Property VlmArbitration
- Namespace
- LMKit.Document.Conversion
- Assembly
- LM-Kit.NET.dll
VlmArbitration
Under VlmOcr, applies the same judgement
as HybridArbitration to every page that carries a text layer: the
vision model's reading is weighed against the text tier's reading of the same page,
and the page's own text is kept when the model's reading dropped or shuffled what
the page states. Defaults to true.
public bool VlmArbitration { get; set; }Property Value
Examples
// Take the vision model's reading of every page unchecked.
var options = new DocumentToMarkdownOptions { VlmArbitration = false };Remarks
A model reading a whole document page by page can lose a page in ways a single
conversion never shows: a completion that stops after the first of two look-alike
figures, a chart region drawn instead of read. The text layer is the reference the
page itself provides, so a reading that reproduces almost none of it gives way to
it. A page without a text layer has no reference and its model reading always
stands. Set to false to take every model reading as it comes.