Struct NlpRequest 

#[non_exhaustive]
pub struct NlpRequest<'a> {
    pub text: &'a str,
    pub tasks: NlpTasks,
    pub turns: Option<&'a [Turn<'a>]>,
    pub schema: Option<&'a ExtractionSchema>,
    pub calibration: Option<&'a NlpCalibrationOverrides>,
}

One NLP analysis request: an input plus the heads to populate.

#[non_exhaustive]: build one with NlpRequest::new (or NlpRequest::from_turns) and chain the with_* setters, so later releases can add inputs without breaking callers.

use uni_xervo::traits::{NlpRequest, NlpTasks};
let req = NlpRequest::new("Ada Lovelace was born in London.", NlpTasks::NER);
assert_eq!(req.tasks, NlpTasks::NER);

Fields (Non-exhaustive)

Non-exhaustive structs could have additional fields added in future. Therefore, non-exhaustive structs cannot be constructed in external crates using the traditional Struct { .. } syntax; cannot be matched against without a wildcard ..; and struct update syntax will not work.

text: &'a str

Input prose. Borrowed so batch callers don’t have to clone.

kniv v7 splits it into paragraphs at blank lines ("\n\n"). Ignored when turns is set.

tasks: NlpTasks

Which heads to populate. Unset flags result in empty fields on the corresponding NlpResult entries.

turns: Option<&'a [Turn<'a>]>

A conversation to analyse instead of text: one (speaker, text) Turn per message, oldest first.

Each turn is its own unit — sentences never cross a turn — and its speaker becomes a "speaker" entity, so relations can be anchored on who said what. Models that do not support conversations ignore it.

schema: Option<&'a ExtractionSchema>

The relations and entity types to extract, for NlpTasks::ENTITIES / NlpTasks::FACTS.

Zero-shot: the schema is supplied per request and scored against the whole window. None leaves facts empty and falls back to the model’s own coarse entity types.

calibration: Option<&'a NlpCalibrationOverrides>

Per-request overrides of the model’s shipped decoding thresholds.

None uses the model’s calibration.json as published, which is what the reported quality figures were measured with.

Implementations

impl<'a> NlpRequest<'a>

pub fn new(text: &'a str, tasks: NlpTasks) -> Self

A request over prose.

pub fn from_turns(turns: &'a [Turn<'a>], tasks: NlpTasks) -> Self

A request over conversation turns.

pub fn with_turns(self, turns: &'a [Turn<'a>]) -> Self

Attach conversation turns, which take precedence over text.

pub fn with_schema(self, schema: &'a ExtractionSchema) -> Self

Attach the extraction schema.

pub fn with_calibration(self, calibration: &'a NlpCalibrationOverrides) -> Self

Attach calibration overrides.

Trait Implementations

impl<'a> Clone for NlpRequest<'a>

fn clone(&self) -> NlpRequest<'a>

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<'a> Debug for NlpRequest<'a>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.