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 +1,63 @@
+import XCTest
+@testable import MarkBase
+
+final class E2BAudioLoadTest: XCTestCase {
+    func testE2BAudioLoad() throws {
+        print("\n═══════════════════════════════════════")
+        print("      E2B AudioTower Loading Test")
+        print("═══════════════════════════════════════\n")
+        
+        let modelDir = "/Users/accusys/.cache/huggingface/hub/models--mlx-community--gemma-4-e2b-it-4bit/snapshots/2c3e507453b4f218d05fe3cc97bea5c5a654257e"
+        
+        print("Step 1: Load AudioConfig...")
+        let acfg = loadAudioConfig(modelDir: modelDir)
+        print("  hiddenSize: \(acfg.hiddenSize)")
+        print("  numHiddenLayers: \(acfg.numHiddenLayers)")
+        
+        print("\nStep 2: Create engine...")
+        let engine = try MarkBaseEngine(autoCompile: true)
+        
+        print("\nStep 3: Load safetensors...")
+        let reader = try SafeTensorsReader(path: modelDir + "/model.safetensors")
+        
+        print("\nStep 4: Load AudioTowerE2B...")
+        do {
+            let tower = try loadAudioTowerE2B(reader: reader, config: acfg, engine: engine)
+            print("  ✓ AudioTowerE2B loaded: \(tower.config.numHiddenLayers) layers")
+            
+            print("\nStep 5: Test forward pass...")
+            let seqLen = 50
+            let nMels = 128
+            var melFeatures: [Float] = Array(repeating: 0.1, count: seqLen * nMels)
+            for i in 0..<melFeatures.count { melFeatures[i] = Float(i % 100) / 100.0 * 0.5 }
+            
+            let inputBuffer = engine.device.makeBuffer(bytes: melFeatures, length: melFeatures.count * 4)!
+            let outputLen = seqLen / 4 * tower.config.outputProjDims
+            let outputBuffer = engine.device.makeBuffer(length: outputLen * 4)!
+            
+            try tower.forward(inputBuffer: inputBuffer, seqLen: seqLen, outputBuffer: outputBuffer)
+            
+            let ptr = outputBuffer.contents().assumingMemoryBound(to: Float.self)
+            var hasNaN = false
+            var maxVal: Float = 0, minVal: Float = 0
+            for i in 0..<outputLen {
+                let v = ptr[i]
+                if v.isNaN { hasNaN = true }
+                if v > maxVal { maxVal = v }
+                if v < minVal { minVal = v }
+            }
+            print("  Output shape: [\(seqLen/4), \(tower.config.outputProjDims)]")
+            print("  Range: [\(minVal), \(maxVal)]")
+            print("  NaN: \(hasNaN)")
+            
+            XCTAssertFalse(hasNaN, "E2B audio output should not have NaN")
+        } catch {
+            print("  ✗ Error: \(error)")
+            throw error
+        }
+        
+        print("\n═══════════════════════════════════════")
+        print("✓ E2B AudioTower test passed")
+        print("═══════════════════════════════════════\n")
+    }
+}