pub struct SamplingOverride {
pub temperature: f64,
pub top_p: f64,
pub frequency_penalty: f64,
pub presence_penalty: f64,
pub logit_biases: Vec<LogitBias>,
}Expand description
Sampling parameter overrides derived from cognitive state.
These parameters modulate model generation at the sampling level. Brain analog: neuromodulators (DA, NE, ACh, 5-HT) don’t carry specific information — they change gain, threshold, and sensitivity of neural circuits. SamplingOverride is Noos’s neuromodulatory output.
Fields§
§temperature: f64Generation temperature. Lower = focused/deterministic, higher = creative/exploratory. Modulated by gain_mode (Aston-Jones 2005) and body_budget (Barrett 2017).
top_p: f64Nucleus sampling threshold. Lower = fewer token candidates, higher = more diverse. Modulated by gain_mode and pe_volatility (Behrens 2007).
frequency_penalty: f64Frequency penalty. Reduces probability of recently-used tokens. Elevated under high arousal + negative valence (LeDoux 1996: tunnel vision → avoid repetition).
presence_penalty: f64Presence penalty. Reduces probability of any token that appeared in output. Elevated under high resource pressure (Sterling 2012: conserve under allostatic load).
logit_biases: Vec<LogitBias>Token-level logit biases. Applied before sampling. Currently empty in Tier 1 Phase 1 (requires tokenizer integration).
Trait Implementations§
Source§impl Clone for SamplingOverride
impl Clone for SamplingOverride
Source§fn clone(&self) -> SamplingOverride
fn clone(&self) -> SamplingOverride
1.0.0 · Source§fn clone_from(&mut self, source: &Self)
fn clone_from(&mut self, source: &Self)
source. Read more