Skip to content

Hybrid Mode

Hybrid mode is the switch that enables EdgeParse’s OCR recovery path for raster-backed tables. In the current runtime, the local rule-based pipeline remains the main extractor and the OCR recovery path only activates when a hybrid backend is configured.

On the official odl-bench board, EdgeParse [hybrid] leads NID, TEDS, MHS, and overall (0.900).

Terminal window
# CLI
edgeparse document.pdf -f json --hybrid docling-fast
import edgeparse
result = edgeparse.convert("document.pdf",
format="json",
hybrid="docling-fast"
)
ScenarioRecommendation
Speed-critical productionStandard mode (0.284 s/doc)
Maximum table accuracyHybrid mode with docling-fast (odl-bench overall 0.900)
No GPU availableStandard mode — No ML stack required for born-digital PDFs; optional in-browser OCR for image tables
Complex academic papersHybrid mode
BackendDescription
docling-fastEnables hybrid mode and OCR recovery for raster-backed tables
  • Speed: Hybrid mode is slower because OCR recovery is enabled on qualifying pages
  • Accuracy: Better recovery on image-backed tables and scanned table regions; leads the official odl-bench board
  • Dependencies: Requires the backend to be installed