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
83 lines
4.1 KiB
Swift
83 lines
4.1 KiB
Swift
import XCTest
|
||
@testable import MarkBase
|
||
|
||
final class OptimizationVerificationTest: XCTestCase {
|
||
|
||
func testOptimizationEffect() throws {
|
||
print("\n═══════════════════════════════════════════════════════════════════")
|
||
print(" Optimization Verification Test")
|
||
print("═══════════════════════════════════════════════════════════════════\n")
|
||
|
||
let modelDir = "/Users/accusys/MarkBaseEngine/models/E4B-MarkBase"
|
||
let engine = try MarkBaseEngine(autoCompile: true)
|
||
let textModel = try E4BModel(modelDir: modelDir, engine: engine, maxContextLength: 256)
|
||
|
||
print("Model loaded: \(textModel.numHiddenLayers) layers\n")
|
||
|
||
// Warm up shaders (run once to compile)
|
||
print("Warm up Metal shaders...")
|
||
_ = try textModel.forward(tokenId: 2, position: 0)
|
||
_ = try textModel.forwardOptimized(tokenId: 2, position: 0)
|
||
print(" ✓ Shaders compiled\n")
|
||
|
||
// Test original forward (10 tokens)
|
||
print("Test 1: Original forward (10 tokens)...")
|
||
var tokens1: [Int] = [2]
|
||
let start1 = Date()
|
||
for _ in 0..<10 {
|
||
let logits = try textModel.forward(tokenId: tokens1.last!, position: tokens1.count - 1)
|
||
var maxIdx = 0
|
||
var maxLogit = logits[0]
|
||
for i in 1..<logits.count {
|
||
if logits[i] > maxLogit { maxLogit = logits[i]; maxIdx = i }
|
||
}
|
||
tokens1.append(maxIdx)
|
||
}
|
||
let time1 = Date().timeIntervalSince(start1) * 1000
|
||
print(" ✓ Time: \(time1) ms (avg \(time1/10) ms/token)")
|
||
|
||
// Test optimized forward (10 tokens)
|
||
print("\nTest 2: Optimized forward (10 tokens)...")
|
||
var tokens2: [Int] = [2]
|
||
let start2 = Date()
|
||
for _ in 0..<10 {
|
||
let logits = try textModel.forwardOptimized(tokenId: tokens2.last!, position: tokens2.count - 1)
|
||
var maxIdx = 0
|
||
var maxLogit = logits[0]
|
||
for i in 1..<logits.count {
|
||
if logits[i] > maxLogit { maxLogit = logits[i]; maxIdx = i }
|
||
}
|
||
tokens2.append(maxIdx)
|
||
}
|
||
let time2 = Date().timeIntervalSince(start2) * 1000
|
||
print(" ✓ Time: \(time2) ms (avg \(time2/10) ms/token)")
|
||
|
||
// Comparison
|
||
print("\nResults:")
|
||
let speedup = time1 / time2
|
||
let improvement = (time1 - time2) / time1 * 100
|
||
print(" Speedup: \(speedup)x")
|
||
print(" Improvement: \(improvement)%")
|
||
print(" Time saved: \(time1 - time2) ms")
|
||
|
||
// Debug: check which path is being used
|
||
print("\nDebug info:")
|
||
print(" Dense layers should have: 0 waits per layer")
|
||
print(" Original: \(textModel.numHiddenLayers) layers × 1 wait = \(textModel.numHiddenLayers) waits")
|
||
print(" Optimized: 1 wait total (at end)")
|
||
print(" Expected speedup: \(textModel.numHiddenLayers)x")
|
||
print(" Actual speedup: \(speedup)x")
|
||
|
||
if speedup < 2.0 {
|
||
print("\n⚠ Warning: Optimization not working as expected!")
|
||
print(" Possible issues:")
|
||
print(" 1. Layer.forwardOptimized not being called")
|
||
print(" 2. Internal helper functions creating their own command buffers")
|
||
print(" 3. MoE layers being used (require router read)")
|
||
}
|
||
|
||
print("\n═══════════════════════════════════════════════════════════════════")
|
||
print("✓ Verification test complete")
|
||
print("═══════════════════════════════════════════════════════════════════\n")
|
||
}
|
||
} |