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 +1,117 @@
+import XCTest
+@testable import MarkBase
+
+final class G12BGeneratorTests: XCTestCase {
+    
+    func testSampler() throws {
+        print("\n═══════════════════════════════════════")
+        print("      Sampler Test")
+        print("═══════════════════════════════════════\n")
+        
+        print("Step 1: Test greedy sampling...")
+        let sampler = Sampler()
+        
+        // Create test logits
+        let logits: [Float] = [1.0, 2.0, 3.0, 0.5, 1.5]
+        
+        let greedyToken = sampler.greedySample(logits: logits)
+        print("  Logits: \(logits)")
+        print("  Greedy token: \(greedyToken)")
+        XCTAssertEqual(greedyToken, 2, "Greedy should pick index 2 (max=3.0)")
+        
+        print("\nStep 2: Test temperature sampling...")
+        // High temperature = more random
+        let highTempToken = sampler.sample(logits: logits, temperature: 2.0)
+        print("  Temperature=2.0 token: \(highTempToken)")
+        
+        // Low temperature = more deterministic
+        let lowTempToken = sampler.sample(logits: logits, temperature: 0.1)
+        print("  Temperature=0.1 token: \(lowTempToken)")
+        XCTAssertEqual(lowTempToken, 2, "Low temperature should be near greedy")
+        
+        print("\nStep 3: Test top-k sampling...")
+        let topKToken = sampler.sample(logits: logits, temperature: 1.0, topK: 2)
+        print("  TopK=2 token: \(topKToken)")
+        // Should be in top 2 indices (2 or 4)
+        XCTAssert(topKToken == 2 || topKToken == 4, "Should be in top-2")
+        
+        print("\nStep 4: Test top-p sampling...")
+        let topPToken = sampler.sample(logits: logits, temperature: 1.0, topP: 0.8)
+        print("  TopP=0.8 token: \(topPToken)")
+        
+        print("\n═══════════════════════════════════════")
+        print("✓ Sampler test passed")
+        print("═══════════════════════════════════════\n")
+    }
+    
+    func testStreamingGenerator() async throws {
+        print("\n═══════════════════════════════════════")
+        print("      Streaming Generator Test")
+        print("═══════════════════════════════════════\n")
+        
+        let modelDir = "/Users/accusys/.cache/huggingface/hub/models--mlx-community--gemma-4-12B-it-4bit/snapshots/73bcf09092aa277861d5a191b989b666f7f32e8f"
+        
+        print("Step 1: Load model and tokenizer...")
+        let engine = try MarkBaseEngine(autoCompile: true)
+        let model = try E4BModel(modelDir: modelDir, engine: engine, maxContextLength: 512)
+        let tokenizer = try TokenizerFactory.load(modelDir: modelDir)
+        
+        let generator = StreamingGenerator(model: model, tokenizer: tokenizer, engine: engine)
+        print("  ✓ Generator created")
+        
+        print("\nStep 2: Test complete generation...")
+        let config = GenerationConfig(maxTokens: 5, temperature: 0.8)
+        let response = try generator.generateComplete(prompt: "Hello", config: config)
+        print("  Prompt: Hello")
+        print("  Response: \(response)")
+        XCTAssertGreaterThan(response.count, 0, "Should generate text")
+        
+        print("\nStep 3: Test streaming generation...")
+        print("  Streaming tokens:")
+        var streamedTokens: [String] = []
+        
+        let streamConfig = GenerationConfig(maxTokens: 5, temperature: 1.0)
+        for await tokenText in generator.generate(prompt: "Test", config: streamConfig) {
+            print("    Token: \(tokenText)")
+            streamedTokens.append(tokenText)
+        }
+        
+        XCTAssertGreaterThan(streamedTokens.count, 0, "Should stream tokens")
+        
+        print("\n═══════════════════════════════════════")
+        print("✓ Streaming generator test passed")
+        print("═══════════════════════════════════════\n")
+    }
+    
+    func testGenerationPerformance() throws {
+        print("\n═══════════════════════════════════════")
+        print("      Generation Performance Test")
+        print("═══════════════════════════════════════\n")
+        
+        let modelDir = "/Users/accusys/.cache/huggingface/hub/models--mlx-community--gemma-4-12B-it-4bit/snapshots/73bcf09092aa277861d5a191b989b666f7f32e8f"
+        
+        print("Step 1: Load model...")
+        let engine = try MarkBaseEngine(autoCompile: true)
+        let model = try E4BModel(modelDir: modelDir, engine: engine, maxContextLength: 512)
+        let tokenizer = try TokenizerFactory.load(modelDir: modelDir)
+        let generator = StreamingGenerator(model: model, tokenizer: tokenizer, engine: engine)
+        
+        print("\nStep 2: Generate 20 tokens...")
+        let config = GenerationConfig(maxTokens: 20, temperature: 0.8)
+        
+        let start = Date()
+        let response = try generator.generateComplete(prompt: "Hello", config: config)
+        let duration = Date().timeIntervalSince(start)
+        
+        print("  Tokens generated: ~20")
+        print("  Time: \(duration) seconds")
+        print("  Speed: \(20.0 / duration) tokens/second")
+        print("  Response: \(response)")
+        
+        XCTAssertGreaterThan(response.count, 0)
+        
+        print("\n═══════════════════════════════════════")
+        print("✓ Generation performance test passed")
+        print("═══════════════════════════════════════\n")
+    }
+}