chandra-ocr-2-GGUF

Chandra-OCR-2

Chandra-OCR-2 is a multimodal OCR-focused vision-language model designed for document understanding, text extraction, layout analysis, and structured information retrieval from visual documents. This repository contains GGUF quantized variants of the model optimized for efficient local inference using llama.cpp.

The model is specifically optimized for extracting and understanding textual content from documents, scanned pages, forms, tables, receipts, invoices, reports, and other document-centric visual inputs. Unlike general-purpose multimodal assistants, Chandra-OCR-2 focuses on high-quality optical character recognition and document intelligence workflows.

The quantized formats significantly reduce memory requirements while preserving OCR accuracy and document understanding capability, enabling practical deployment on consumer hardware and edge environments.

---

Model Overview

• Model Name: Chandra-OCR-2
• Base Model: datalab-to/chandra-ocr-2
• Architecture: Vision-Language Model
• Parameter Count: Approximately 4 Billion Parameters
• Modalities: Text, Image
• Languages: Multilingual
• Developer: DataLab
• License: Apache 2.0

---

Quantization Formats

This repository provides various GGUF quantized versions of the Chandra-OCR-2 model optimized for efficient local inference using llama.cpp. Below are the details of the available quantization formats.

IQ3_M

• Size reduction of approx 74.92% (2.32 GB) compared to 16-bit (9.25 GB)
• Aggressive 3-bit quantization optimized for maximum memory efficiency
• Suitable for lightweight OCR pipelines and low-memory deployment environments
• Enables practical execution of document-understanding workloads on consumer hardware
• Fine-grained text extraction accuracy and complex document interpretation may experience reduced fidelity compared to higher-precision variants

IQ4_NL

• Size reduction of approx 69.30% (2.84 GB) compared to 16-bit (9.25 GB)
• Advanced 4-bit non-linear quantization designed to better preserve OCR quality and document understanding performance
• Better suited for document extraction, structured information retrieval, and layout-aware analysis workflows
• Designed to reduce quantization loss compared to more aggressive formats
• May require slightly increased computational overhead during inference

IQ4_XS

• Size reduction of approx 70.49% (2.73 GB) compared to 16-bit (9.25 GB)
• Balanced 4-bit quantization focused on efficient inference and dependable OCR performance
• Provides a practical balance between memory efficiency, extraction quality, and runtime speed
• Suitable for invoices, forms, receipts, reports, and general document-processing workflows
• Maintains stable performance across most practical OCR and document-understanding tasks

Q6_K

• Size reduction of approx 58.92% (3.80 GB) compared to 16-bit (9.25 GB)
• Higher-precision 6-bit K-Quant format designed to preserve OCR fidelity and document interpretation quality
• Better suited for demanding workloads involving dense documents, complex layouts, multilingual content, and structured extraction tasks
• Provides stronger output consistency and improved retention of the original model capabilities compared to lower-bit formats
• Requires higher memory resources but offers enhanced extraction quality and document-understanding performance

---

Training Background (Original Model)

Chandra-OCR-2 is trained with an emphasis on document intelligence, optical character recognition, layout understanding, and multimodal information extraction across a diverse range of document types.

Pretraining

• Large-scale multimodal pretraining across document-centric image and text datasets
• Focus on text recognition, layout understanding, and document representation learning
• Optimized for downstream OCR and document-analysis workloads

Instruction Tuning

• Refined using document understanding and extraction-oriented datasets
• Enhanced for structured information retrieval and visual-text comprehension
• Improved consistency for OCR, layout reasoning, and document-processing workflows

---

Key Capabilities

• Optical Character Recognition (OCR) Extracts textual information from scanned documents, images, and visual content.

• Document Understanding Interprets document structure, context, and relationships between textual elements.

• Layout Analysis Understands page layouts, sections, tables, forms, and structured document formats.

• Structured Information Extraction Supports extraction of key-value pairs, fields, and document metadata.

• Multilingual Processing Handles multilingual document understanding and text extraction workflows including the Indic Languages.

• Efficient Local Deployment Quantized variants enable practical OCR inference on consumer hardware.

---

Usage Example

Using llama.cpp

./llama-mtmd-cli \
  -m SandlogicTechnologies/Chandra-OCR-2_IQ4_NL.gguf \
  --mmproj SandlogicTechnologies/chandra-ocr-2.mmproj-f16.gguf \
  --image invoice.png \
  -p "Extract all invoice fields and return them in structured JSON format."

---

Recommended Usecases

• Document Digitization Convert scanned documents into searchable and structured text.

• Invoice and Receipt Processing Extract fields, totals, dates, and metadata from financial documents.

• Form Processing Automate extraction of information from structured forms and applications.

• Document Intelligence Systems Build OCR-powered enterprise document-processing workflows.

• Research and Experimentation Evaluate OCR, layout understanding, and multimodal document-analysis techniques.

---

Acknowledgments

These quantized models are based on the original work by the DataLab development team.

Special thanks to:

• The DataLab team for developing and releasing the Chandra-OCR-2 model.

• Georgi Gerganov and the llama.cpp open-source community for enabling efficient quantization and inference via the GGUF format.

---

Contact

For questions, feedback, or support, please reach out at [email protected] or visit https://www.sandlogic.com/