Lift is a vision-language document parsing model developed by DataLab, designed for document understanding, structured document reconstruction, layout analysis, and intelligent content extraction. This repository contains GGUF quantized variants of the model optimized for efficient local inference using llama.cpp.

Unlike traditional OCR models that primarily focus on recognizing text, Lift emphasizes reconstructing complete document structure while preserving reading order, hierarchy, tables, figures, mathematical expressions, forms, and semantic relationships between document elements. The model enables complex documents to be transformed into structured machine-readable representations suitable for downstream AI workflows.

The quantized formats significantly reduce memory requirements while maintaining strong document parsing capability, making the model suitable for enterprise document intelligence, retrieval systems, and large-scale document conversion pipelines.

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

bash

./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/

SandLogicTechnologies/lift-GGUF

Lift