Skip to main content

docling-pp-ocrv6

A Docling OCR plugin for PaddlePaddle's PP-OCRv6 models.

This plugin seamlessly integrates with Docling's standard pipeline to provide local OCR capabilities using PaddlePaddle's PP-OCRv6 models. It runs the PP-OCRv6 detection and recognition ONNX checkpoints locally through RapidOCR (onnxruntime), so OCR happens inside the docling worker — no external service required.


GitHub  |  PyPI


PyPI version Python versions License CI Ruff Coverage

GPU acceleration is automatic when the docling accelerator device resolves to CUDA and onnxruntime-gpu is installed.

Installation

Pick exactly one onnxruntime extra — installing both the CPU and GPU wheels at once is unsupported and prevents the CUDA provider from registering:

pip install "docling-pp-ocrv6[cpu]"     # CPU (onnxruntime)
pip install "docling-pp-ocrv6[gpu]"     # CUDA (onnxruntime-gpu)

The detection and recognition ONNX models are downloaded from HuggingFace on first use and cached under docling's model cache. To pre-fetch them (e.g. at container build time):

from docling_pp_ocrv6 import PPOCRv6Model

PPOCRv6Model.download_models()

Usage

from docling.document_converter import DocumentConverter, PdfFormatOption
from docling.datamodel.base_models import InputFormat
from docling.datamodel.pipeline_options import PdfPipelineOptions
from docling_pp_ocrv6 import PPOCRv6Options

pipeline_options = PdfPipelineOptions(do_ocr=True)
pipeline_options.ocr_options = PPOCRv6Options()

converter = DocumentConverter(format_options={InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options)})
result = converter.convert("scanned.pdf")
print(result.document.export_to_markdown())

With docling-serve, request the engine by its kind:

{ "options": { "ocr": true, "ocr_engine": "pp-ocrv6" } }

(DOCLING_SERVE_ALLOW_EXTERNAL_PLUGINS=true must be set for the plugin to load.)

Configuration

All options are settable via PPOCRv6Options(...) or environment variables:

Option Env var Default
lang PPOCRV6_LANG de,en,fr,it,es,nl,pt,... (German-led European set)
text_score PPOCRV6_TEXT_SCORE 0.5
use_det / use_cls / use_rec PPOCRV6_USE_DET / _CLS / _REC true
det_repo PPOCRV6_DET_REPO PaddlePaddle/PP-OCRv6_medium_det_onnx
rec_repo PPOCRV6_REC_REPO PaddlePaddle/PP-OCRv6_medium_rec_onnx
det_model_path / rec_model_path / rec_keys_path / cls_model_path PPOCRV6_*_MODEL_PATH / _KEYS_PATH auto

The recognition character dictionary is extracted automatically from the recognition model's inference.yml. Angle classification uses RapidOCR's bundled cls model unless cls_model_path is set.

Fast RapidOCR engine (rapidocr-fast)

The plugin also registers rapidocr-fast: docling's own RapidOCR engine with the same text and a faster page loop. It takes every option of docling's rapidocr kind and changes only how a page batch is run:

  • each page is rendered once at the OCR scale and every OCR rectangle is cropped from it (docling renders the whole page once per rectangle);
  • with RapidOCR's PyTorch engine (backend: torch), all rectangles of a page batch go through one pass: image normalisation, padding and the CTC argmax on the GPU, and the detection, angle and recognition forwards of all rectangles queued back to back before their results are read. The batches stay RapidOCR's own: on the GPU a convolution's result can depend on the batch size, so other batches would change the text. RapidOCR's own pre- and post-processing does the rest. Other backends, right-to-left languages, word boxes and rapidocr releases other than 3.10 call RapidOCR per image as docling does.

For docling-serve, as an OCR preset:

DOCLING_SERVE_CUSTOM_OCR_PRESETS='{"rapidocr": {"kind": "rapidocr-fast", "backend": "torch"}}'

Written against rapidocr 3.10 (checked at runtime) and docling-parse 7.x.

Development

make install   # uv sync + pre-commit
make check     # ruff lint + format check + ty type check
make test      # pytest with coverage

License

MIT

Metadata

Release files for docling-pp-ocrv6 0.3.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for docling-pp-ocrv6 0.3.0
File Size Uploaded Provenance Actions
docling_pp_ocrv6-0.3.0.tar.gz 224.8 kB Without provenance Details for docling_pp_ocrv6-0.3.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for docling-pp-ocrv6 0.3.0
File Interpreter ABI Platform Actions
docling_pp_ocrv6-0.3.0-py3-none-any.whl Python 3 none any Details for docling_pp_ocrv6-0.3.0-py3-none-any.whl

Total release size: 258.5 kB

Release files / docling_pp_ocrv6-0.3.0.tar.gz

Download URL docling_pp_ocrv6-0.3.0.tar.gz
Size 224.8 kB
Tags Source
SHA-256 checksum
How to use checksums
8184a625f8fc12c4e36a9caa0ad6e16f725173842c38687691eedc2a41e7848f
BLAKE2b-256 checksum
How to use checksums
1d50de5616e8dfc714912a0b06468614c8409967368de67549ab0752802f57fb
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via uv/0.13.7 {"installer":{"name":"uv","version":"0.13.7","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"26.04","id":"resolute","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

Release files / docling_pp_ocrv6-0.3.0-py3-none-any.whl

Download URL docling_pp_ocrv6-0.3.0-py3-none-any.whl
Size 33.8 kB
Tags Python 3
SHA-256 checksum
How to use checksums
e46866b5ccfa1e62c92c20a15fa1e456cb07ee508fa7c7f4b8abce8c7ef03b49
BLAKE2b-256 checksum
How to use checksums
3b6e9f74dd49ee0b6dc03e8da278598f04cffa472e123cfc6a081977def5df39
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via uv/0.13.7 {"installer":{"name":"uv","version":"0.13.7","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"26.04","id":"resolute","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

Security for docling-pp-ocrv6 0.3.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page