Struct SecDecodingSteerer 

pub struct SecDecodingSteerer { /* private fields */ }

SafetyMode::SecDecoding — base-vs-safety-model logit steering.

FOLLOW-UP (ADR-005): requires two ~1B models run on Candle. Until model assets are wired, this returns no adjustment and reports its intended mode, so callers can select it without it silently mis-steering.

Implementations

impl SecDecodingSteerer

pub fn placeholder() -> Self

Trait Implementations

impl SafetySteerer for SecDecodingSteerer

fn adjust(&self, _recent: &[Token]) -> LogitAdjustment

fn mode(&self) -> SafetyMode

fn adjust_with_logits( &self, recent_tokens: &[Token], _base_logits: &[i32], ) -> LogitAdjustment

Logit-aware steering (ADR-013). Given the base model’s next-token logits for this step, return the adjustment to apply (still after the grammar mask, before sampling — the ADR-005 order is unchanged).