and recognition workflows.
./llama-cli \
-m SandLogicTechnologies/MiniCPM5-2B_IQ4_NL.gguf \
-mmproj SandLogicTechnologies/mmproj-paddleocr-VL-1.6b-f16.gguf
Document Digitization Convert scanned pages and document images into machine-readable text and structured representations.
OCR and Text Extraction Extract textual information from documents containing standard or challenging visual layouts.
Table and Form Processing Recognize structured tables and extract information from document forms.
Formula and Chart Processing Analyze mathematical formulas, charts, and other non-standard document elements.
Enterprise Document Intelligence Build document-processing pipelines for invoices, reports, forms, archives, and other business documents.
Document Research and Evaluation Evaluate compact vision-language models on OCR, document parsing, element recognition, and multimodal understanding tasks.
These GGUF quantized models are based on the original PaddleOCR-VL-1.6 model developed by the PaddlePaddle team.
Special thanks to:
The PaddlePaddle and PaddleOCR teams for developing and releasing PaddleOCR-VL-1.6 and its document-intelligence ecosystem.
Georgi Gerganov and the llama.cpp open-source community for enabling efficient local inference through the GGUF ecosystem.
For any inquiries or support, please contact us at support@sandlogic.com or visit our Website.