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 +1,109 @@
+import XCTest
+@testable import MarkBase
+
+final class MoERouterOnlyTest: XCTestCase {
+    
+    func testRouterProjectionOnly() throws {
+        print("\n═══════════════════════════════════════")
+        print("      Router Projection Only Test")
+        print("═══════════════════════════════════════\n")
+        
+        let modelDir = "/Users/accusys/MarkBaseEngine/models/gemma-4-26b-a4b-it-4bit"
+        
+        print("Step 1: Load model...")
+        let engine = try MarkBaseEngine(autoCompile: true)
+        let model = try E4BModel(modelDir: modelDir, engine: engine, maxContextLength: 32)
+        print("  ✓ Model loaded")
+        
+        print("\nStep 2: Get router from layer 0...")
+        let layer0 = model.layers[0]
+        
+        guard let router = layer0.routerProj else {
+            XCTFail("Router not present")
+            return
+        }
+        print("  ✓ Router retrieved")
+        print("    Router bits: \(router.bits)")
+        print("    Router inDim: \(router.inDim)")
+        print("    Router outDim: \(router.outDim)")
+        
+        print("\nStep 3: Create minimal buffers...")
+        let hs = model.hiddenSize
+        
+        // Create input buffer (normalized layer output)
+        let inputBuffer = engine.device.makeBuffer(length: hs * MemoryLayout<Float>.size)!
+        var inputData = [Float](repeating: 0.1, count: hs)  // Small values to avoid overflow
+        memcpy(inputBuffer.contents(), &inputData, inputData.count * MemoryLayout<Float>.size)
+        
+        // Create output buffer for router logits
+        let routerOutDim = router.outDim  // 128 experts
+        let routerOutputBuffer = engine.device.makeBuffer(length: routerOutDim * MemoryLayout<Float>.size)!
+        
+        print("  ✓ Buffers created")
+        print("    Input: \(hs) floats")
+        print("    Output: \(routerOutDim) floats (expert scores)")
+        
+        print("\nStep 4: Test router projection alone...")
+        let start = Date()
+        
+        do {
+            print("  Calling quantizedMatmul...")
+            fflush(stdout)
+            
+            let cmdBuf = engine.commandQueue.makeCommandBuffer()!
+            try layer0.quantizedMatmul(
+                engine: engine,
+                cmdBuf: cmdBuf,
+                input: inputBuffer,
+                weights: router,
+                output: routerOutputBuffer
+            )
+            
+            print("  Command buffer created, committing...")
+            fflush(stdout)
+            
+            cmdBuf.commit()
+            print("  Waiting for completion...")
+            fflush(stdout)
+            
+            cmdBuf.waitUntilCompleted()
+            
+            let elapsed = Date().timeIntervalSince(start)
+            print("  ✓ Router projection completed in \(String(format: "%.3f", elapsed))s")
+            print("    Command buffer status: \(cmdBuf.status.rawValue)")
+            
+            if elapsed > 5.0 {
+                print("  ⚠️ Router took >5s - potential issue")
+            }
+            
+            if cmdBuf.status != .completed {
+                let errorMsg = cmdBuf.error?.localizedDescription ?? "unknown"
+                print("  ❌ Router command failed: \(errorMsg)")
+                XCTFail("Router projection failed")
+                return
+            }
+            
+            // Read router output
+            let routerOutput = engine.readFloats(from: routerOutputBuffer, count: routerOutDim)
+            print("\n  Router logits (first 10): \(routerOutput[0..<10])")
+            print("  Router logits max: \(routerOutput.max() ?? 0)")
+            print("  Router logits min: \(routerOutput.min() ?? 0)")
+            
+            let hasNaN = routerOutput.contains { $0.isNaN }
+            if hasNaN {
+                print("  ❌ NaN detected in router output!")
+                XCTFail("Router output has NaN")
+            } else {
+                print("  ✓ Router output valid (no NaN)")
+            }
+            
+        } catch {
+            print("  ❌ Router projection failed: \(error)")
+            throw error
+        }
+        
+        print("\n═══════════════════════════════════════")
+        print("✓ Router projection works!")
+        print("═══════════════════════════════════════\n")
+    }
+}