Accelerate DeepSeek DSpark inference
This commit is contained in:
@@ -90,7 +90,7 @@ impl SpeculativePreferences {
|
||||
glm_mtp: self.glm_mtp,
|
||||
glm_mtp_timing: self.glm_mtp_timing,
|
||||
dspark: self.dspark_enabled,
|
||||
dspark_confidence_threshold: self.dspark_confidence_threshold.unwrap_or(0.6),
|
||||
dspark_confidence_threshold: self.dspark_confidence_threshold.unwrap_or(0.8),
|
||||
dspark_confidence_threshold_set: self.dspark_confidence_threshold.is_some(),
|
||||
dspark_strict: self.dspark_strict,
|
||||
dspark_exact_sampling: self.dspark_exact_sampling,
|
||||
@@ -787,11 +787,11 @@ mod tests {
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn speculative_settings_match_ds4_defaults_and_dependencies() {
|
||||
fn speculative_settings_match_acceleration_defaults_and_dependencies() {
|
||||
let defaults = SpeculativePreferences::default();
|
||||
let engine = defaults.engine_settings();
|
||||
assert_eq!((engine.mtp_draft_tokens, engine.mtp_margin), (1, 3.0));
|
||||
assert_eq!(engine.dspark_confidence_threshold, 0.6);
|
||||
assert_eq!(engine.dspark_confidence_threshold, 0.8);
|
||||
assert!(!engine.dspark_confidence_threshold_set);
|
||||
|
||||
let tuned = SpeculativePreferences {
|
||||
|
||||
Reference in New Issue
Block a user