File size: 11,577 Bytes
afd2cc6 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 |
#!/usr/bin/env python
"""
Benchmark script to compare performance between standard Transformers
and CTranslate2 optimized translation models.
"""
import argparse
import json
import logging
import os
import sys
import time
from typing import Dict, List, Tuple
import numpy as np
import torch
import tqdm
# Add project root to path for imports
sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), '..')))
# Configure logging
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
)
logger = logging.getLogger(__name__)
# Import our models
try:
from app.models.translation_model import TranslationModel # Standard model
from app.models.translation_model_ct2 import TranslationModelCT2 # CTranslate2 model
except ImportError:
logger.error("Could not import translation models. Make sure you're running this script from the project root.")
sys.exit(1)
# Test sentences for various languages
TEST_SENTENCES = {
"en-es": [
"Hello, how are you today?",
"I would like to book a flight to Madrid for next week.",
"The quick brown fox jumps over the lazy dog.",
"Artificial intelligence is transforming the way we live and work.",
"Please contact our customer service if you have any questions."
],
"en-fr": [
"Hello, how are you today?",
"I would like to book a flight to Paris for next week.",
"The quick brown fox jumps over the lazy dog.",
"Artificial intelligence is transforming the way we live and work.",
"Please contact our customer service if you have any questions."
],
"en-de": [
"Hello, how are you today?",
"I would like to book a flight to Berlin for next week.",
"The quick brown fox jumps over the lazy dog.",
"Artificial intelligence is transforming the way we live and work.",
"Please contact our customer service if you have any questions."
],
"en-dra": [
"Hello, how are you today?",
"I would like to book a flight to Chennai for next week.",
"The quick brown fox jumps over the lazy dog.",
"Artificial intelligence is transforming the way we live and work.",
"Please contact our customer service if you have any questions."
]
}
def benchmark_standard_model(
src_lang: str,
tgt_lang: str,
sentences: List[str],
num_runs: int = 5,
warm_up: int = 2
) -> Dict:
"""Benchmark the standard Transformers model."""
logger.info(f"Benchmarking standard Transformers model for {src_lang}-{tgt_lang}")
# Initialize model
model = TranslationModel()
# Warm-up runs
logger.info(f"Performing {warm_up} warm-up runs...")
for _ in range(warm_up):
for sentence in sentences[:2]: # Use only first 2 sentences for warm-up
model.translate(sentence, src_lang, tgt_lang)
# Actual benchmark
logger.info(f"Performing {num_runs} benchmark runs...")
times = []
translations = []
for run in range(num_runs):
run_times = []
run_translations = []
for sentence in tqdm.tqdm(sentences, desc=f"Run {run+1}/{num_runs}"):
start_time = time.time()
translation = model.translate(sentence, src_lang, tgt_lang)
elapsed_time = time.time() - start_time
run_times.append(elapsed_time)
run_translations.append(translation)
times.append(run_times)
# Only keep translations from the first run
if run == 0:
translations = run_translations
# Calculate statistics
all_times = np.array(times).flatten()
stats = {
"mean_time": float(np.mean(all_times)),
"median_time": float(np.median(all_times)),
"std_dev": float(np.std(all_times)),
"min_time": float(np.min(all_times)),
"max_time": float(np.max(all_times)),
"total_time": float(np.sum(all_times)),
"num_sentences": len(sentences) * num_runs,
"translations": translations
}
return stats
def benchmark_ct2_model(
src_lang: str,
tgt_lang: str,
sentences: List[str],
num_runs: int = 5,
warm_up: int = 2
) -> Dict:
"""Benchmark the CTranslate2 optimized model."""
logger.info(f"Benchmarking CTranslate2 model for {src_lang}-{tgt_lang}")
# Initialize model
model = TranslationModelCT2()
# Warm-up runs
logger.info(f"Performing {warm_up} warm-up runs...")
for _ in range(warm_up):
for sentence in sentences[:2]: # Use only first 2 sentences for warm-up
model.translate(sentence, src_lang, tgt_lang)
# Actual benchmark
logger.info(f"Performing {num_runs} benchmark runs...")
times = []
translations = []
for run in range(num_runs):
run_times = []
run_translations = []
for sentence in tqdm.tqdm(sentences, desc=f"Run {run+1}/{num_runs}"):
start_time = time.time()
translation = model.translate(sentence, src_lang, tgt_lang)
elapsed_time = time.time() - start_time
run_times.append(elapsed_time)
run_translations.append(translation)
times.append(run_times)
# Only keep translations from the first run
if run == 0:
translations = run_translations
# Calculate statistics
all_times = np.array(times).flatten()
stats = {
"mean_time": float(np.mean(all_times)),
"median_time": float(np.median(all_times)),
"std_dev": float(np.std(all_times)),
"min_time": float(np.min(all_times)),
"max_time": float(np.max(all_times)),
"total_time": float(np.sum(all_times)),
"num_sentences": len(sentences) * num_runs,
"translations": translations
}
return stats
def benchmark_batch(
src_lang: str,
tgt_lang: str,
sentences: List[str],
num_runs: int = 5,
warm_up: int = 2
) -> Dict:
"""Benchmark batch translation with CTranslate2."""
logger.info(f"Benchmarking CTranslate2 batch translation for {src_lang}-{tgt_lang}")
# Initialize model
model = TranslationModelCT2()
# Warm-up runs
logger.info(f"Performing {warm_up} warm-up runs...")
for _ in range(warm_up):
model.translate_batch(sentences[:2], src_lang, tgt_lang)
# Actual benchmark
logger.info(f"Performing {num_runs} benchmark runs...")
times = []
translations = []
for run in range(num_runs):
start_time = time.time()
batch_translations = model.translate_batch(sentences, src_lang, tgt_lang)
elapsed_time = time.time() - start_time
times.append(elapsed_time)
# Only keep translations from the first run
if run == 0:
translations = batch_translations
# Calculate statistics
stats = {
"mean_time": float(np.mean(times)),
"median_time": float(np.median(times)),
"std_dev": float(np.std(times)),
"min_time": float(np.min(times)),
"max_time": float(np.max(times)),
"total_time": float(np.sum(times)),
"num_sentences": len(sentences),
"num_batches": num_runs,
"translations": translations
}
return stats
def run_benchmarks(
lang_pairs: List[Tuple[str, str]],
num_runs: int = 5,
warm_up: int = 2,
output_file: str = "benchmark_results.json"
) -> Dict:
"""Run benchmarks for specified language pairs."""
device = "cuda" if torch.cuda.is_available() else "cpu"
logger.info(f"Running benchmarks on {device}")
results = {
"device": device,
"cuda_available": torch.cuda.is_available(),
"cuda_version": torch.version.cuda if torch.cuda.is_available() else None,
"num_runs": num_runs,
"warm_up_runs": warm_up,
"language_pairs": {}
}
for src_lang, tgt_lang in lang_pairs:
model_key = f"{src_lang}-{tgt_lang}"
if model_key not in TEST_SENTENCES:
logger.warning(f"No test sentences available for {model_key}, skipping...")
continue
logger.info(f"Benchmarking {model_key}...")
sentences = TEST_SENTENCES[model_key]
# Run standard model benchmark
standard_stats = benchmark_standard_model(
src_lang, tgt_lang, sentences, num_runs, warm_up
)
# Run CTranslate2 model benchmark
ct2_stats = benchmark_ct2_model(
src_lang, tgt_lang, sentences, num_runs, warm_up
)
# Run batch translation benchmark
batch_stats = benchmark_batch(
src_lang, tgt_lang, sentences, num_runs, warm_up
)
# Calculate speedup
speedup = standard_stats["mean_time"] / ct2_stats["mean_time"]
batch_speedup = standard_stats["mean_time"] * len(sentences) / batch_stats["mean_time"]
results["language_pairs"][model_key] = {
"standard_model": standard_stats,
"ct2_model": ct2_stats,
"batch_translation": batch_stats,
"speedup": float(speedup),
"batch_speedup": float(batch_speedup)
}
# Print summary
logger.info(f"\nResults for {model_key}:")
logger.info(f" Standard model average time: {standard_stats['mean_time']:.4f}s")
logger.info(f" CTranslate2 model average time: {ct2_stats['mean_time']:.4f}s")
logger.info(f" Batch translation average time: {batch_stats['mean_time']:.4f}s (for {len(sentences)} sentences)")
logger.info(f" Speedup: {speedup:.2f}x")
logger.info(f" Batch speedup: {batch_speedup:.2f}x")
# Save results to file
with open(output_file, "w") as f:
json.dump(results, f, indent=2)
logger.info(f"Benchmark results saved to {output_file}")
return results
def main():
"""Main entry point for the benchmark script."""
parser = argparse.ArgumentParser(
description="Benchmark translation models performance"
)
parser.add_argument(
"--lang-pairs",
type=str,
nargs="+",
default=["en-es", "en-fr", "en-de", "en-dra"],
help="Language pairs to benchmark (e.g., 'en-es en-fr')"
)
parser.add_argument(
"--runs",
type=int,
default=5,
help="Number of benchmark runs"
)
parser.add_argument(
"--warm-up",
type=int,
default=2,
help="Number of warm-up runs"
)
parser.add_argument(
"--output",
type=str,
default="benchmark_results.json",
help="Output file for benchmark results"
)
args = parser.parse_args()
# Parse language pairs
lang_pairs = []
for pair in args.lang_pairs:
if "-" in pair:
src, tgt = pair.split("-")
lang_pairs.append((src, tgt))
else:
logger.warning(f"Invalid language pair format: {pair}, skipping...")
if not lang_pairs:
logger.error("No valid language pairs specified")
return 1
# Run benchmarks
run_benchmarks(
lang_pairs=lang_pairs,
num_runs=args.runs,
warm_up=args.warm_up,
output_file=args.output
)
return 0
if __name__ == "__main__":
sys.exit(main())
|