Remove deprecated Flash preview and legacy MTP

src/engine/metal/checkpoint.rs

@@ -301,10 +301,6 @@ impl DeepSeekExecutor {
         self.tokens = tokens;
         self.logits = logits;
         self.checkpoint_tag = checkpoint_tag;
-        if let Some(mtp) = &mut self.legacy_mtp {
-            mtp.draft_token = None;
-            mtp.raw_rows = 0;
-        }
         if let Some(dspark) = &mut self.dspark {
             dspark.capture_mask = 0;
             dspark.cache_start = 0;

src/engine/metal/glm.rs

@@ -398,8 +398,6 @@ impl GlmExecutor {
             quality,
             ssd,
             EngineSpeculativeSettings {
-                mtp_draft_tokens: 1,
-                mtp_margin: 3.0,
                 glm_mtp: false,
                 glm_mtp_timing: false,
                 dspark: false,
@@ -3243,8 +3241,7 @@ mod tests {
     use crate::settings::EngineSsdSettings;
 
     fn installed_glm_path() -> std::path::PathBuf {
-        crate::model::engine_artifacts(ModelChoice::Glm52, false, false, &crate::app::models_path())
-            .model
+        crate::model::engine_artifacts(ModelChoice::Glm52, false, &crate::app::models_path()).model
     }
 
     #[test]
@@ -3433,8 +3430,6 @@ mod tests {
                     preload_experts: 0,
                 },
                 EngineSpeculativeSettings {
-                    mtp_draft_tokens: 2,
-                    mtp_margin: 3.0,
                     glm_mtp: enabled,
                     glm_mtp_timing: false,
                     dspark: false,

src/engine/metal/profile.rs

@@ -288,9 +288,10 @@ mod tests {
             std::process::id(),
             std::thread::current().name().unwrap_or("test")
         ));
-        let mut profile = ExpertProfile::new(path.to_str(), ModelChoice::DeepSeekV4Flash, 1, 8, 2)
-            .unwrap()
-            .unwrap();
+        let mut profile =
+            ExpertProfile::new(path.to_str(), ModelChoice::DeepSeekV4Flash0731, 1, 8, 2)
+                .unwrap()
+                .unwrap();
         profile
             .record_row(0, 10, &[1, 2], &[0.6, 0.4], false)
             .unwrap();