Lift GGUF

Model Overview

• Model Name: Lift
• Base Model: datalab-to/lift
• Architecture: Vision-Language Document Parsing Model
• Parameter Count: Approximately 8 Billion Parameters
• Modalities: Text, Image
• Primary Languages: Multilingual
• Developer: DataLab
• License: Apache 2.0

Quantization Formats

This repository provides various GGUF quantized versions of the Lift model optimized for efficient local inference using llama.cpp.

IQ3_M

• Size reduction of approx 77.12% (4.21 GB) compared to 16-bit (18.40 GB)
• Aggressive 3-bit quantization optimized for memory-efficient document parsing workloads
• Suitable for large-scale document conversion pipelines operating on resource-constrained hardware
• Enables efficient processing of books, reports, forms, technical papers, and visually rich documents
• Preservation of fine document hierarchy, layout relationships, and complex structural elements may reduce compared to higher-precision variants

IQ4_NL

• Size reduction of approx 71.74% (5.20 GB) compared to 16-bit (18.40 GB)
• Advanced 4-bit non-linear quantization designed to preserve document structure and parsing fidelity
• Better suited for layout-aware document conversion, semantic reconstruction, and structured content extraction
• Provides improved consistency when interpreting complex page layouts and document organization
• May require slightly increased computational overhead during inference

IQ4_XS

• Size reduction of approx 72.88% (4.99 GB) compared to 16-bit (18.40 GB)
• Balanced 4-bit quantization optimized for efficient document parsing and reliable structural reconstruction
• Provides a practical balance between memory efficiency, document understanding quality, and inference speed
• Suitable for enterprise document processing, PDF conversion, document indexing, and knowledge-base preparation
• Maintains dependable performance across diverse document layouts and formatting styles

Training Background (Original Model)

Lift is trained with an emphasis on document parsing, layout understanding, semantic reconstruction, and multimodal document intelligence across diverse document collections.

Pretraining

• Large-scale multimodal pretraining using document-centric image and text datasets
• Focus on visual-text alignment, layout representation learning, and document structure modeling
• Optimized for downstream document conversion and structured document understanding

Document Parsing Optimization

• Enhanced for preserving reading order, semantic hierarchy, and document structure
• Optimized for complex layouts containing tables, figures, forms, mathematical expressions, and mixed-content pages
• Improved consistency for document reconstruction and structured representation generation

Key Capabilities

• Document Parsing : Reconstructs complete document structure while preserving semantic organization.

• Layout Understanding : Identifies relationships between headings, paragraphs, tables, figures, and other document elements.

• Structured Document Conversion : Converts complex documents into machine-readable structured representations.

• Semantic Content Extraction : Extracts contextual information while maintaining document hierarchy.

• Complex Layout Processing : Handles technical documents, reports, books, forms, and visually rich page layouts.

• Efficient Local Deployment : Quantized variants enable practical document intelligence workloads on consumer hardware.

Usage Example

Using llama.cpp

./llama-mtmd-cli \
  -m SandlogicTechnologies/Lift_IQ4_NL.gguf \
  --mmproj SandlogicTechnologies/lift.mmproj-f16.gguf \
  --image technical_report.pdf \
  -p "Convert this document into structured Markdown while preserving headings, tables, and figures."

Recommended Usecases

• Document Conversion : Transform complex documents into structured machine-readable formats.

• Layout-Aware Processing : Preserve document hierarchy, formatting, and reading order during extraction.

• Knowledge Base Generation : Prepare structured documents for enterprise search and retrieval systems.

• Technical Document Processing : Parse scientific papers, manuals, books, reports, and technical documentation.

• Enterprise Document Intelligence : Build automated document understanding and indexing workflows.

• RAG Data Preparation : Generate structured representations suitable for Retrieval-Augmented Generation systems.

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 Lift 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 support@sandlogic.com or visit https://www.sandlogic.com/

lift-IQ4_XS.gguf

gguf · 4,99 GB · SHA-256 043758a07f12…6dc0 · Hugging Face