Struct AnswerSpan 

pub struct AnswerSpan {
    pub start: Vec<u32>,
    pub end: u32,
    pub values: Option<TokenText>,
}

Supervised chat records scored on the assistant’s answer only: every target from the first token after start (the rendered generation prompt, e.g. <|Bot|></think>) through the first end token (the turn terminator) carries the LM loss; the prompt is context. This is the multi-token sibling of focus_tokens — a JSON verdict, a number, a sentence — where a one-token label does not fit.

Fields

start: Vec<u32>

Token ids of the generation prompt; the answer starts right after its LAST occurrence in the record.

end: u32

The turn terminator (one token); the last scored target.

values: Option<TokenText>

Decoded text of every vocabulary id. When present, the answer’s top-level JSON values are located in it and the held score grades whole verdicts (answer_values): a question counts as correct only when every token of its value is the argmax. Token accuracy over the span alone is dominated by the braces, quotes and question ids the model copies from the prompt.

Trait Implementations

impl Clone for AnswerSpan

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for AnswerSpan

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

Formats the value using the given formatter.

impl Eq for AnswerSpan

impl PartialEq for AnswerSpan

fn eq(&self, other: &Self) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for AnswerSpan