Skip to main content

​
Docling

To fine-tune the machine resources, and be able to use all machine resources in ingestion process, we provide the following configuration to configure based on the machine:

​
Performance Configuration

  1. Number of workers: This is the configuration that determines how many workers Docling will have. We recommend leaving it at 2, where 1 worker will process large files and the other will process smaller files.
  2. Number of threads: This configuration will be used to limit the capacity that our internal extraction service will use. Consider that this number should be lower to the maximum cores that machine has.
Important: The number of threads must be divisible by the number of workers. In case you want to test and improve the performance, consider configure as: 
external:
  docling:
    numWorkers: M
    numThreads: N

​
Optional Configuration

  1. Table mode: Controls the table extraction strategy:
    • accurate: Maximum precision for complex tables. Higher processing time. Default option.
    • fast: Optimized for speed with good accuracy. Lower processing time.
    • none: Disables table extraction. No processing time.
  2. Do cell matching: Enables matching of table cells to improve structure recognition. Recommended for complex tables. By default is true.
  3. Force full page OCR: Forces OCR processing on entire pages instead of selective regions. Use when standard extraction misses content. This will increase the processing time significantly. By default is false. 
external:
  docling:
    tableMode: "accurate"
    doCellMatching: true
    forceFullPageOCR: false