nougat-ocr-cli 0.1.0

Simple CLI wrapper for Nougat OCR with GPU acceleration support

Nougat OCR CLI

Simple, batteries-included CLI wrapper for Nougat OCR with GPU acceleration.

Features

• GPU acceleration (CUDA & Apple Metal)
• Simple CLI interface
• Batch processing support
• Clean Markdown output
• Automatic model downloading
• Python API with type hints

Installation

From PyPI

pip install nougat-ocr-cli

From GitHub

pip install git+https://github.com/rubenffuertes/nougat-ocr-cli.git

From source

git clone https://github.com/rubenffuertes/nougat-ocr-cli.git
cd nougat-ocr-cli
uv pip install -e .

CLI Usage

# Basic usage - outputs to current directory
nougat-ocr-cli document.pdf

# Specify output directory
nougat-ocr-cli document.pdf -o output/

# Process specific pages (zero-indexed)
nougat-ocr-cli document.pdf --pages 0-5
nougat-ocr-cli document.pdf --pages 1,3,5,7

# Use smaller model for faster processing
nougat-ocr-cli document.pdf --model 0.1.0-small

# Use full precision (FP32) for better accuracy
nougat-ocr-cli document.pdf --full-precision

# Set batch size manually
nougat-ocr-cli document.pdf --batch-size 4

CLI Options

Option	Description
input	Input PDF file to process
-o, --output	Output directory (default: current directory)
--model	Model version (default: 0.1.0-base)
--batch-size	Batch size for processing (auto-detected)
--full-precision	Use FP32 instead of BF16
--no-markdown	Disable markdown post-processing
--pages	Page range (e.g., '0-5' or '1,3,5')

Python API

from nougat_wrapper import NougatOCR
from pathlib import Path

# Initialize (loads model to GPU automatically)
ocr = NougatOCR()

# Extract text from PDF
result = ocr.extract_text(Path("paper.pdf"))

print(f"Extracted {result.pages} pages")
print(f"Failed pages: {result.placeholder_pages}")
print(result.text)  # Markdown output

Advanced Usage

ocr = NougatOCR(
    model_tag="0.1.0-small",  # Use smaller model
    batch_size=4,              # Process 4 pages at once
    full_precision=True,       # Use FP32 instead of BF16
)

# Only OCR pages 0, 1, 2 (zero-indexed)
result = ocr.extract_text(pdf_path, pages=[0, 1, 2])

Requirements

• Python 3.11+
• GPU recommended (CUDA or Apple Metal)
• ~1.3 GB for model weights (auto-downloaded)

License

MIT