MarkBase Admin
ac75faa0cc
CI / build-and-test (push) Has been cancelled
- 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
54 lines
1.9 KiB
Swift
54 lines
1.9 KiB
Swift
import Foundation
|
|
import MarkBase
|
|
|
|
let modelDir = "/Users/accusys/MarkBaseEngine/models/gemma-4-31b-it-4bit"
|
|
|
|
guard FileManager.default.fileExists(atPath: modelDir + "/config.json") else {
|
|
print("✗ 31B model not found at \(modelDir)")
|
|
exit(1)
|
|
}
|
|
|
|
print("Loading engine...")
|
|
let engine = try E4BEngine(autoCompile: true)
|
|
print("✓ Engine created")
|
|
|
|
print("\nLoading 31B model (~18 GB, may take 2-3 minutes)...")
|
|
let model = try E4BModel(modelDir: modelDir, engine: engine, maxContextLength: 128)
|
|
|
|
print("✓ Model loaded!")
|
|
print(" Layers: \(model.numHiddenLayers)")
|
|
print(" Hidden size: \(model.hiddenSize)")
|
|
print(" Vocab size: \(model.vocabSize)")
|
|
print(" Layer types: \(model.layerTypesIsFull.count) total, \(model.layerTypesIsFull.filter { $0 }.count) full, \(model.layerTypesIsFull.filter { !$0 }.count) sliding")
|
|
|
|
print("\nLoading tokenizer...")
|
|
let tokenizer = try TokenizerFactory.load(modelDir: modelDir)
|
|
print("✓ Tokenizer loaded")
|
|
|
|
// Test forward pass
|
|
print("\n=== Forward pass test ===")
|
|
let tokens = tokenizer.encode(text: "Hello")
|
|
print("Tokenized 'Hello': \(tokens)")
|
|
|
|
let logits = try model.forward(tokenId: tokens[0], position: 0)
|
|
print("Forward pass complete: \(logits.count) logits")
|
|
let maxLogit = logits.max() ?? -999
|
|
print("Max logit: \(maxLogit)")
|
|
|
|
let sorted = logits.enumerated().sorted { $0.element > $1.element }
|
|
let top5 = sorted.prefix(5)
|
|
print("Top 5 predictions:")
|
|
for (idx, val) in top5 {
|
|
let tokenStr = tokenizer.decode(tokens: [idx])
|
|
print(" Token \(idx) ('\(tokenStr)'): \(val)")
|
|
}
|
|
|
|
// Generation test
|
|
print("\n=== Generation test ===")
|
|
let genConfig = GenerationConfig(maxTokens: 5, temperature: 0.7)
|
|
let generator = StreamingGenerator(model: model, tokenizer: tokenizer, engine: engine)
|
|
let response = try generator.generateComplete(prompt: "Hello", config: genConfig)
|
|
print("Generated: '\(response)'")
|
|
|
|
print("\n✅ 31B test complete!")
|