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
91 lines
4.6 KiB
Swift
91 lines
4.6 KiB
Swift
import XCTest
|
|
@testable import MarkBase
|
|
|
|
class A4BComparisonTest: XCTestCase {
|
|
|
|
func testCompareEmbeddingTensors() throws {
|
|
print("\n═══════════════════════════════════════════════════════════════════")
|
|
print(" 26B-A4B vs 26B-Standard Embedding Tensor Comparison")
|
|
print("═══════════════════════════════════════════════════════════════════\n")
|
|
|
|
// Load 26B-A4B
|
|
let a4bPath = "/Users/accusys/MarkBaseEngine/models/gemma-4-26b-a4b-it-4bit"
|
|
let a4bIndex = try SafeTensorsIndex(modelDir: a4bPath)
|
|
var a4bReaders: [String: SafeTensorsReader] = [:]
|
|
for shardFile in a4bIndex.weightMap.values {
|
|
if a4bReaders[shardFile] == nil {
|
|
a4bReaders[shardFile] = try SafeTensorsReader(path: "\(a4bPath)/\(shardFile)")
|
|
}
|
|
}
|
|
let a4bTensors = a4bReaders.values.flatMap { $0.allTensors }
|
|
|
|
// Load 26B-Standard (single file)
|
|
let standardPath = "/Users/accusys/MarkBaseEngine/models/gemma-4-26b-standard"
|
|
let standardReader = try SafeTensorsReader(path: "\(standardPath)/model.safetensors")
|
|
let standardTensors = standardReader.allTensors
|
|
|
|
print("26B-A4B tensors: \(a4bTensors.count)")
|
|
print("26B-Standard tensors: \(standardTensors.count)")
|
|
|
|
// Compare embed_tokens
|
|
print("\nEmbedding tensor comparison:")
|
|
|
|
// A4B
|
|
let a4bWeight = a4bTensors.first { $0.name.contains("embed_tokens.weight") }
|
|
let a4bScales = a4bTensors.first { $0.name.contains("embed_tokens.scales") }
|
|
let a4bBiases = a4bTensors.first { $0.name.contains("embed_tokens.biases") }
|
|
|
|
print("\n26B-A4B:")
|
|
if let w = a4bWeight { print(" weight: \(w.shape), \(w.dtype)") }
|
|
if let s = a4bScales { print(" scales: \(s.shape), \(s.dtype)") }
|
|
if let b = a4bBiases { print(" biases: \(b.shape), \(b.dtype)") }
|
|
|
|
// Standard
|
|
let standardWeight = standardTensors.first { $0.name.contains("embed_tokens.weight") }
|
|
let standardScales = standardTensors.first { $0.name.contains("embed_tokens.scales") }
|
|
let standardBiases = standardTensors.first { $0.name.contains("embed_tokens.biases") }
|
|
|
|
print("\n26B-Standard:")
|
|
if let w = standardWeight { print(" weight: \(w.shape), \(w.dtype)") }
|
|
if let s = standardScales { print(" scales: \(s.shape), \(s.dtype)") }
|
|
if let b = standardBiases { print(" biases: \(b.shape), \(b.dtype)") }
|
|
|
|
// Compare first few weight values (if uint32, sample them)
|
|
if let w = a4bWeight, w.dtype == .u32 {
|
|
let reader = a4bReaders[a4bIndex.weightMap[w.name] ?? "model-00001-of-00003.safetensors"]!
|
|
let data = try reader.read(tensor: w)
|
|
let uint32s = data.withUnsafeBytes { ptr in
|
|
Array(ptr.bindMemory(to: UInt32.self).prefix(20))
|
|
}
|
|
print("\n26B-A4B weight sample (first 20 uint32): \(uint32s)")
|
|
}
|
|
|
|
if let w = standardWeight, w.dtype == .u32 {
|
|
let data = try standardReader.read(tensor: w)
|
|
let uint32s = data.withUnsafeBytes { ptr in
|
|
Array(ptr.bindMemory(to: UInt32.self).prefix(20))
|
|
}
|
|
print("\n26B-Standard weight sample (first 20 uint32): \(uint32s)")
|
|
}
|
|
|
|
// Compare scales
|
|
if let s = a4bScales {
|
|
let reader = a4bReaders[a4bIndex.weightMap[s.name] ?? "model-00001-of-00003.safetensors"]!
|
|
let data = try reader.read(tensor: s)
|
|
let scales = data.withUnsafeBytes { ptr in
|
|
Array(ptr.assumingMemoryBound(to: Float.self).prefix(20))
|
|
}
|
|
print("\n26B-A4B scales sample (first 20): \(scales)")
|
|
}
|
|
|
|
if let s = standardScales {
|
|
let data = try standardReader.read(tensor: s)
|
|
let scales = data.withUnsafeBytes { ptr in
|
|
Array(ptr.assumingMemoryBound(to: Float.self).prefix(20))
|
|
}
|
|
print("\n26B-Standard scales sample (first 20): \(scales)")
|
|
}
|
|
|
|
print("\n═══════════════════════════════════════════════════════════════════")
|
|
}
|
|
} |