mirror of
https://github.com/MLSysBook/TinyTorch.git
synced 2026-03-12 03:03:37 -05:00
Optimization Level 0: Baseline
Results: - Perceptron: ✅ (1.92s) 100.0% - XOR: ✅ (1.87s) 54.5% - MNIST: ✅ (1.96s) 11.5% - CIFAR: ❌ (60.00s) - TinyGPT: ✅ (1.92s)
This commit is contained in:
Vijay Janapa Reddi
@@ -14,3 +14,4 @@ Testing Optimization Level 0: Baseline
|
||||
[2025-09-28 21:42:40] ✅ Complete in 1.85s
|
||||
[2025-09-28 21:42:40]
|
||||
Committing results for Baseline...
|
||||
[2025-09-28 21:42:40] Committed results
|
||||
|
||||
19
optimization_log_20250928_214329.txt
Normal file
19
optimization_log_20250928_214329.txt
Normal file
@@ -0,0 +1,19 @@
|
||||
[2025-09-28 21:43:29] ================================================================================
|
||||
[2025-09-28 21:43:29] PHASE 2: OPTIMIZATION TESTING
|
||||
[2025-09-28 21:43:29] ================================================================================
|
||||
[2025-09-28 21:43:29]
|
||||
Testing Optimization Level 0: Baseline
|
||||
[2025-09-28 21:43:29] Description: No optimizations
|
||||
[2025-09-28 21:43:29] ------------------------------------------------------------
|
||||
[2025-09-28 21:43:29] Testing Perceptron with Baseline...
|
||||
[2025-09-28 21:43:31] ✅ Complete in 1.92s
|
||||
[2025-09-28 21:43:31] Testing XOR with Baseline...
|
||||
[2025-09-28 21:43:33] ✅ Complete in 1.87s
|
||||
[2025-09-28 21:43:33] Testing MNIST with Baseline...
|
||||
[2025-09-28 21:43:35] ✅ Complete in 1.96s
|
||||
[2025-09-28 21:43:35] Testing CIFAR with Baseline...
|
||||
[2025-09-28 21:44:05] ⏱️ Timeout after 60s
|
||||
[2025-09-28 21:44:05] Testing TinyGPT with Baseline...
|
||||
[2025-09-28 21:44:07] ✅ Complete in 1.92s
|
||||
[2025-09-28 21:44:07]
|
||||
Committing results for Baseline...
|
||||
@@ -41,7 +41,7 @@ class OptimizationTester:
|
||||
{
|
||||
'name': 'CIFAR',
|
||||
'path': 'examples/cifar_cnn_modern/train_cnn.py',
|
||||
'args': '--test-only', # Quick test for now
|
||||
'args': '--test-only', # Architecture test only
|
||||
'metrics': ['forward_pass', 'time']
|
||||
},
|
||||
{
|
||||
@@ -52,7 +52,7 @@ class OptimizationTester:
|
||||
}
|
||||
]
|
||||
|
||||
# Define optimization levels (modules 15-20)
|
||||
# Define optimization levels (modules 14-19 based on actual TinyTorch structure)
|
||||
self.optimizations = [
|
||||
{
|
||||
'level': 0,
|
||||
@@ -60,41 +60,41 @@ class OptimizationTester:
|
||||
'description': 'No optimizations',
|
||||
'module': None
|
||||
},
|
||||
{
|
||||
'level': 14,
|
||||
'name': 'Profiling',
|
||||
'description': 'Module 14: Performance profiling and analysis',
|
||||
'module': 'profiling'
|
||||
},
|
||||
{
|
||||
'level': 15,
|
||||
'name': 'Memory Optimization',
|
||||
'description': 'Module 15: Memory-efficient operations',
|
||||
'module': 'memory_opt'
|
||||
'name': 'Acceleration',
|
||||
'description': 'Module 15: Hardware acceleration optimizations',
|
||||
'module': 'acceleration'
|
||||
},
|
||||
{
|
||||
'level': 16,
|
||||
'name': 'Compute Optimization',
|
||||
'description': 'Module 16: Vectorization and parallelization',
|
||||
'module': 'compute_opt'
|
||||
'name': 'Quantization',
|
||||
'description': 'Module 16: Quantization and compression',
|
||||
'module': 'quantization'
|
||||
},
|
||||
{
|
||||
'level': 17,
|
||||
'name': 'Cache Optimization',
|
||||
'description': 'Module 17: Cache-friendly operations',
|
||||
'module': 'cache_opt'
|
||||
'name': 'Compression',
|
||||
'description': 'Module 17: Model compression techniques',
|
||||
'module': 'compression'
|
||||
},
|
||||
{
|
||||
'level': 18,
|
||||
'name': 'Kernel Fusion',
|
||||
'description': 'Module 18: Fused operations',
|
||||
'module': 'kernel_fusion'
|
||||
'name': 'Caching',
|
||||
'description': 'Module 18: Caching and memory optimization',
|
||||
'module': 'caching'
|
||||
},
|
||||
{
|
||||
'level': 19,
|
||||
'name': 'Mixed Precision',
|
||||
'description': 'Module 19: FP16/BF16 operations',
|
||||
'module': 'mixed_precision'
|
||||
},
|
||||
{
|
||||
'level': 20,
|
||||
'name': 'Full Optimization',
|
||||
'description': 'Module 20: All optimizations combined',
|
||||
'module': 'full_opt'
|
||||
'name': 'Benchmarking',
|
||||
'description': 'Module 19: Advanced benchmarking suite',
|
||||
'module': 'benchmarking'
|
||||
}
|
||||
]
|
||||
|
||||
@@ -118,13 +118,15 @@ class OptimizationTester:
|
||||
env['TINYTORCH_OPT'] = optimization['module']
|
||||
|
||||
try:
|
||||
# Use shorter timeout for CIFAR architecture test
|
||||
timeout_val = 30 if example['name'] == 'CIFAR' else 60
|
||||
cmd = f"python {example['path']} {example['args']}"
|
||||
result = subprocess.run(
|
||||
cmd,
|
||||
shell=True,
|
||||
capture_output=True,
|
||||
text=True,
|
||||
timeout=60,
|
||||
timeout=timeout_val,
|
||||
env=env
|
||||
)
|
||||
|
||||
@@ -275,19 +277,27 @@ class OptimizationTester:
|
||||
self.log("\nMatrix saved to optimization_matrix.md")
|
||||
|
||||
if __name__ == "__main__":
|
||||
import sys
|
||||
|
||||
tester = OptimizationTester()
|
||||
|
||||
# For now, just test baseline
|
||||
print("\nStarting with BASELINE performance testing...")
|
||||
baseline = tester.optimizations[0]
|
||||
baseline_results = tester.test_optimization_level(baseline)
|
||||
|
||||
# Save baseline results
|
||||
tester.commit_results(baseline, baseline_results)
|
||||
|
||||
print("\n" + "="*60)
|
||||
print("BASELINE TESTING COMPLETE")
|
||||
print("="*60)
|
||||
print("\nBaseline results committed.")
|
||||
print("Ready to proceed with optimization testing.")
|
||||
print("\nTo run full suite: python optimization_test_framework.py --full")
|
||||
|
||||
# Check if user wants full suite
|
||||
if '--full' in sys.argv:
|
||||
print("\n🚀 RUNNING FULL OPTIMIZATION TEST SUITE...")
|
||||
print("Testing all optimization levels: Baseline → Profiling → Acceleration → Quantization → Compression → Caching → Benchmarking")
|
||||
all_results = tester.run_full_test_suite()
|
||||
else:
|
||||
# Just test baseline
|
||||
print("\nStarting with BASELINE performance testing...")
|
||||
baseline = tester.optimizations[0]
|
||||
baseline_results = tester.test_optimization_level(baseline)
|
||||
|
||||
# Save baseline results
|
||||
tester.commit_results(baseline, baseline_results)
|
||||
|
||||
print("\n" + "="*60)
|
||||
print("BASELINE TESTING COMPLETE")
|
||||
print("="*60)
|
||||
print("\nBaseline results committed.")
|
||||
print("Ready to proceed with optimization testing.")
|
||||
print("\nTo run full suite: python optimization_test_framework.py --full")
|
||||
|
||||
@@ -1,24 +1,24 @@
|
||||
{
|
||||
"Perceptron": {
|
||||
"success": true,
|
||||
"time": 1.8552052974700928,
|
||||
"time": 1.924880027770996,
|
||||
"output_preview": "ion\n\n\ud83d\ude80 Next Steps:\n \u2022 Continue to XOR 1969 milestone after Module 06 (Autograd)\n \u2022 YOUR foundation enables solving non-linear problems!\n \u2022 With 100.0% accuracy, YOUR perceptron works perfectly!\n",
|
||||
"loss": 0.2038,
|
||||
"accuracy": 100.0
|
||||
},
|
||||
"XOR": {
|
||||
"success": true,
|
||||
"time": 1.9177570343017578,
|
||||
"time": 1.8728010654449463,
|
||||
"output_preview": "ayer networks\n\n\ud83d\ude80 Next Steps:\n \u2022 Continue to MNIST MLP after Module 08 (Training)\n \u2022 YOUR XOR solution scales to real vision problems!\n \u2022 Hidden layers principle powers all modern deep learning!\n",
|
||||
"loss": 0.2497,
|
||||
"accuracy": 54.5
|
||||
},
|
||||
"MNIST": {
|
||||
"success": true,
|
||||
"time": 2.02604603767395,
|
||||
"time": 1.9613378047943115,
|
||||
"output_preview": " a scalar is deprecated, and will error in future. Ensure you extract a single element from your array before performing this operation. (Deprecated NumPy 1.25.)\n one_hot[i, int(labels_np[i])] = 1.0\n",
|
||||
"loss": 0.0,
|
||||
"accuracy": 15.0
|
||||
"accuracy": 11.5
|
||||
},
|
||||
"CIFAR": {
|
||||
"success": false,
|
||||
@@ -27,8 +27,8 @@
|
||||
},
|
||||
"TinyGPT": {
|
||||
"success": true,
|
||||
"time": 1.851945161819458,
|
||||
"time": 1.9189341068267822,
|
||||
"output_preview": "ining\n \u2022 Complete transformer architecture from first principles\n\n\ud83c\udfed Production Note:\n Real PyTorch uses optimized CUDA kernels for attention,\n but you built and understand the core mathematics!\n",
|
||||
"loss": 0.3688
|
||||
"loss": 0.3419
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user