PaddleOCRVLDocumentConverter

Extract text from PDF files and images using PaddleOCR's vision-language document parsing API.

Key Features

Extracts text and structured content from PDFs and images using PaddleOCR's VL model.
Supports layout detection, chart recognition, seal recognition, and table extraction.
Auto-detects file type from file extension or MIME type.
Returns both Haystack Document objects and raw PaddleOCR API responses.
Supports optional document orientation classification and image unwarping for scanned documents.
Configurable VLM sampling parameters such as temperature, top_p, and max_new_tokens.

Configuration

Sign in to AI Studio and get an access token.
Deploy a PaddleOCR model endpoint and note the base URL.
Set the PADDLEOCR_ACCESS_TOKEN environment variable. For instructions, see Create Secrets.
Configure PaddleOCRVLDocumentConverter in your pipeline YAML with the base_url of your deployed endpoint.

Connections

PaddleOCRVLDocumentConverter accepts file paths, Path objects, or ByteStream inputs as sources, and optional metadata as meta. It outputs a list of Document objects (documents) containing extracted text as Markdown with page breaks, and a list of raw API responses (raw_paddleocr_responses).

Connect the pipeline's file input to the sources input. Connect the documents output to DocumentSplitter or DocumentWriter for further processing.

Source Code

To check this component's source code, open paddleocr_vl_document_converter.py in the Haystack Core Integrations repository.

Usage Examples

Basic Configuration

  paddleocr_converter:
    type: haystack_integrations.components.converters.paddleocr.paddleocr_vl_document_converter.PaddleOCRVLDocumentConverter
    init_parameters:
      base_url: http://your-paddleocr-endpoint.aistudio-app.com
      access_token:
        type: env_var
        env_vars:
        - PADDLEOCR_ACCESS_TOKEN
        strict: false

Using the Component in a Pipeline

# haystack-pipeline
components:
  paddleocr_converter:
    type: haystack_integrations.components.converters.paddleocr.paddleocr_vl_document_converter.PaddleOCRVLDocumentConverter
    init_parameters:
      base_url: http://your-paddleocr-endpoint.aistudio-app.com
      access_token:
        type: env_var
        env_vars:
        - PADDLEOCR_ACCESS_TOKEN
        strict: false
  splitter:
    type: haystack.components.preprocessors.document_splitter.DocumentSplitter
    init_parameters:
      split_by: word
      split_length: 250
      split_overlap: 30
  writer:
    type: haystack.components.writers.document_writer.DocumentWriter
    init_parameters:
      document_store:
        type: haystack_integrations.document_stores.opensearch.document_store.OpenSearchDocumentStore
        init_parameters:
          hosts:
          - ${OPENSEARCH_HOST}
          index: paddleocr-demo
          embedding_dim: 768
          create_index: true

connections:
- sender: paddleocr_converter.documents
  receiver: splitter.documents
- sender: splitter.documents
  receiver: writer.documents

inputs:
  files:
  - paddleocr_converter.sources

max_runs_per_component: 100

metadata: {}

Parameters

Init Parameters

These are the parameters you can configure in Pipeline Builder:

Parameter	Type	Default	Description
`base_url`	str \| None	`None`	Base URL of the PaddleOCR API endpoint. Falls back to the `PADDLEOCR_BASE_URL` environment variable if not set.
`access_token`	Secret	`Secret.from_env_var(["PADDLEOCR_ACCESS_TOKEN", "AISTUDIO_ACCESS_TOKEN"])`	Access token for the PaddleOCR API.
`model`	str	`Model.PADDLE_OCR_VL_16`	Document parsing model to use.
`file_type`	str \| None	`None`	Input file type: `pdf`, `image`, or `None` for auto-detection from file extension or MIME type.
`use_doc_orientation_classify`	bool \| None	`False`	Enable document orientation classification for rotated scans.
`use_doc_unwarping`	bool \| None	`False`	Enable text image unwarping for curved or bent documents.
`use_layout_detection`	bool \| None	`None`	Enable layout detection.
`use_chart_recognition`	bool \| None	`None`	Enable chart recognition.
`use_seal_recognition`	bool \| None	`None`	Enable seal recognition.
`use_ocr_for_image_block`	bool \| None	`None`	Recognize text inside image blocks within documents.
`temperature`	float \| None	`None`	Temperature for VLM sampling.
`max_new_tokens`	int \| None	`None`	Maximum number of tokens to generate.
`merge_tables`	bool \| None	`None`	Merge tables across pages.
`additional_params`	dict \| None	`None`	Extra options passed to `PaddleOCRVLOptions.extra_options`.

Was this page helpful?

Key Features​

Configuration​

Connections​

Source Code​

Usage Examples​

Basic Configuration​

Using the Component in a Pipeline​

Parameters​

Init Parameters​

Related Information​