DocumentSplitSettings
Settings for DocumentSplit. Values fall back through three levels: document → SDK → built-in default. Writes target the document only when set on a document’s settings, otherwise the SDK globally when set on SdkSettings.
using Nutrient;The SDK creates this class through factory methods or other SDK objects.
Properties
BoundaryThreshold
public float BoundaryThreshold { get; set; }Calibrated per-page confidence, in [0,1], at or above which a page starts a new sub-document. The first page is always a boundary. Higher values yield fewer, larger segments (the model must be more certain to cut); lower values split more eagerly. Default 0.5.
Type: float
EnableDocumentDecisionSplit
public bool EnableDocumentDecisionSplit { get; set; }When true, compute per-page boundary confidences via the nutrient-document-decision GGUF model’s chunk-based shared-state design instead of the SigLIP2/e5/CRF ONNX pipeline. Resolved from the CDN (Vision/DocumentDecision/*). Slower than the specialist — opt-in, not a default. Default false.
Type: bool
IncludePageConfidences
public bool IncludePageConfidences { get; set; }Diagnostic opt-in. When true, the split artifact additionally carries the calibrated boundary confidence for every page (pageConfidences) and, when StreamCandidateLabels is set, every page’s document type and its confidence (pageDocumentTypes, pageDocumentTypeConfidences), letting a caller re-split the stream offline without re-running the model. Default false so production artifacts stay lean.
Type: bool
RenderDpi
public int RenderDpi { get; set; }DPI used to render each page for the image tower. Default 200.
Type: int
StreamCandidateLabels
public string? StreamCandidateLabels { get; set; }JSON array of candidate document-type labels, each {“key”: string, “description”: string|null} (at least 2 entries). Setting this classifies every page against these labels (one ‘choice’ question per page, fused into the same pass that decides boundaries) and splits wherever the page document type changes, in addition to the detected boundaries, so every segment has one document type. Labels can therefore only add splits. Zero-shot, not a trained task. Requires EnableDocumentDecisionSplit. Left unset (the default), no classification runs and the split is boundaries only.
Type: string?
Resource management
public void Dispose()DocumentSplitSettings implements IDisposable. Call Dispose() or use a C# using declaration to release its native handle right away. If you don’t, the handle is released when the object is garbage collected.