Clear cache between benchmark runs

Author: DePasqualeOrg

Tests/Benchmarks/ModelLoadingBenchmarks.swift

@@ -1,5 +1,6 @@
 import Foundation
 import Hub
+import MLX
 import MLXLLM
 import MLXLMCommon
 import MLXVLM
@@ -38,6 +39,9 @@ struct ModelLoadingBenchmarks {
             print("LLM load run \(i): \(String(format: "%.0f", elapsed))ms")
 
             _ = modelContext  // silence unused warning
+
+            // Clear GPU cache to ensure independent measurements
+            GPU.clearCache()
         }
 
         let avg = times.reduce(0, +) / Double(times.count)
@@ -72,6 +76,9 @@ struct ModelLoadingBenchmarks {
             print("VLM load run \(i): \(String(format: "%.0f", elapsed))ms")
 
             _ = modelContext  // silence unused warning
+
+            // Clear GPU cache to ensure independent measurements
+            GPU.clearCache()
         }
 
         let avg = times.reduce(0, +) / Double(times.count)