warren/markbaseengine
1
0
Fork 0
Code Issues Pull Requests Actions 15 Packages Projects Releases Wiki Activity

Initial commit: E4B-MarkBase model integration with passing tests
CI / build-and-test (push) Has been cancelled
Details

Browse Source 
- 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
This commit is contained in:
MarkBase Admin
2026-06-23 18:12:35 +08:00
commit ac75faa0cc
301 changed files with 63426 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Tests/MarkBaseTests/OptimizedForwardTest.swift
+124
View File
@@ -0,0 +1,124 @@
import XCTest
@testable import MarkBase
final class OptimizedForwardTest: XCTestCase {
func testE4BOptimizedForward() throws {
print("\n═══════════════════════════════════════════════════════════════════")
print(" E4B Optimized Forward Pass Test")
print("═══════════════════════════════════════════════════════════════════\n")
let modelDir = "/Users/accusys/MarkBaseEngine/models/E4B-MarkBase"
print("Step 1: Load TEXT model...")
let engine = try MarkBaseEngine(autoCompile: true)
let loadStart = Date()
let textModel = try E4BModel(modelDir: modelDir, engine: engine, maxContextLength: 256)
let loadTime = Date().timeIntervalSince(loadStart) * 1000
print(" ✓ Model loaded in \(loadTime) ms")
print(" Hidden: \(textModel.hiddenSize), Layers: \(textModel.numHiddenLayers)")
print("\nStep 2: Test ORIGINAL forward (baseline)...")
var tokens1: [Int] = [2]
let start1 = Date()
let logits1 = try textModel.forward(tokenId: 2, position: 0)
let time1 = Date().timeIntervalSince(start1) * 1000
print(" ✓ Original forward: \(time1) ms")
print(" Logits: max=\(logits1.max() ?? 0), min=\(logits1.min() ?? 0)")
print(" Top 5: \(logits1.enumerated().sorted(by: { $0.element > $1.element }).prefix(5).map { "\($0.offset): \($0.element)" })")
print("\nStep 3: Test OPTIMIZED forward...")
let start2 = Date()
let logits2 = try textModel.forwardOptimized(tokenId: 2, position: 0)
let time2 = Date().timeIntervalSince(start2) * 1000
print(" ✓ Optimized forward: \(time2) ms")
print(" Logits: max=\(logits2.max() ?? 0), min=\(logits2.min() ?? 0)")
print(" Top 5: \(logits2.enumerated().sorted(by: { $0.element > $1.element }).prefix(5).map { "\($0.offset): \($0.element)" })")
print("\nStep 4: Compare results...")
let speedup = time1 / time2
let maxDiff = abs(logits1.max()! - logits2.max()!)
let minDiff = abs(logits1.min()! - logits2.min()!)
print(" Speedup: \(speedup)x")
print(" Max difference: \(maxDiff)")
print(" Min difference: \(minDiff)")
// Compare top tokens
let top1 = logits1.enumerated().sorted(by: { $0.element > $1.element }).prefix(5).map { $0.offset }
let top2 = logits2.enumerated().sorted(by: { $0.element > $1.element }).prefix(5).map { $0.offset }
print(" Top 5 tokens match: \(top1 == top2)")
print("\nStep 5: Generate 10 tokens (optimized)...")
var tokens: [Int] = [2]
let genStart = Date()
for _ in 0..<10 {
let logits = try textModel.forwardOptimized(tokenId: tokens.last!, position: tokens.count - 1)
var maxIdx = 0
var maxLogit = logits[0]
for i in 1..<logits.count {
if logits[i] > maxLogit {
maxLogit = logits[i]
maxIdx = i
}
}
tokens.append(maxIdx)
}
let genTime = Date().timeIntervalSince(genStart) * 1000
print(" ✓ Generated 10 tokens in \(genTime) ms")
print(" Tokens: \(tokens.suffix(10))")
print(" Average token time: \(genTime / 10) ms")
print("\n═══════════════════════════════════════════════════════════════════")
print("✓ Optimized forward test passed")
print(" Load: \(loadTime) ms")
print(" Single forward: Original=\(time1)ms, Optimized=\(time2)ms")
print(" 10 tokens: \(genTime) ms (avg \(genTime/10) ms/token)")
print(" Speedup: \(speedup)x")
print("═══════════════════════════════════════════════════════════════════\n")
XCTAssertFalse(logits2.contains { $0.isNaN }, "Optimized logits should not have NaN")
XCTAssertLessThan(maxDiff, 1.0, "Logits should be similar")
}
func test12BOptimizedForward() throws {
print("\n═══════════════════════════════════════════════════════════════════")
print(" 12B Optimized Forward Pass Test")
print("═══════════════════════════════════════════════════════════════════\n")
let modelDir = "/Users/accusys/.cache/huggingface/hub/models--mlx-community--gemma-4-12B-it-4bit/snapshots/73bcf09092aa277861d5a191b989b666f7f32e8f"
print("Step 1: Load TEXT model...")
let engine = try MarkBaseEngine(autoCompile: true)
let loadStart = Date()
let textModel = try E4BModel(modelDir: modelDir, engine: engine, maxContextLength: 256)
let loadTime = Date().timeIntervalSince(loadStart) * 1000
print(" ✓ Model loaded in \(loadTime) ms")
print(" Hidden: \(textModel.hiddenSize), Layers: \(textModel.numHiddenLayers)")
print("\nStep 2: Generate 10 tokens (optimized)...")
var tokens: [Int] = [2]
let genStart = Date()
for _ in 0..<10 {
let logits = try textModel.forwardOptimized(tokenId: tokens.last!, position: tokens.count - 1)
var maxIdx = 0
var maxLogit = logits[0]
for i in 1..<logits.count {
if logits[i] > maxLogit {
maxLogit = logits[i]
maxIdx = i
}
}
tokens.append(maxIdx)
}
let genTime = Date().timeIntervalSince(genStart) * 1000
print(" ✓ Generated 10 tokens in \(genTime) ms")
print(" Tokens: \(tokens.suffix(10))")
print(" Average token time: \(genTime / 10) ms")
print("\n═══════════════════════════════════════════════════════════════════")
print("✓ 12B Optimized forward test passed")
print(" Load: \(loadTime) ms")
print(" 10 tokens: \(genTime) ms (avg \(genTime/10) ms/token)")
print("═══════════════════════════════════════════════════════════════════\n")
}
}