Struct NerEngine 

pub struct NerEngine { /* private fields */ }

NER engine wrapping gline-rs GLiNER in span mode.

Uses onnx-community/gliner_large-v2.1 (or any span-based GLiNER ONNX model).

Implementations

impl NerEngine

pub fn new( model_path: &Path, tokenizer_path: &Path, threshold: f32, ) -> Result<Self, NerError>

Create a new NER engine from model and tokenizer paths.

• model_path: path to model.onnx (or model_int8.onnx)
• tokenizer_path: path to tokenizer.json
• threshold: minimum GLiNER span probability (0.0–1.0). Parameters::default uses 0.5 which is too aggressive for domain-specific labels; use a lower value like 0.1–0.3 and let the pipeline apply its own threshold on top.

pub fn extract( &self, text: &str, labels: &[&str], label_to_type: Option<&HashMap<&str, &str>>, ) -> Result<Vec<ExtractedEntity>, NerError>

Extract entities from text using the given labels.

label_to_type is an optional mapping from GLiNER label string → canonical entity type key. Pass None to use the label string as-is for entity_type. Pass Some(pairs) when using natural-language descriptions as labels so the returned entity_type is the short canonical key (e.g. “Database”).