Spaces:
Runtime error
Runtime error
File size: 12,061 Bytes
1f39ae1 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 |
#!/usr/bin/env python3
"""
Performance comparison script for different training approaches
"""
import time
import torch
import os
import sys
from pathlib import Path
def benchmark_original_training():
"""Benchmark the original training approach"""
print("=== Benchmarking Original Training ===")
try:
# Import original training
from train_morphological import create_model, train_epoch, validate
from morphological_dataset import MorphologicalDataset, build_vocabulary, collate_fn
# Setup minimal data for benchmarking
os.makedirs("benchmark_data", exist_ok=True)
with open("benchmark_data/test.src", "w") as f:
f.write("hello world\n" * 100)
with open("benchmark_data/test.tgt", "w") as f:
f.write("hola mundo\n" * 100)
# Build vocabulary
src_vocab = build_vocabulary(["benchmark_data/test.src"])
tgt_vocab = build_vocabulary(["benchmark_data/test.tgt"])
# Create dataset
dataset = MorphologicalDataset("benchmark_data/test.src", "benchmark_data/test.tgt",
src_vocab, tgt_vocab, max_length=50)
# Create dataloader
from torch.utils.data import DataLoader
dataloader = DataLoader(
dataset,
batch_size=400,
shuffle=True,
collate_fn=lambda batch: collate_fn(batch, src_vocab, tgt_vocab, 50),
num_workers=4
)
# Create model
config = {
'embed_dim': 256,
'nb_heads': 4,
'src_hid_size': 1024,
'src_nb_layers': 4,
'trg_hid_size': 1024,
'trg_nb_layers': 4,
'dropout_p': 0.1,
'tie_trg_embed': True,
'label_smooth': 0.1,
}
model = create_model(config, src_vocab, tgt_vocab)
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = model.to(device)
# Create optimizer
optimizer = torch.optim.Adam(model.parameters(), lr=0.001)
# Benchmark training
start_time = time.time()
# Run a few epochs for benchmarking
for epoch in range(3):
train_loss, _ = train_epoch(
model, dataloader, optimizer, None, device, epoch, config
)
end_time = time.time()
total_time = end_time - start_time
print(f"Original training: {total_time:.2f}s for 3 epochs")
# Cleanup
import shutil
shutil.rmtree("benchmark_data")
return total_time
except Exception as e:
print(f"Original training benchmark failed: {e}")
return None
def benchmark_optimized_training():
"""Benchmark the optimized training approach"""
print("\n=== Benchmarking Optimized Training ===")
try:
# Import optimized training
from train_morphological_fast import create_optimized_model, train_epoch_ultra_fast, validate_fast
from morphological_dataset import MorphologicalDataset, build_vocabulary, collate_fn
# Setup minimal data for benchmarking
os.makedirs("benchmark_data", exist_ok=True)
with open("benchmark_data/test.src", "w") as f:
f.write("hello world\n" * 100)
with open("benchmark_data/test.tgt", "w") as f:
f.write("hola mundo\n" * 100)
# Build vocabulary
src_vocab = build_vocabulary(["benchmark_data/test.src"])
tgt_vocab = build_vocabulary(["benchmark_data/test.tgt"])
# Create dataset
dataset = MorphologicalDataset("benchmark_data/test.src", "benchmark_data/test.tgt",
src_vocab, tgt_vocab, max_length=50)
# Create dataloader
from torch.utils.data import DataLoader
dataloader = DataLoader(
dataset,
batch_size=800,
shuffle=True,
collate_fn=lambda batch: collate_fn(batch, src_vocab, tgt_vocab, 50),
num_workers=8,
pin_memory=True,
persistent_workers=True,
prefetch_factor=4,
drop_last=True
)
# Create model
config = {
'embed_dim': 256,
'nb_heads': 4,
'src_hid_size': 1024,
'src_nb_layers': 4,
'trg_hid_size': 1024,
'trg_nb_layers': 4,
'dropout_p': 0.1,
'tie_trg_embed': True,
'label_smooth': 0.1,
'use_amp': True,
'gradient_accumulation_steps': 1,
}
model = create_optimized_model(config, src_vocab, tgt_vocab)
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = model.to(device)
# Create optimizer
optimizer = torch.optim.AdamW(model.parameters(), lr=0.001, foreach=True)
# Create scaler
from torch.cuda.amp import GradScaler
scaler = GradScaler(enabled=True)
# Benchmark training
start_time = time.time()
# Run a few epochs for benchmarking
for epoch in range(3):
train_loss = train_epoch_ultra_fast(
model, dataloader, optimizer, device, epoch, config, scaler
)
end_time = time.time()
total_time = end_time - start_time
print(f"Optimized training: {total_time:.2f}s for 3 epochs")
# Cleanup
import shutil
shutil.rmtree("benchmark_data")
return total_time
except Exception as e:
print(f"Optimized training benchmark failed: {e}")
return None
def benchmark_cuda_training():
"""Benchmark the CUDA-optimized training approach"""
print("\n=== Benchmarking CUDA-Optimized Training ===")
try:
# Import CUDA-optimized training
from train_morphological_cuda import create_cuda_optimized_model, train_epoch_cuda, validate_cuda
from morphological_dataset import MorphologicalDataset, build_vocabulary, collate_fn
# Setup minimal data for benchmarking
os.makedirs("benchmark_data", exist_ok=True)
with open("benchmark_data/test.src", "w") as f:
f.write("hello world\n" * 100)
with open("benchmark_data/test.tgt", "w") as f:
f.write("hola mundo\n" * 100)
# Build vocabulary
src_vocab = build_vocabulary(["benchmark_data/test.src"])
tgt_vocab = build_vocabulary(["benchmark_data/test.tgt"])
# Create dataset
dataset = MorphologicalDataset("benchmark_data/test.src", "benchmark_data/test.tgt",
src_vocab, tgt_vocab, max_length=50)
# Create dataloader
from torch.utils.data import DataLoader
dataloader = DataLoader(
dataset,
batch_size=1024,
shuffle=True,
collate_fn=lambda batch: collate_fn(batch, src_vocab, tgt_vocab, 50),
num_workers=16,
pin_memory=True,
persistent_workers=True,
prefetch_factor=8,
drop_last=True,
multiprocessing_context='spawn'
)
# Create model
config = {
'embed_dim': 256,
'nb_heads': 4,
'src_hid_size': 1024,
'src_nb_layers': 4,
'trg_hid_size': 1024,
'trg_nb_layers': 4,
'dropout_p': 0.1,
'tie_trg_embed': True,
'label_smooth': 0.1,
'use_amp': True,
'gradient_accumulation_steps': 1,
}
model = create_cuda_optimized_model(config, src_vocab, tgt_vocab)
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = model.to(device, memory_format=torch.channels_last)
# Create optimizer
optimizer = torch.optim.AdamW(model.parameters(), lr=0.001, foreach=True, fused=True)
# Create scaler
from torch.cuda.amp import GradScaler
scaler = GradScaler(enabled=True)
# Benchmark training
start_time = time.time()
# Run a few epochs for benchmarking
for epoch in range(3):
train_loss = train_epoch_cuda(
model, dataloader, optimizer, device, epoch, config, scaler
)
end_time = time.time()
total_time = end_time - start_time
print(f"CUDA-optimized training: {total_time:.2f}s for 3 epochs")
# Cleanup
import shutil
shutil.rmtree("benchmark_data")
return total_time
except Exception as e:
print(f"CUDA training benchmark failed: {e}")
return None
def run_performance_comparison():
"""Run complete performance comparison"""
print("🚀 Performance Comparison: Training Approaches")
print("=" * 60)
# Check CUDA availability
if torch.cuda.is_available():
print(f"✓ CUDA available: {torch.cuda.get_device_name()}")
print(f"✓ CUDA Memory: {torch.cuda.get_device_properties(0).total_memory / 1024**3:.1f} GB")
else:
print("⚠ CUDA not available - some optimizations will be disabled")
print()
# Run benchmarks
results = {}
# Original training
original_time = benchmark_original_training()
if original_time:
results['Original'] = original_time
# Optimized training
optimized_time = benchmark_optimized_training()
if optimized_time:
results['Optimized'] = optimized_time
# CUDA-optimized training
cuda_time = benchmark_cuda_training()
if cuda_time:
results['CUDA-Optimized'] = cuda_time
# Print results
print("\n" + "=" * 60)
print("📊 PERFORMANCE COMPARISON RESULTS")
print("=" * 60)
if results:
# Sort by time (fastest first)
sorted_results = sorted(results.items(), key=lambda x: x[1])
fastest = sorted_results[0]
print(f"🏆 Fastest: {fastest[0]} ({fastest[1]:.2f}s)")
print("\nAll Results:")
for i, (name, time_taken) in enumerate(sorted_results):
if i == 0:
print(f"🥇 {name}: {time_taken:.2f}s (Baseline)")
elif i == 1:
print(f"🥈 {name}: {time_taken:.2f}s")
else:
print(f"🥉 {name}: {time_taken:.2f}s")
if i > 0:
speedup = fastest[1] / time_taken
print(f" Speedup: {speedup:.1f}x slower than {fastest[0]}")
# Calculate improvements
if len(results) >= 2:
print(f"\n🚀 Performance Improvements:")
baseline = results['Original'] if 'Original' in results else fastest[1]
for name, time_taken in results.items():
if name != 'Original':
improvement = baseline / time_taken
print(f" {name}: {improvement:.1f}x faster than baseline")
else:
print("❌ No benchmarks completed successfully")
print("\n" + "=" * 60)
print("💡 Recommendations:")
if 'CUDA-Optimized' in results:
print(" • Use CUDA-optimized training for maximum speed")
elif 'Optimized' in results:
print(" • Use optimized training for better performance")
else:
print(" • Consider upgrading PyTorch or checking dependencies")
print(" • Monitor GPU memory usage during training")
print(" • Adjust batch size based on your GPU memory")
return results
if __name__ == '__main__':
run_performance_comparison()
|