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 +1,82 @@
+import XCTest
+@testable import MarkBase
+
+class Model12BVisionCheckTest: XCTestCase {
+    
+    func test12BVisionWeights() throws {
+        print("\n═══════════════════════════════════════════════════════════════════")
+        print("  12B Standard Vision Weights 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
+        }
+        
+        // Check shard 2 (VisionTower12B.load uses model-00002-of-00002.safetensors)
+        let shard2Path = "\(modelPath)/model-00002-of-00002.safetensors"
+        guard FileManager.default.fileExists(atPath: shard2Path) else {
+            print("⚠ Shard 2 not found")
+            return
+        }
+        
+        let reader = try SafeTensorsReader(path: shard2Path)
+        let tensors = reader.allTensors
+        
+        print("1. Shard 2 Total Tensors: \(tensors.count)")
+        
+        // Check vision_embedder weights
+        print("\n2. Vision Embedder Weights:")
+        let visionEmbedderTensors = tensors.filter { $0.name.contains("vision_embedder") }
+        print("  vision_embedder tensors: \(visionEmbedderTensors.count)")
+        for tensor in visionEmbedderTensors {
+            print("    \(tensor.name): shape=\(tensor.shape), dtype=\(tensor.dtype)")
+        }
+        
+        // Check embed_vision weights
+        print("\n3. Embed Vision Weights:")
+        let embedVisionTensors = tensors.filter { $0.name.contains("embed_vision") }
+        print("  embed_vision tensors: \(embedVisionTensors.count)")
+        for tensor in embedVisionTensors {
+            print("    \(tensor.name): shape=\(tensor.shape), dtype=\(tensor.dtype)")
+        }
+        
+        // Check audio weights
+        print("\n4. Audio Weights:")
+        let audioTensors = tensors.filter { $0.name.contains("audio") }
+        print("  audio tensors: \(audioTensors.count)")
+        for tensor in audioTensors.prefix(10) {
+            print("    \(tensor.name): shape=\(tensor.shape)")
+        }
+        
+        // Summary
+        print("\n═══════════════════════════════════════════════════════════════════")
+        print("  Vision Weights Summary")
+        print("═══════════════════════════════════════════════════════════════════\n")
+        
+        if visionEmbedderTensors.count > 0 {
+            print("✓ 12B HAS VISION EMBEDDER: \(visionEmbedderTensors.count) tensors")
+        } else {
+            print("✗ 12B NO VISION EMBEDDER")
+        }
+        
+        if embedVisionTensors.count > 0 {
+            print("✓ 12B HAS EMBED VISION: \(embedVisionTensors.count) tensors")
+        } else {
+            print("✗ 12B NO EMBED VISION")
+        }
+        
+        if audioTensors.count > 0 {
+            print("✓ 12B HAS AUDIO: \(audioTensors.count) tensors")
+        } else {
+            print("✗ 12B NO AUDIO")
+        }
+        
+        print("\nVision Capability: \(visionEmbedderTensors.count > 0 && embedVisionTensors.count > 0 ? "YES ✓" : "NO ✗")")
+        print("Audio Capability: \(audioTensors.count > 0 ? "YES ✓" : "NO ✗")")
+        
+        print("\n═══════════════════════════════════════════════════════════════════")
+    }
+}