docling-ibm-models

What it is and what it does

This package provides pre-trained neural networks for understanding document structure, with two main components: TableFormer, which identifies table boundaries and cell structure in document images, and a Layout model that detects tables and other page elements. It wraps transformer-based models trained on datasets including PubTabNet, FinTabNet, and TableBank, designed to integrate with document conversion pipelines.

The package depends on torch, transformers, and huggingface_hub to load and run inference. You install it with an optional choice between opencv-python and opencv-python-headless, then instantiate predictors and call them on image data. Models are downloaded from huggingface_hub on first use.

Use it for:

• Extract table structure and cell locations from scanned PDF pages for downstream OCR or data extraction.
• Detect and segment page layout regions to guide document parsing and conversion.
• Batch process document images to identify which pages contain tables before specialized recognition.
• Build a document conversion pipeline that preserves table formatting when converting PDFs to structured formats.
• Analyze financial or scientific documents where accurate table extraction is critical for data integrity.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Provides AI models for table structure recognition and page layout detection to support PDF document conversion.

Yes, if you are building a document processing pipeline that needs table and layout detection. The package is actively maintained, has no known vulnerabilities, and integrates with the broader ecosystem. The MIT license poses no barrier. Install friction is low, though torch and transformers are substantial dependencies—acceptable for ML workloads but not for lightweight applications. Suitable for production use.

Install

docling-ibm-models on PyPI

pip install docling-ibm-models

uv add docling-ibm-models

poetry add docling-ibm-models

Installing docling-ibm-models

Before you install

Low friction install with a pure-Python wheel. Active maintenance with a release 3 days old. Requires 13 runtime dependencies including torch, transformers, and huggingface_hub, which are substantial but standard for ML inference workloads.

License in practice

MIT license permits commercial and private use with minimal restrictions—suitable for most projects.

Quickstart

pip install "docling-ibm-models[opencv-python-headless]"

from docling_ibm_models import LayoutPredictor
predictor = LayoutPredictor.from_pretrained()
result = predictor.predict(image)

Requires Python 3.10 or later. torch and torchvision are heavy dependencies; accelerate is optional but recommended for inference speed.

Verify before relying

• Whether pre-trained model weights are automatically downloaded or require manual setup beyond huggingface_hub.
• Performance characteristics (inference latency, memory footprint) for typical document sizes.
• Whether CPU-only inference is supported or CUDA is required.

Package facts

Evidence: docling_ibm_models-3.14.0-py3-none-any.whl

Keywords: docling, convert, document, pdf, layout model, segmentation, table structure, table former

Tags

More Artificial Intelligence packages

Further reading