Initial commit: E4B-MarkBase model integration with passing tests

- E4B-MarkBase model (42 layers, 4.4GB) loaded successfully - All Phase 1-6 tests passed (model loading, forward pass, vision/audio towers, token generation, performance) - All stress tests passed (5/5 in 127.6s) - Concurrent inference - Memory stress (67.5 tok/s, 0 NaN) - Continuous generation - Batch processing - Long-running stability - Swift Metal inference engine with multimodal support
2026-06-23 18:12:35 +08:00
commit ac75faa0cc
301 changed files with 63426 additions and 0 deletions
@@ -0,0 +1,66 @@
+import XCTest
+@testable import MarkBase
+
+final class CleanMoETest: XCTestCase {
+    func testCleanMoEPerformance() throws {
+        print("\n" + String(repeating: "=", count: 60))
+        print("Clean MoE Performance Test (No Debug Output)")
+        print(String(repeating: "=", count: 60))
+        
+        let modelDir = "/Users/accusys/MarkBaseEngine/models/gemma-4-26b-a4b-it-4bit"
+        let engine = try MarkBaseEngine(autoCompile: true)
+        let model = try E4BModel(modelDir: modelDir, engine: engine, maxContextLength: 128)
+        
+        print("\n✓ Model loaded")
+        
+        // Benchmark without debug output
+        let inputIds: [Int32] = [1, 3234, 357, 659, 198]
+        let iterations = 3
+        
+        var totalSpeed: Double = 0
+        
+        for iter in 0..<iterations {
+            print("\nIteration \(iter + 1):")
+            
+            let start = Date()
+            
+            // Process input tokens
+            for (index, tokenId) in inputIds.enumerated() {
+                let logits = try model.forward(tokenId: Int(tokenId), position: index)
+                let nanCount = logits.filter { $0.isNaN }.count
+                if nanCount > 0 {
+                    print("  ERROR: NaN detected in logits!")
+                    return
+                }
+            }
+            
+            // Generate 5 new tokens
+            var lastToken = inputIds.last!
+            for i in 0..<5 {
+                let logits = try model.forward(tokenId: Int(lastToken), position: inputIds.count + i)
+                let nanCount = logits.filter { $0.isNaN }.count
+                if nanCount > 0 {
+                    print("  ERROR: NaN detected in generation!")
+                    return
+                }
+                let maxIdx = logits.enumerated().max(by: { $0.element < $1.element })!.offset
+                lastToken = Int32(maxIdx)
+            }
+            
+            let elapsed = Date().timeIntervalSince(start)
+            let tokens = inputIds.count + 5
+            let speed = Double(tokens) / elapsed
+            
+            print("  Tokens: \(tokens), Time: \(String(format: "%.3f", elapsed))s")
+            print("  Speed: \(String(format: "%.2f", speed)) tok/s")
+            
+            totalSpeed += speed
+        }
+        
+        let avgSpeed = totalSpeed / Double(iterations)
+        print("\n✓ Average speed: \(String(format: "%.2f", avgSpeed)) tok/s")
+        print("✓ No NaN detected in all iterations!")
+        
+        print("\n" + String(repeating: "=", count: 60))
+    }
+}