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

测试26B加载.swift

@@ -0,0 +1,68 @@
+import Foundation
+import MarkBase
+
+print("=== 测试 Gemma-4 26B 模型加载 ===")
+
+let modelDir = "/Users/accusys/.cache/huggingface/hub/models--mlx-community--gemma-4-26b-a4b-mxfp4"
+
+print("\n步骤 1: 检查文件")
+let fm = FileManager.default
+
+let files = [
+    "config.json",
+    "tokenizer.json",
+    "model-00001-of-00003.safetensors",
+    "model-00002-of-00003.safetensors",
+    "model-00003-of-00003.safetensors",
+    "model.safetensors.index.json"
+]
+
+for file in files {
+    let path = modelDir + "/" + file
+    if fm.fileExists(atPath: path) {
+        let size = try! fm.attributesOfItem(atPath: path)[.size] as! Int64
+        print("  ✓ \(file): \(size / 1024 / 1024) MB")
+    } else {
+        print("  ✗ \(file): NOT FOUND")
+    }
+}
+
+print("\n步骤 2: 加载 Engine")
+do {
+    let engine = try E4BEngine(autoCompile: true)
+    print("  ✓ Engine created")
+    
+    print("\n步骤 3: 加载 Model (26B)")
+    print("  警告: 这可能需要几分钟...")
+    
+    // Try loading with smaller context to save memory
+    let model = try E4BModel(modelDir: modelDir, engine: engine, maxContextLength: 128)
+    
+    print("  ✓ Model loaded!")
+    print("  Layers: \(model.numHiddenLayers)")
+    print("  Hidden size: \(model.hiddenSize)")
+    print("  Vocab size: \(model.vocabSize)")
+    
+    print("\n步骤 4: 加载 Tokenizer")
+    let tokenizer = try TokenizerFactory.load(modelDir: modelDir)
+    print("  ✓ Tokenizer loaded")
+    
+    print("\n步骤 5: 测试推理")
+    let tokens = tokenizer.encode(text: "Hello")
+    print("  Test tokens: \(tokens)")
+    
+    print("\n=== 26B 模型加载成功 ===")
+    
+} catch {
+    print("\n✗ 错误: \(error)")
+    print("\n可能的错误原因:")
+    print("  1. 权重命名不匹配 (language_model.model.layers vs layers)")
+    print("  2. MoE 结构不支持 (需要专家路由)")
+    print("  3. Memory 不足 (26B 需要 ~16GB)")
+    print("  4. 配置格式不兼容")
+    
+    print("\n建议:")
+    print("  1. 检查权重命名并适配")
+    print("  2. 实现 MoE 支持")
+    print("  3. 转换为我们的标准格式")
+}