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 +1,114 @@
+import Foundation
+
+/// Single source of truth for Metal kernel source code.
+/// Tests use these constants instead of duplicating inline strings.
+public enum MetalKernels {
+    public static let vectorAdd = """
+    #include <metal_stdlib>
+    using namespace metal;
+    kernel void vector_add(
+        device const float *a [[buffer(0)]],
+        device const float *b [[buffer(1)]],
+        device float *c [[buffer(2)]],
+        constant uint &n [[buffer(3)]],
+        uint id [[thread_position_in_grid]]
+    ) {
+        if (id < n) c[id] = a[id] + b[id];
+    }
+    """
+
+    public static let matmulF32 = """
+    #include <metal_stdlib>
+    using namespace metal;
+    kernel void matmul_f32(
+        device const float *A [[buffer(0)]],
+        device const float *B [[buffer(1)]],
+        device float *C [[buffer(2)]],
+        constant uint &M [[buffer(3)]],
+        constant uint &N [[buffer(4)]],
+        constant uint &K [[buffer(5)]],
+        uint2 gid [[thread_position_in_grid]]
+    ) {
+        if (gid.x >= N || gid.y >= M) return;
+        float sum = 0.0;
+        for (uint k = 0; k < K; k++)
+            sum += A[gid.y * K + k] * B[k * N + gid.x];
+        C[gid.y * N + gid.x] = sum;
+    }
+    """
+
+    /// Full E4B inference kernel source (reads from .metal file at runtime).
+    /// Use for JIT compilation in tests.
+    public static var e4bSource: String {
+        let url = URL(fileURLWithPath: #filePath)
+            .deletingLastPathComponent()
+            .appendingPathComponent("Metal/MetalKernels.metal")
+        return try! String(contentsOf: url, encoding: .utf8)
+    }
+    
+    /// Optimized SIMD kernel source for Phase 1.
+    /// Includes attention, matmul, and norm optimizations.
+    public static var optimizedSource: String {
+        let url = URL(fileURLWithPath: #filePath)
+            .deletingLastPathComponent()
+            .appendingPathComponent("Metal/OptimizedKernels.metal")
+        return try! String(contentsOf: url, encoding: .utf8)
+    }
+    
+    /// Combined source: original + optimized kernels.
+    /// Use for production deployment.
+    public static var combinedSource: String {
+        return e4bSource + "\n" + optimizedSource
+    }
+    
+    /// Fusion kernel source for Phase 1.3.
+    /// Includes kernel fusion optimizations.
+    public static var fusionSource: String {
+        let url = URL(fileURLWithPath: #filePath)
+            .deletingLastPathComponent()
+            .appendingPathComponent("Metal/FusionKernels.metal")
+        return try! String(contentsOf: url, encoding: .utf8)
+    }
+    
+    /// Fused kernel source for Phase 2.
+    /// Includes advanced kernel fusion for embedding, norm, and matmul.
+    public static var fusedKernelsSource: String {
+        let url = URL(fileURLWithPath: #filePath)
+            .deletingLastPathComponent()
+            .appendingPathComponent("Metal/FusedKernels.metal")
+        return try! String(contentsOf: url, encoding: .utf8)
+    }
+    
+    /// Full optimized source: original + SIMD + fusion.
+    /// Maximum optimization without MPS.
+    public static var fullOptimizedSource: String {
+        return combinedSource + "\n" + fusionSource + "\n" + fusedKernelsSource
+    }
+    
+    /// Full optimized source with all kernels.
+    /// Strips duplicate #include and using namespace from subsequent files.
+    public static var fullOptimizedSourceWithFusion: String {
+        // Start with first file (includes its #include and using namespace)
+        var result = e4bSource
+        
+        // Strip preamble from optimized source
+        let optStripped = optimizedSource
+            .replacingOccurrences(of: "#include <metal_stdlib>\n", with: "")
+            .replacingOccurrences(of: "using namespace metal;\n", with: "")
+        result += "\n" + optStripped
+        
+        // Strip preamble from fusion source
+        let fusionStripped = fusionSource
+            .replacingOccurrences(of: "#include <metal_stdlib>\n", with: "")
+            .replacingOccurrences(of: "using namespace metal;\n", with: "")
+        result += "\n" + fusionStripped
+        
+        // Strip preamble from fused kernels source
+        let fusedStripped = fusedKernelsSource
+            .replacingOccurrences(of: "#include <metal_stdlib>\n", with: "")
+            .replacingOccurrences(of: "using namespace metal;\n", with: "")
+        result += "\n" + fusedStripped
+        
+        return result
+    }
+}