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
40 lines
1.6 KiB
Swift
40 lines
1.6 KiB
Swift
import XCTest
|
|
@testable import MarkBase
|
|
|
|
final class Layer4NaNTest: XCTestCase {
|
|
func testLayer4Expert127NaN() throws {
|
|
print("\n" + String(repeating: "=", count: 60))
|
|
print("Layer 4 Expert 127 NaN Debug Test")
|
|
print(String(repeating: "=", count: 60))
|
|
|
|
let modelDir = "/Users/accusys/MarkBaseEngine/models/gemma-4-26b-a4b-it-4bit"
|
|
let engine = try MarkBaseEngine(autoCompile: true)
|
|
let model = try E4BModel(modelDir: modelDir, engine: engine, maxContextLength: 128)
|
|
|
|
print("\nModel loaded successfully")
|
|
|
|
// Run one forward pass to trigger Layer 4 Expert 127
|
|
let inputIds: [Int32] = [1, 3234, 357, 659, 198]
|
|
print("\nRunning forward pass with input: \(inputIds)")
|
|
|
|
// Run multiple forward passes like generation
|
|
do {
|
|
for (index, tokenId) in inputIds.enumerated() {
|
|
print("\n--- Processing token \(index): \(tokenId) ---")
|
|
let logits = try model.forward(tokenId: Int(tokenId), position: index)
|
|
let logitsNaNCount = logits.filter { $0.isNaN }.count
|
|
print("Logits NaN count: \(logitsNaNCount)/\(logits.count)")
|
|
|
|
if logitsNaNCount > 0 {
|
|
print("ERROR: NaN detected in logits!")
|
|
break
|
|
}
|
|
}
|
|
print("\nAll forward passes completed successfully")
|
|
} catch {
|
|
print("Forward pass failed: \(error)")
|
|
}
|
|
|
|
print("\n" + String(repeating: "=", count: 60))
|
|
}
|
|
} |