Trait NlpModel 

pub trait NlpModel: ModelInfo {
    // Required methods
    fn analyze<'life0, 'life1, 'async_trait>(
        &'life0 self,
        requests: Vec<NlpRequest<'life1>>,
    ) -> Pin<Box<dyn Future<Output = Result<Vec<NlpResult>>> + Send + 'async_trait>>
       where Self: 'async_trait,
             'life0: 'async_trait,
             'life1: 'async_trait;
    fn supported_tasks(&self) -> NlpTasks;

    // Provided methods
    fn label_maps(&self) -> Option<&NlpLabelMaps> { ... }
    fn warmup<'life0, 'async_trait>(
        &'life0 self,
    ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>
       where Self: Sync + 'async_trait,
             'life0: 'async_trait { ... }
}

A multi-head structured-NLP model.

Canonical default model

The reference implementation is dragonscale-ai/kniv-deberta-nlp-base-en-xsmall: a DeBERTa-v3-xsmall multi-head cascade producing POS / NER / DEP / SRL / CLS in a single forward pass. It will ship as the canonical nlp/default catalog alias once the local/onnx provider gains an NlpModel implementation in a follow-up release.

Tagsets used by the reference model:

• POS: Universal Dependencies English EWT (17 classes)
• NER: OntoNotes (18 entity types)
• DEP: Universal Dependencies English EWT
• SRL: PropBank (42 classes)
• CLS: 8 dialog acts (internal dataset)

Language: English only. Maximum input: 128 tokens — provider-side chunking is the implementation’s responsibility; the trait contract keeps token offsets stable in whole-document UTF-8 byte coordinates regardless of internal chunking.

Required Methods

fn analyze<'life0, 'life1, 'async_trait>( &'life0 self, requests: Vec<NlpRequest<'life1>>, ) -> Pin<Box<dyn Future<Output = Result<Vec<NlpResult>>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Analyze a batch of texts, returning one NlpResult per request.

Errors

Returns an error if a requested task is not in supported_tasks, or if the provider fails internally.

fn supported_tasks(&self) -> NlpTasks

Heads this model can populate. Callers may request a subset via NlpRequest::tasks.

Provided Methods

fn label_maps(&self) -> Option<&NlpLabelMaps>

The label vocabularies this model decodes against, if it exposes them.

Returns Some for models with a fixed, introspectable label set (e.g. the ONNX cascade), letting callers map labels to class ids and read SpeechAct::scores against NlpLabelMaps::cls. The default is None for models that do not expose vocabularies (e.g. remote or mock).

fn warmup<'life0, 'async_trait>( &'life0 self, ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>

where Self: Sync + 'async_trait, 'life0: 'async_trait,

Optional warmup hook. The default is a no-op.

Errors

Returns an error if the underlying model fails to initialize.

Dyn Compatibility

This trait is dyn compatible.

In older versions of Rust, dyn compatibility was called "object safety".

Implementors

impl NlpModel for InstrumentedNlpModel