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 +1,110 @@
+import XCTest
+@testable import MarkBase
+
+class LongTextLimitTest: XCTestCase {
+    
+    func testLongTextLimit() throws {
+        print("\n═══════════════════════════════════════════════════════════════════")
+        print("  Long Text Limit Test")
+        print("═══════════════════════════════════════════════════════════════════\n")
+        
+        let modelPath = "/Users/accusys/MarkBaseEngine/models/gemma-4-26b-standard"
+        
+        guard FileManager.default.fileExists(atPath: modelPath) else {
+            print("⚠ Model not found")
+            return
+        }
+        
+        let engine = try MarkBaseEngine(autoCompile: true)
+        
+        // Test different context lengths
+        let contextLengths = [128, 256, 512, 1024, 2048]
+        
+        for maxLen in contextLengths {
+            print("\n───────────────────────────────────────────────────────────────────")
+            print("Testing maxContextLength: \(maxLen)")
+            print("───────────────────────────────────────────────────────────────────")
+            
+            do {
+                let model = try E4BModel(modelDir: modelPath, engine: engine, maxContextLength: maxLen)
+                print("✓ Model loaded with maxContextLength=\(maxLen)")
+                print("  Layers: \(model.numHiddenLayers)")
+                print("  Hidden: \(model.hiddenSize)")
+                
+                // Test forward pass at position 0
+                let result0 = try model.forwardOptimized(tokenId: 2, position: 0)
+                let nan0 = result0.filter { $0.isNaN }.count
+                print("  Forward at pos 0: NaN=\(nan0)/\(result0.count)")
+                
+                // Test forward pass at max position
+                let maxPos = maxLen - 1
+                if maxPos < 1000 {  // Limit test time
+                    let resultMax = try model.forwardOptimized(tokenId: 2, position: maxPos)
+                    let nanMax = resultMax.filter { $0.isNaN }.count
+                    print("  Forward at pos \(maxPos): NaN=\(nanMax)/\(resultMax.count)")
+                    
+                    if nanMax == 0 {
+                        print("  ✓✓✓ Position \(maxPos) OK")
+                    } else {
+                        print("  ✗ NaN at position \(maxPos)")
+                    }
+                }
+                
+            } catch {
+                print("✗ Failed: \(error)")
+            }
+        }
+        
+        print("\n═══════════════════════════════════════════════════════════════════")
+        print("  Test completed")
+        print("═══════════════════════════════════════════════════════════════════\n")
+    }
+    
+    func testMemoryUsage() throws {
+        print("\n═══════════════════════════════════════════════════════════════════")
+        print("  Memory Usage Test")
+        print("═══════════════════════════════════════════════════════════════════\n")
+        
+        let modelPath = "/Users/accusys/MarkBaseEngine/models/gemma-4-26b-standard"
+        
+        guard FileManager.default.fileExists(atPath: modelPath) else {
+            print("⚠ Model not found")
+            return
+        }
+        
+        let engine = try MarkBaseEngine(autoCompile: true)
+        
+        // Calculate theoretical memory usage
+        let configs = [
+            ("128 context", 128),
+            ("512 context", 512),
+            ("1024 context", 1024),
+            ("2048 context", 2048),
+            ("4096 context", 4096),
+            ("8192 context", 8192)
+        ]
+        
+        print("\nTheoretical KV Cache Memory Usage:")
+        print("Model: 26B-Standard (30 layers, hidden=2816, nKvHeads=4, headDim=256)")
+        
+        for (name, maxLen) in configs {
+            // KV cache memory: 2 * numLayers * maxLen * nKvHeads * headDim * 4 bytes
+            let kvMemory = 2 * 30 * maxLen * 4 * 256 * 4  // K + V, 30 layers, Float32
+            let kvMemoryMB = Double(kvMemory) / 1024.0 / 1024.0
+            
+            print("  \(name): \(String(format: "%.2f", kvMemoryMB)) MB")
+            
+            if kvMemoryMB < 100 {
+                print("    ✓ Safe for 128GB unified memory")
+            } else if kvMemoryMB < 500 {
+                print("    ⚠ Moderate usage")
+            } else {
+                print("    ✗ High memory usage")
+            }
+        }
+        
+        print("\n═══════════════════════════════════════════════════════════════════")
+        print("  Test completed")
+        print("═══════════════════════════════════════════════════════════════════\n")
+    }
+}