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
77 lines
4.0 KiB
Swift
77 lines
4.0 KiB
Swift
import XCTest
|
|
@testable import MarkBase
|
|
|
|
final class G12BLoadingTests: XCTestCase {
|
|
|
|
func testModelConfigLoading() throws {
|
|
print("\n═══════════════════════════════════════")
|
|
print(" 12B Model Config Loading Test")
|
|
print("═══════════════════════════════════════\n")
|
|
|
|
let modelDir = "/Users/accusys/.cache/huggingface/hub/models--mlx-community--gemma-4-12B-it-4bit/snapshots/73bcf09092aa277861d5a191b989b666f7f32e8f"
|
|
|
|
print("Loading config from: \(modelDir)")
|
|
let cfg = try ModelConfig.load(from: modelDir)
|
|
|
|
print("\nConfig loaded successfully:")
|
|
print(" hidden_size: \(cfg.hiddenSize ?? 0)")
|
|
print(" num_layers: \(cfg.numHiddenLayers ?? 0)")
|
|
print(" num_heads: \(cfg.numAttentionHeads ?? 0)")
|
|
print(" num_kv_heads: \(cfg.numKeyValueHeads ?? 0)")
|
|
print(" intermediate_size: \(cfg.intermediateSize ?? 0)")
|
|
print(" head_dim: \(cfg.headDim ?? 0)")
|
|
print(" vocab_size: \(cfg.vocabSize ?? 0)")
|
|
print(" num_kv_shared: \(cfg.numKvSharedLayers ?? 0)")
|
|
print(" sliding_window: \(cfg.slidingWindow ?? 0)")
|
|
print(" max_position: \(cfg.maxPositionEmbeddings ?? 0)")
|
|
|
|
// Verify 12B parameters
|
|
XCTAssertEqual(cfg.hiddenSize, 3840, "12B should have hidden_size=3840")
|
|
XCTAssertEqual(cfg.numHiddenLayers, 48, "12B should have 48 layers")
|
|
XCTAssertEqual(cfg.numAttentionHeads, 16, "12B should have 16 heads")
|
|
XCTAssertEqual(cfg.numKeyValueHeads, 8, "12B should have 8 kv_heads")
|
|
XCTAssertEqual(cfg.numKvSharedLayers, 0, "12B should have NO KV sharing")
|
|
XCTAssertEqual(cfg.slidingWindow, 1024, "12B sliding_window should be 1024")
|
|
|
|
print("\n═══════════════════════════════════════")
|
|
print("✓ Config validation passed")
|
|
print("═══════════════════════════════════════\n")
|
|
}
|
|
|
|
func testShardDetection() throws {
|
|
print("\n═══════════════════════════════════════")
|
|
print(" 12B Shard Detection Test")
|
|
print("═══════════════════════════════════════\n")
|
|
|
|
let modelDir = "/Users/accusys/.cache/huggingface/hub/models--mlx-community--gemma-4-12B-it-4bit/snapshots/73bcf09092aa277861d5a191b989b666f7f32e8f"
|
|
|
|
// Check for index file
|
|
let indexPath = "\(modelDir)/model.safetensors.index.json"
|
|
let hasIndex = FileManager.default.fileExists(atPath: indexPath)
|
|
print("Index file exists: \(hasIndex) ✓")
|
|
|
|
if hasIndex {
|
|
// Load index
|
|
let index = try SafeTensorsIndex(modelDir: modelDir)
|
|
print("\nShard files found:")
|
|
for shard in index.shardFiles.sorted() {
|
|
print(" - \(shard)")
|
|
}
|
|
|
|
print("\nTotal tensors: \(index.allTensors.count)")
|
|
print("Sample tensors:")
|
|
let sampleTensors = index.allTensors.prefix(5)
|
|
for tensor in sampleTensors {
|
|
if let shard = index.weightMap[tensor] {
|
|
print(" \(tensor) → \(shard)")
|
|
}
|
|
}
|
|
|
|
XCTAssertGreaterThan(index.shardFiles.count, 1, "12B should have multiple shards")
|
|
}
|
|
|
|
print("\n═══════════════════════════════════════")
|
|
print("✓ Shard detection passed")
|
|
print("═══════════════════════════════════════\n")
|
|
}
|
|
} |