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 +1,86 @@
+import XCTest
+@testable import MarkBase
+
+class Model12BArchitectureTest: XCTestCase {
+    
+    func testModel12BFullArchitecture() throws {
+        print("\n═══════════════════════════════════════════════════════════════════")
+        print("  12B Standard Complete Architecture Check")
+        print("═══════════════════════════════════════════════════════════════════\n")
+        
+        let modelPath = "/Users/accusys/.cache/huggingface/hub/models--mlx-community--gemma-4-12B-it-4bit/snapshots/73bcf09092aa277861d5a191b989b666f7f32e8f"
+        
+        guard FileManager.default.fileExists(atPath: modelPath) else {
+            print("⚠ Model not found")
+            return
+        }
+        
+        // Load tensors
+        let index = try SafeTensorsIndex(modelDir: modelPath)
+        var readers: [String: SafeTensorsReader] = [:]
+        for shardFile in Set(index.weightMap.values) {
+            readers[shardFile] = try SafeTensorsReader(path: "\(modelPath)/\(shardFile)")
+        }
+        
+        let allTensors = readers.values.flatMap { $0.allTensors }
+        
+        print("1. Total Tensors: \(allTensors.count)")
+        
+        // Audio tower analysis
+        print("\n2. Audio Tower:")
+        let audioTensors = allTensors.filter { $0.name.contains("audio_tower") || $0.name.contains("audio_model") }
+        print("  Audio tensors: \(audioTensors.count)")
+        if !audioTensors.isEmpty {
+            print("  Sample tensors:")
+            for tensor in audioTensors.prefix(10) {
+                print("    \(tensor.name): shape=\(tensor.shape)")
+            }
+        }
+        
+        // Vision tower analysis
+        print("\n3. Vision Tower:")
+        let visionTensors = allTensors.filter { $0.name.contains("vision_tower") || $0.name.contains("vision_model") }
+        print("  Vision tensors: \(visionTensors.count)")
+        if !visionTensors.isEmpty {
+            print("  Sample tensors:")
+            for tensor in visionTensors.prefix(10) {
+                print("    \(tensor.name): shape=\(tensor.shape)")
+            }
+        }
+        
+        // TEXT model analysis
+        print("\n4. TEXT Model:")
+        let textTensors = allTensors.filter { $0.name.contains("language_model") || $0.name.contains("layers.") }
+        print("  TEXT-related tensors: \(textTensors.count)")
+        
+        // Embed tokens
+        let embedTokens = allTensors.filter { $0.name.contains("embed_tokens") }
+        print("  Embed tokens: \(embedTokens.count)")
+        for tensor in embedTokens.prefix(5) {
+            print("    \(tensor.name): shape=\(tensor.shape)")
+        }
+        
+        // Summary
+        print("\n═══════════════════════════════════════════════════════════════════")
+        print("  Architecture Summary")
+        print("═══════════════════════════════════════════════════════════════════\n")
+        
+        if audioTensors.count > 0 {
+            print("✓ 12B HAS AUDIO TOWER: \(audioTensors.count) tensors")
+        } else {
+            print("✗ 12B NO AUDIO TOWER")
+        }
+        
+        if visionTensors.count > 0 {
+            print("✓ 12B HAS VISION TOWER: \(visionTensors.count) tensors")
+        } else {
+            print("✗ 12B NO VISION TOWER")
+        }
+        
+        print("✓ TEXT Model: \(textTensors.count) tensors")
+        
+        print("\nModel Type: \(audioTensors.count > 0 || visionTensors.count > 0 ? "Multimodal" : "Pure TEXT")")
+        
+        print("\n═══════════════════════════════════════════════════════════════════")
+    }
+}