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
84 lines
3.5 KiB
Swift
84 lines
3.5 KiB
Swift
import XCTest
|
|
@testable import MarkBase
|
|
|
|
class E4BMarkBaseTest: XCTestCase {
|
|
|
|
func testE4BTextPerformance() throws {
|
|
print("\n═══════════════════════════════════════════════════════════════════")
|
|
print(" E4B-MarkBase TEXT Performance Test")
|
|
print("═══════════════════════════════════════════════════════════════════\n")
|
|
|
|
let modelPath = "/Users/accusys/MarkBaseEngine/models/E4B-MarkBase"
|
|
|
|
guard FileManager.default.fileExists(atPath: modelPath) else {
|
|
print("⚠ Model not found")
|
|
return
|
|
}
|
|
|
|
// Check scales quality
|
|
print("1. Scales Quality Check:")
|
|
let reader = try SafeTensorsReader(path: "\(modelPath)/model.safetensors")
|
|
let tensors = reader.allTensors
|
|
|
|
let embedScales = tensors.first { $0.name.contains("embed_tokens.scales") }
|
|
if let s = embedScales {
|
|
let data = try reader.read(tensor: s)
|
|
let scales = data.withUnsafeBytes { ptr in
|
|
Array(ptr.assumingMemoryBound(to: Float.self).prefix(20))
|
|
}
|
|
print(" Scales shape: \(s.shape), dtype: \(s.dtype)")
|
|
print(" Sample: \(scales)")
|
|
let negCount = scales.filter { $0 < 0 }.count
|
|
print(" Negative: \(negCount)")
|
|
}
|
|
|
|
// Test TEXT inference
|
|
print("\n2. TEXT Inference Test:")
|
|
let engine = try MarkBaseEngine(autoCompile: true)
|
|
let model = try E4BModel(modelDir: modelPath, engine: engine, maxContextLength: 128)
|
|
print(" ✓ Model loaded (Layers: \(model.numHiddenLayers), Hidden: \(model.hiddenSize))")
|
|
|
|
// Warmup
|
|
_ = try model.forwardOptimized(tokenId: 2, position: 0)
|
|
|
|
// Test NaN
|
|
print("\n3. NaN Test (tokenIds 0-5):")
|
|
var nanCount = 0
|
|
for tokenId in 0..<5 {
|
|
let result = try model.forwardOptimized(tokenId: tokenId, position: 0)
|
|
let nans = result.filter { $0.isNaN }.count
|
|
if nans > 0 { nanCount += nans }
|
|
}
|
|
print(" NaN count: \(nanCount)")
|
|
|
|
// Test speed
|
|
print("\n4. Speed Test (10 tokens):")
|
|
let testStart = Date()
|
|
var currentToken = 2
|
|
for i in 0..<10 {
|
|
let result = try model.forwardOptimized(tokenId: currentToken, position: i)
|
|
var maxIdx = 0
|
|
var maxVal = result[0]
|
|
for j in 1..<result.count {
|
|
if result[j] > maxVal {
|
|
maxVal = result[j]
|
|
maxIdx = j
|
|
}
|
|
}
|
|
currentToken = maxIdx
|
|
}
|
|
let testTime = Date().timeIntervalSince(testStart) * 1000
|
|
let avgTime = testTime / 10.0
|
|
|
|
print(" Average: \(String(format: "%.1f", avgTime))ms per token")
|
|
print(" Speed: \(String(format: "%.1f", 1000.0 / avgTime)) tok/s")
|
|
|
|
if avgTime < 100 && nanCount == 0 {
|
|
print("\n✓✓✓ E4B-MarkBase PRODUCTION READY")
|
|
} else {
|
|
print("\n⚠ Issues detected")
|
|
}
|
|
|
|
print("\n═══════════════════════════════════════════════════════════════════")
|
|
}
|
|
} |