Skip to content

armgit / uzu

Public Imported from GitHub
0

A high-performance inference engine for AI models

remove cli settings hardcode

eugene49bdc9dc9af6
diff --git a/crates/uzu/src/cli/components/preferences.rs b/crates/uzu/src/cli/components/preferences.rs
index 470f0ed73..a1855fc44 100644
--- a/crates/uzu/src/cli/components/preferences.rs
+++ b/crates/uzu/src/cli/components/preferences.rs
@@ -50,7 +50,7 @@ impl SamplingMode {
}
}
-#[derive(Debug, Clone, Copy, PartialEq, Serialize, Deserialize)]
+#[derive(Debug, Clone, Copy, PartialEq, Serialize, Deserialize, Default)]
#[serde(default)]
pub struct SamplingPreferences {
pub mode: SamplingMode,
@@ -67,25 +67,6 @@ pub struct SamplingPreferences {
pub suffix_repetition_length: i64,
}
-impl Default for SamplingPreferences {
- fn default() -> Self {
- Self {
- mode: SamplingMode::default(),
- temperature_enabled: false,
- temperature: 0.8,
- top_k_enabled: false,
- top_k: 64,
- top_p_enabled: false,
- top_p: 0.95,
- min_p_enabled: false,
- min_p: 0.05,
- repetition_penalty_enabled: false,
- repetition_penalty: 1.1,
- suffix_repetition_length: 1024,
- }
- }
-}
-
impl SamplingPreferences {
pub fn suffix_repetition_length_enabled(&self) -> bool {
self.repetition_penalty_enabled
@@ -190,66 +171,3 @@ pub(super) fn cycle<T: Copy + PartialEq>(
let next = (index + delta).rem_euclid(length) as usize;
values[next]
}
-
-#[cfg(test)]
-mod tests {
- use super::*;
-
- #[test]
- fn model_default_maps_to_default_policy() {
- let prefs = SamplingPreferences {
- mode: SamplingMode::ModelDefault,
- ..SamplingPreferences::default()
- };
- assert!(matches!(prefs.policy(), SamplingPolicy::Default {}));
- }
-
- #[test]
- fn greedy_maps_to_greedy_method() {
- let prefs = SamplingPreferences {
- mode: SamplingMode::Greedy,
- ..SamplingPreferences::default()
- };
- assert!(matches!(
- prefs.policy(),
- SamplingPolicy::Custom {
- method: SamplingMethod::Greedy {},
- }
- ));
- }
-
- #[test]
- fn stochastic_only_forwards_enabled_parameters() {
- let prefs = SamplingPreferences {
- mode: SamplingMode::Stochastic,
- temperature_enabled: true,
- temperature: 0.7,
- top_k_enabled: false,
- top_p_enabled: true,
- top_p: 0.9,
- min_p_enabled: false,
- ..SamplingPreferences::default()
- };
- match prefs.policy() {
- SamplingPolicy::Custom {
- method:
- SamplingMethod::Stochastic {
- temperature,
- top_k,
- top_p,
- min_p,
- repetition_penalty,
- suffix_repetition_length,
- },
- } => {
- assert_eq!(temperature, Some(0.7));
- assert_eq!(top_k, None);
- assert_eq!(top_p, Some(0.9));
- assert_eq!(min_p, None);
- assert_eq!(repetition_penalty, None);
- assert_eq!(suffix_repetition_length, None);
- },
- _ => panic!("stochastic mode must map to a custom stochastic policy"),
- }
- }
-}
diff --git a/crates/uzu/src/cli/flows/settings.rs b/crates/uzu/src/cli/flows/settings.rs
index 2620a79e6..05ba8eee9 100644
--- a/crates/uzu/src/cli/flows/settings.rs
+++ b/crates/uzu/src/cli/flows/settings.rs
@@ -113,27 +113,48 @@ fn toggle(
preferences: &mut Preferences,
field: Field,
support: ThinkingSupport,
+ defaults: ModelSamplingDefaults,
) {
+ let sampling = &mut preferences.sampling;
match field {
Field::Thinking => adjust(preferences, Field::Thinking, 1, support),
- Field::SamplingMode => preferences.sampling.mode = preferences.sampling.mode.next(),
- Field::Temperature => preferences.sampling.temperature_enabled = !preferences.sampling.temperature_enabled,
- Field::TopK => preferences.sampling.top_k_enabled = !preferences.sampling.top_k_enabled,
- Field::TopP => preferences.sampling.top_p_enabled = !preferences.sampling.top_p_enabled,
- Field::MinP => preferences.sampling.min_p_enabled = !preferences.sampling.min_p_enabled,
+ Field::SamplingMode => sampling.mode = sampling.mode.next(),
+ Field::Temperature => {
+ toggle_value(&mut sampling.temperature_enabled, &mut sampling.temperature, defaults.temperature)
+ },
+ Field::TopK => toggle_value(&mut sampling.top_k_enabled, &mut sampling.top_k, defaults.top_k),
+ Field::TopP => toggle_value(&mut sampling.top_p_enabled, &mut sampling.top_p, defaults.top_p),
+ Field::MinP => toggle_value(&mut sampling.min_p_enabled, &mut sampling.min_p, defaults.min_p),
Field::RepetitionPenalty => {
- preferences.sampling.repetition_penalty_enabled = !preferences.sampling.repetition_penalty_enabled
+ toggle_value(
+ &mut sampling.repetition_penalty_enabled,
+ &mut sampling.repetition_penalty,
+ defaults.repetition_penalty,
+ );
+ if sampling.repetition_penalty_enabled
+ && let Some(value) = defaults.suffix_repetition_length
+ {
+ sampling.suffix_repetition_length = value;
+ }
},
Field::SuffixRepetitionLength => {
- assert!(
- preferences.sampling.suffix_repetition_length_enabled(),
- "suffix_repetition_length cannot be enabled"
- );
- preferences.sampling.suffix_repetition_length_disable();
+ assert!(sampling.suffix_repetition_length_enabled(), "suffix_repetition_length cannot be enabled");
+ sampling.suffix_repetition_length_disable();
},
}
}
+fn toggle_value<T>(
+ enabled: &mut bool,
+ value: &mut T,
+ default: Option<T>,
+) {
+ *enabled = !*enabled;
+ if *enabled && let Some(default) = default {
+ *value = default;
+ }
+}
+
fn step_f64(
value: f64,
delta: i64,
@@ -246,7 +267,7 @@ fn SettingsFlowView(
},
KeyCode::Char(' ') => {
let mut preferences = draft.get();
- toggle(&mut preferences, field, support);
+ toggle(&mut preferences, field, support, defaults);
draft.set(preferences);
},
KeyCode::Enter => {