File size: 51,389 Bytes
ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 1bc7e54 ea0b2a0 1bc7e54 ea0b2a0 454e47c ea0b2a0 454e47c 1bc7e54 ea0b2a0 454e47c 1bc7e54 454e47c ea0b2a0 1bc7e54 ea0b2a0 454e47c ea0b2a0 1bc7e54 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 1bc7e54 454e47c ea0b2a0 454e47c 1bc7e54 454e47c 1bc7e54 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 1bc7e54 ea0b2a0 1bc7e54 ea0b2a0 1bc7e54 ea0b2a0 1bc7e54 454e47c ea0b2a0 1bc7e54 ea0b2a0 454e47c ea0b2a0 1bc7e54 ea0b2a0 1bc7e54 ea0b2a0 1bc7e54 ea0b2a0 1bc7e54 ea0b2a0 1bc7e54 ea0b2a0 1bc7e54 ea0b2a0 1bc7e54 ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 1bc7e54 ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c 1bc7e54 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 1bc7e54 ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c 1bc7e54 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 1bc7e54 454e47c 1bc7e54 454e47c 1bc7e54 ea0b2a0 1bc7e54 454e47c ea0b2a0 1bc7e54 454e47c 1bc7e54 454e47c 1bc7e54 454e47c 1bc7e54 ea0b2a0 454e47c 1bc7e54 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 1bc7e54 454e47c ea0b2a0 1bc7e54 454e47c 1bc7e54 ea0b2a0 454e47c 1bc7e54 ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c 1bc7e54 454e47c ea0b2a0 454e47c 1bc7e54 ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c 1bc7e54 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c ea0b2a0 454e47c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 |
"""
Comprehensive Model Evaluation Script for Code-Specialized Embedding Models.
This script evaluates embedding models on both task performance and operational metrics:
Task Performance:
- CodeSearchNet evaluation (NDCG, MRR, Recall metrics)
- Code search accuracy across programming languages
Operational Performance:
- Inference speed (latency and throughput)
- Memory efficiency (RAM and GPU usage)
- Model size and storage requirements
- CPU vs GPU performance scaling
Usage:
distiller evaluate [--use-beam] [--skip-benchmark] # Run evaluation locally or on Beam
"""
# Try to import flash_attn to check if it's available
import contextlib
import importlib.util
import json
import logging
import os
import time
import traceback
from pathlib import Path
from typing import Any
import numpy as np
import pandas as pd
import psutil
import torch
import typer
from beam import function
from datasets import Dataset, load_dataset
from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity
from tqdm import tqdm
from .beam_utils import download_specific_evaluation_file
from .config import (
DEFAULT_EVALUATION_MODELS,
codesearchnet_config,
directories,
get_evaluation_function_kwargs,
get_safe_model_name,
get_volume_config,
languages_config,
)
# Check if flash_attn is available and compatible
FLASH_ATTN_AVAILABLE = importlib.util.find_spec("flash_attn") is not None
logger = logging.getLogger(__name__)
# =============================================================================
# EVALUATION CONFIGURATION
# =============================================================================
BATCH_SIZE = 10
LOCAL_EVALUATION_DIR = directories.evaluation_results
LOCAL_BENCHMARK_DIR = directories.benchmark_results
LOCAL_MODELS_DIR = directories.final
VOLUME_CONFIG = get_volume_config()
# =============================================================================
# CORE EVALUATION CLASSES
# =============================================================================
# Sample texts for benchmarking (various lengths)
BENCHMARK_TEXTS = {
"short": [
"def add(a, b): return a + b",
"function multiply(x, y) { return x * y; }",
"class Calculator { public int subtract(int a, int b) { return a - b; } }",
]
* 100, # 300 short texts
"medium": [
"def fibonacci(n):\n if n <= 1:\n return n\n return fibonacci(n-1) + fibonacci(n-2)",
"function quickSort(arr) {\n if (arr.length <= 1) return arr;\n const pivot = arr[arr.length - 1];\n const left = [], right = [];\n for (let i = 0; i < arr.length - 1; i++) {\n if (arr[i] < pivot) left.push(arr[i]);\n else right.push(arr[i]);\n }\n return [...quickSort(left), pivot, ...quickSort(right)];\n}",
]
* 50, # 100 medium texts
"long": [
"""
def complex_algorithm(data, config):
'''
Complex data processing algorithm with multiple steps.
'''
results = []
# Data validation and processing steps...
return results
""".strip(),
]
* 20, # 20 long texts
}
def reset_cuda_state() -> None:
"""Aggressively reset CUDA state after memory allocation errors."""
if not torch.cuda.is_available():
return
try:
# Clear all CUDA caches
torch.cuda.empty_cache()
torch.cuda.ipc_collect()
torch.cuda.reset_peak_memory_stats()
# Try to force garbage collection
import gc
gc.collect()
logger.info("π§Ή CUDA state reset completed")
except Exception as e:
logger.warning(f"β οΈ Could not fully reset CUDA state: {e}")
def configure_flash_attention() -> dict[str, Any]:
"""Configure flash attention settings and return model kwargs."""
model_kwargs: dict[str, Any] = {}
if not FLASH_ATTN_AVAILABLE:
logger.info("β οΈ Flash attention not available - using standard attention")
return model_kwargs
# Set environment variables for flash attention and CUDA memory management
os.environ["TOKENIZERS_PARALLELISM"] = "false"
os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "expandable_segments:True"
# Check if we're on a compatible GPU
try:
if torch.cuda.is_available():
device_capability = torch.cuda.get_device_capability()
# Flash attention requires compute capability >= 7.5 (Turing, Ampere, Ada, Hopper)
if device_capability[0] >= 7 and (device_capability[0] > 7 or device_capability[1] >= 5):
logger.info("β
Flash attention available - compatible GPU detected")
# For SentenceTransformer, we'll use environment variables to enable flash attention
os.environ["TRANSFORMERS_FLASH_ATTENTION"] = "1"
else:
logger.info(f"β οΈ GPU compute capability {device_capability} < 7.5 - flash attention disabled")
else:
logger.info("β οΈ No CUDA available - flash attention disabled")
except Exception as e:
logger.warning(f"β οΈ Failed to check GPU compatibility: {e} - flash attention disabled")
return model_kwargs
def load_model_with_flash_attention(model_path: str, device: str = "auto") -> SentenceTransformer:
"""Load a SentenceTransformer model with flash attention if available."""
# Convert "auto" device to actual device
target_device = "cuda" if device == "auto" and torch.cuda.is_available() else device
if device == "auto" and not torch.cuda.is_available():
target_device = "cpu"
# Configure flash attention via environment variables
configure_flash_attention()
# Load model with standard SentenceTransformer initialization
logger.info(f"π Loading model: {Path(model_path).name}")
try:
# Try loading directly to target device first
model = SentenceTransformer(model_path, device=target_device, trust_remote_code=True)
logger.info(f"β
Model loaded successfully on {target_device}")
return model
except (torch.OutOfMemoryError, RuntimeError) as oom_error:
# Handle both torch.OutOfMemoryError and RuntimeError (CUDA driver errors)
is_oom = isinstance(oom_error, torch.OutOfMemoryError) or "out of memory" in str(oom_error).lower()
if is_oom and target_device != "cpu":
logger.warning(f"β οΈ OOM loading directly to {target_device}, trying CPU first: {oom_error}")
try:
# Clear CUDA cache more aggressively after OOM
reset_cuda_state()
logger.info("π Loading model on CPU first, then trying to move to GPU...")
model = SentenceTransformer(model_path, device="cpu", trust_remote_code=True)
logger.info("π¦ Model loaded on CPU, attempting GPU transfer...")
# Try moving to GPU with additional error handling
try:
model = model.to(target_device)
logger.info(f"β
Model successfully moved to {target_device}")
return model
except (RuntimeError, AssertionError) as gpu_move_error:
# Handle PyTorch internal errors and CUDA allocator issues
logger.warning(f"β οΈ GPU transfer failed: {gpu_move_error}")
if "INTERNAL ASSERT FAILED" in str(gpu_move_error) or "handles_" in str(gpu_move_error):
logger.warning("π§ Detected CUDA allocator corruption, resetting and staying on CPU")
# Try to reset CUDA context
reset_cuda_state()
else:
# Re-raise unexpected GPU transfer errors
raise
logger.info("β
Model will remain on CPU due to GPU memory issues")
return model
except Exception as fallback_error:
logger.warning(f"β οΈ CPU fallback failed: {fallback_error}, loading fresh on CPU")
# Clear any remaining CUDA state
reset_cuda_state()
model = SentenceTransformer(model_path, device="cpu", trust_remote_code=True)
logger.info("β
Model loaded on CPU (all GPU attempts failed)")
return model
else:
# Re-raise if not OOM or already on CPU
raise
except ValueError as e:
if "'MaxSim' is not a valid SimilarityFunction" in str(e):
logger.warning(f"β οΈ Model {Path(model_path).name} uses unsupported MaxSim similarity function")
logger.info("π§ Attempting workaround by loading with custom config...")
# Try loading with similarity function override
try:
# Load model components manually and override similarity function
import json
import tempfile
from pathlib import Path as PathLib
# Create temporary directory for modified config
with tempfile.TemporaryDirectory() as temp_dir:
temp_path = PathLib(temp_dir) / "temp_model"
# Download/copy model files
if model_path.startswith("http") or ("/" in model_path and not PathLib(model_path).exists()):
# It's a HuggingFace model ID
from huggingface_hub import snapshot_download
snapshot_download(model_path, local_dir=temp_path, ignore_patterns=["*.bin"])
else:
# It's a local path
import shutil
shutil.copytree(model_path, temp_path)
# Modify config to use supported similarity function
config_path = temp_path / "config_sentence_transformers.json"
if config_path.exists():
with config_path.open() as f:
config = json.load(f)
# Override similarity function to 'cosine' (supported)
if "similarity_fn_name" in config:
logger.info(
f"π§ Changing similarity function from '{config['similarity_fn_name']}' to 'cosine'"
)
config["similarity_fn_name"] = "cosine"
with config_path.open("w") as f:
json.dump(config, f, indent=2)
# Load model with modified config
model = SentenceTransformer(str(temp_path), device=device, trust_remote_code=True)
logger.info("β
Model loaded successfully with similarity function workaround")
return model
except Exception as workaround_error:
logger.warning(f"β οΈ Similarity function workaround failed: {workaround_error}")
logger.info("π§ Attempting direct model component loading...")
# Last resort: try loading model components directly
try:
from sentence_transformers.models import Pooling, Transformer
# Load model components manually
logger.info("π Loading model components directly...")
# Create SentenceTransformer components using model path
transformer = Transformer(model_path)
pooling = Pooling(transformer.get_word_embedding_dimension())
# Create SentenceTransformer with manual components
model = SentenceTransformer(modules=[transformer, pooling], device=device)
logger.info("β
Model loaded successfully with direct component loading")
return model
except Exception as direct_error:
logger.warning(f"β οΈ Direct component loading failed: {direct_error}")
logger.exception(f"β All loading methods failed for {Path(model_path).name}")
raise e from direct_error
else:
raise
except Exception:
logger.exception(f"β Failed to load model {Path(model_path).name}")
raise
class PerformanceBenchmark:
"""Comprehensive performance benchmarking for embedding models."""
def __init__(self, model_path: str, model_name: str | None = None) -> None:
"""Initialize benchmarker with model."""
self.model_path = model_path
self.model_name = model_name or Path(model_path).name
self.model: SentenceTransformer | None = None
self.device = "cuda" if torch.cuda.is_available() else "cpu"
self.results: dict[str, Any] = {}
def load_model(self) -> None:
"""Load the embedding model."""
logger.info(f"Loading model from {self.model_path}")
start_time = time.time()
try:
self.model = load_model_with_flash_attention(self.model_path, self.device)
load_time = time.time() - start_time
logger.info(f"β
Model loaded in {load_time:.2f}s on {self.device}")
self.results["model_load_time"] = load_time
except Exception:
logger.exception("β Failed to load model")
self.results["error"] = traceback.format_exc()
raise
def measure_model_size(self) -> dict[str, float]:
"""Measure model size metrics."""
logger.info("π Measuring model size...")
size_metrics: dict[str, Any] = {}
# Disk size
try:
if Path(self.model_path).is_dir():
# Local directory - calculate size of model files only
model_extensions = {".safetensors", ".bin", ".json", ".txt", ".tokenizer"}
total_size = 0
model_dir = Path(self.model_path)
for file_path in model_dir.rglob("*"):
if file_path.is_file() and (
file_path.suffix.lower() in model_extensions or "tokenizer" in file_path.name.lower()
):
total_size += file_path.stat().st_size
size_metrics["disk_size_mb"] = total_size / (1024 * 1024)
# HuggingFace model - estimate based on model parameters
elif self.model is not None:
param_count = sum(p.numel() for p in self.model.parameters())
# Rough estimate: 4 bytes per parameter (float32)
estimated_size = param_count * 4
size_metrics["disk_size_mb"] = estimated_size / (1024 * 1024)
else:
size_metrics["disk_size_mb"] = 0.0
except Exception as e:
logger.warning(f"β οΈ Could not calculate disk size: {e}")
size_metrics["disk_size_mb"] = 0.0
# Memory size (if model is loaded)
if self.model is not None:
try:
# Parameter count
param_count = sum(p.numel() for p in self.model.parameters())
size_metrics["parameter_count"] = param_count
size_metrics["parameters_millions"] = param_count / 1e6
# Memory usage estimate
param_size = sum(p.numel() * p.element_size() for p in self.model.parameters())
buffer_size = sum(b.numel() * b.element_size() for b in self.model.buffers())
size_metrics["memory_size_mb"] = (param_size + buffer_size) / (1024 * 1024)
size_metrics["ram_usage_mb"] = size_metrics["memory_size_mb"]
# GPU memory if using CUDA
if self.device == "cuda" and torch.cuda.is_available():
size_metrics["gpu_memory_mb"] = torch.cuda.memory_allocated() / (1024 * 1024)
size_metrics["gpu_name"] = torch.cuda.get_device_name(0)
# Embedding dimension if available
if hasattr(self.model, "get_sentence_embedding_dimension"):
size_metrics["embedding_dim"] = self.model.get_sentence_embedding_dimension()
except Exception as e:
logger.warning(f"β οΈ Could not calculate memory size: {e}")
# Update results
self.results["size_metrics"] = size_metrics
return size_metrics
def benchmark_inference_speed(self, batch_sizes: list[int] | None = None) -> dict[str, Any]:
"""Benchmark inference speed with different batch sizes."""
if batch_sizes is None:
batch_sizes = [1, 8, 16, 32]
logger.info(f"β‘ Benchmarking inference speed with batch sizes: {batch_sizes}")
if self.model is None:
self.load_model()
speed_results: dict[str, Any] = {"medium": {}}
# Use medium-length texts for speed testing
test_texts = BENCHMARK_TEXTS["medium"]
for batch_size in batch_sizes:
logger.info(f" π Testing batch size: {batch_size}")
# Prepare batch
batch = (
test_texts[:batch_size]
if batch_size <= len(test_texts)
else test_texts * ((batch_size // len(test_texts)) + 1)
)
batch = batch[:batch_size]
# Warmup
if self.model is not None:
_ = self.model.encode(batch[: min(4, len(batch))], convert_to_tensor=False)
# Benchmark multiple runs
latencies = []
num_runs = max(3, 20 // batch_size) # More runs for smaller batches
for _ in range(num_runs):
start_time = time.time()
if self.model is not None:
_ = self.model.encode(batch, convert_to_tensor=False, normalize_embeddings=True)
end_time = time.time()
latencies.append(end_time - start_time)
# Calculate metrics
avg_latency = sum(latencies) / len(latencies)
throughput = batch_size / avg_latency
time_per_text_ms = (avg_latency / batch_size) * 1000
batch_key = f"batch_{batch_size}"
speed_results["medium"][batch_key] = {
"time_per_text_ms": time_per_text_ms,
"texts_per_second": throughput,
"tokens_per_second": throughput * 50, # Estimate 50 tokens per text
}
logger.info(f" β‘ Latency: {avg_latency:.3f}s, Throughput: {throughput:.1f} texts/sec")
# Update results
self.results["speed_benchmarks"] = speed_results
return speed_results
def benchmark_memory_scaling(self, batch_sizes: list[int] | None = None) -> dict[str, Any]:
"""Benchmark memory usage scaling with batch size."""
if batch_sizes is None:
batch_sizes = [1, 8, 16, 32]
logger.info(f"π§ Benchmarking memory scaling with batch sizes: {batch_sizes}")
if self.model is None:
self.load_model()
memory_results: dict[str, Any] = {}
test_texts = BENCHMARK_TEXTS["medium"]
for batch_size in batch_sizes:
logger.info(f" π Testing memory with batch size: {batch_size}")
# Prepare batch
batch = (
test_texts[:batch_size]
if batch_size <= len(test_texts)
else test_texts * ((batch_size // len(test_texts)) + 1)
)
batch = batch[:batch_size]
# Clear GPU cache if using CUDA
if torch.cuda.is_available():
torch.cuda.empty_cache()
torch.cuda.reset_peak_memory_stats()
try:
# Run inference
if self.model is not None:
_ = self.model.encode(batch, convert_to_tensor=False)
# Measure peak memory
if torch.cuda.is_available():
peak_memory = torch.cuda.max_memory_allocated() / (1024 * 1024)
memory_per_text = peak_memory / batch_size
else:
# Use psutil for CPU memory (less accurate)
peak_memory = psutil.virtual_memory().used / (1024 * 1024)
memory_per_text = 0 # Can't accurately measure per-text on CPU
batch_key = f"batch_{batch_size}"
memory_results[batch_key] = {
"memory_used_mb": peak_memory,
"memory_per_text_mb": memory_per_text,
"oom": False,
}
logger.info(f" π§ Peak memory: {peak_memory:.1f}MB, Per text: {memory_per_text:.2f}MB")
except Exception as e:
logger.warning(f"β οΈ Memory benchmark failed for batch {batch_size}: {e}")
batch_key = f"batch_{batch_size}"
memory_results[batch_key] = {
"oom": True,
"error": str(e),
}
self.results["memory_benchmarks"] = memory_results
return memory_results
def benchmark_cpu_vs_gpu(self) -> dict[str, Any]:
"""Compare CPU vs GPU performance."""
logger.info("βοΈ Benchmarking CPU vs GPU performance")
if not torch.cuda.is_available():
logger.warning("β οΈ CUDA not available - skipping GPU benchmark")
return {}
comparison_results: dict[str, Any] = {}
test_texts = BENCHMARK_TEXTS["medium"][:16] # Use 16 texts for comparison
for device in ["cpu", "cuda"]:
logger.info(f" π Testing on {device.upper()}")
try:
model = load_model_with_flash_attention(self.model_path, device)
# Warmup
_ = model.encode(test_texts[:4], convert_to_tensor=False)
# Benchmark
start_time = time.time()
_ = model.encode(test_texts, convert_to_tensor=False, normalize_embeddings=True)
end_time = time.time()
latency = end_time - start_time
throughput = len(test_texts) / latency
comparison_results[device] = {
"texts_per_second": throughput,
}
logger.info(f" β‘ {device.upper()}: {latency:.3f}s, {throughput:.1f} texts/sec")
# Clean up
del model
if device == "cuda":
torch.cuda.empty_cache()
except Exception as e:
logger.warning(f"β οΈ Failed to benchmark {device}: {e}")
comparison_results[device] = {"error": str(e)}
# Calculate speedup
if "cpu" in comparison_results and "cuda" in comparison_results:
cpu_throughput = comparison_results["cpu"].get("texts_per_second", 0)
gpu_throughput = comparison_results["cuda"].get("texts_per_second", 0)
if cpu_throughput > 0:
speedup = gpu_throughput / cpu_throughput
comparison_results["gpu_speedup"] = speedup
logger.info(f" π GPU Speedup: {speedup:.1f}x")
self.results["cpu_vs_gpu"] = comparison_results
return comparison_results
def run_comprehensive_benchmark(self) -> dict[str, Any]:
"""Run all benchmarks and return comprehensive results."""
logger.info(f"π Starting comprehensive benchmark for {self.model_name}")
# Model information
self.results["model_name"] = self.model_name
self.results["model_path"] = self.model_path
self.results["timestamp"] = time.strftime("%Y-%m-%d %H:%M:%S")
# Run all benchmarks
try:
self.load_model()
self.measure_model_size()
self.benchmark_inference_speed([1, 8, 16, 32])
self.benchmark_memory_scaling([1, 8, 16, 32])
self.benchmark_cpu_vs_gpu()
logger.info(f"β
Comprehensive benchmark completed for {self.model_name}")
except Exception:
logger.exception(f"β Benchmark failed for {self.model_name}")
self.results["error"] = traceback.format_exc()
return self.results
class CodeSearchNetEvaluator:
"""Evaluator for CodeSearchNet-style code search tasks."""
def __init__(self, model_path: str, model_name: str | None = None) -> None:
"""Initialize the evaluator with a model."""
self.model_path = model_path
self.model_name = model_name or Path(model_path).name
self.model: SentenceTransformer | None = None
self._load_model()
def _load_model(self) -> None:
"""Load the embedding model."""
logger.info(f"Loading model from {self.model_path}")
try:
self.model = load_model_with_flash_attention(self.model_path)
logger.info(f"Successfully loaded model: {self.model_name}")
except Exception:
logger.exception(f"Failed to load model from {self.model_path}")
raise
def encode_texts(self, texts: list[str], desc: str = "Encoding") -> np.ndarray:
"""Encode texts into embeddings with batching and memory management."""
if self.model is None:
msg = "Model not loaded"
raise RuntimeError(msg)
embeddings = []
# Use smaller batch size to avoid OOM
effective_batch_size = min(BATCH_SIZE, 5) # Limit to 5 for large models
for i in tqdm(range(0, len(texts), effective_batch_size), desc=desc):
batch = texts[i : i + effective_batch_size]
try:
batch_embeddings = self.model.encode(batch, convert_to_tensor=False, normalize_embeddings=True)
embeddings.append(batch_embeddings)
# Clear CUDA cache periodically to prevent memory buildup
if torch.cuda.is_available() and i > 0 and i % (effective_batch_size * 4) == 0:
torch.cuda.empty_cache()
except (torch.OutOfMemoryError, RuntimeError) as e:
# Handle both torch.OutOfMemoryError and RuntimeError (CUDA driver errors)
is_oom = isinstance(e, torch.OutOfMemoryError) or "out of memory" in str(e).lower()
if is_oom:
logger.warning(
f"β οΈ OOM during encoding batch {i // effective_batch_size + 1}, trying smaller batch..."
)
# Try encoding one at a time
for single_text in batch:
try:
single_embedding = self.model.encode(
[single_text], convert_to_tensor=False, normalize_embeddings=True
)
embeddings.append(single_embedding)
if torch.cuda.is_available():
torch.cuda.empty_cache()
except (torch.OutOfMemoryError, RuntimeError) as single_e:
if isinstance(single_e, torch.OutOfMemoryError) or "out of memory" in str(single_e).lower():
logger.exception("β Cannot encode even single text, model too large for GPU")
raise
raise
else:
raise
return np.vstack(embeddings)
def evaluate_language(self, language: str, max_queries: int = 100) -> dict[str, Any]:
"""Evaluate on a specific programming language."""
logger.info(f"Evaluating on {language} language (max {max_queries} queries)")
try:
# Load ONLY test split for the language with streaming to avoid loading full dataset
logger.info(f"π₯ Loading test split for {language}...")
dataset = load_dataset(
codesearchnet_config.dataset_name,
language,
split=f"test[:{max_queries * 10}]", # Load 10x more than needed to ensure we get enough valid pairs
trust_remote_code=True,
)
if not isinstance(dataset, Dataset):
logger.error(f"Unexpected dataset type for {language}: {type(dataset)}")
return {}
logger.info(f"π Loaded {len(dataset)} examples from {language} test split")
queries: list[str] = []
codes: list[str] = []
query_ids: list[str] = []
# Process examples and stop once we have enough valid pairs
for i, example in enumerate(dataset):
if len(queries) >= max_queries: # Stop once we have enough
break
doc_string = example.get("func_documentation_string", "").strip()
code_string = example.get("func_code_string", "").strip()
if doc_string and code_string and len(doc_string.split()) >= 3:
queries.append(doc_string)
codes.append(code_string)
query_ids.append(f"{language}_{i}")
if len(queries) == 0:
logger.warning(f"No valid query-code pairs found for {language}")
return {}
# Truncate to exactly max_queries if we have more
if len(queries) > max_queries:
queries = queries[:max_queries]
codes = codes[:max_queries]
query_ids = query_ids[:max_queries]
logger.info(f"Found {len(queries)} valid query-code pairs for {language}")
# Check available memory before encoding
if torch.cuda.is_available():
free_memory = torch.cuda.get_device_properties(0).total_memory - torch.cuda.memory_allocated()
free_gb = free_memory / (1024**3)
logger.info(f"πΎ Available GPU memory before encoding: {free_gb:.1f} GB")
if free_gb < 2.0: # Less than 2GB free
logger.warning(f"β οΈ Low GPU memory ({free_gb:.1f} GB), using conservative encoding")
torch.cuda.empty_cache()
# Encode queries and codes
start_time = time.time()
query_embeddings = self.encode_texts(queries, f"Encoding {language} queries")
code_embeddings = self.encode_texts(codes, f"Encoding {language} code")
encoding_time = time.time() - start_time
# Compute similarities and metrics
similarities = cosine_similarity(query_embeddings, code_embeddings)
metrics = self._compute_retrieval_metrics(similarities)
# Prepare results
results = {
"language": language,
"model_name": self.model_name,
"num_queries": len(queries),
"encoding_time_seconds": encoding_time,
"metrics": metrics,
}
logger.info(f"β
{language} evaluation completed in {encoding_time:.2f}s")
return results
except Exception:
logger.exception(f"β Failed to evaluate {language}")
return {}
def _compute_retrieval_metrics(self, similarities: np.ndarray) -> dict[str, float]:
"""Compute retrieval metrics from similarity matrix."""
n_queries = similarities.shape[0]
# For each query, the correct code is at the same index
correct_indices = np.arange(n_queries)
# Rank all codes for each query
ranked_indices = np.argsort(similarities, axis=1)[:, ::-1]
metrics = {}
# Compute metrics for different k values
for k in [1, 5, 10]:
if k <= similarities.shape[1]:
# Recall@k
recall_k = np.mean([correct_indices[i] in ranked_indices[i, :k] for i in range(n_queries)])
metrics[f"recall@{k}"] = recall_k
# NDCG@k
ndcg_k = np.mean(
[self._compute_ndcg(ranked_indices[i], correct_indices[i], k) for i in range(n_queries)]
)
metrics[f"ndcg@{k}"] = ndcg_k
# Mean Reciprocal Rank
reciprocal_ranks = []
for i in range(n_queries):
rank = np.where(ranked_indices[i] == correct_indices[i])[0]
if len(rank) > 0:
reciprocal_ranks.append(1.0 / (rank[0] + 1))
else:
reciprocal_ranks.append(0.0)
metrics["mrr"] = np.mean(reciprocal_ranks)
# Add mean rank and median rank
mean_ranks = []
for i in range(n_queries):
rank = np.where(ranked_indices[i] == correct_indices[i])[0]
if len(rank) > 0:
mean_ranks.append(rank[0] + 1) # 1-indexed
else:
mean_ranks.append(similarities.shape[1]) # Worst possible rank
metrics["mean_rank"] = np.mean(mean_ranks)
metrics["median_rank"] = np.median(mean_ranks)
# Ensure all values are float
return {k: float(v) for k, v in metrics.items()}
def _compute_ndcg(self, ranked_indices: np.ndarray, correct_idx: int, k: int) -> float:
"""Compute NDCG@k for a single query."""
if correct_idx in ranked_indices[:k]:
rank = np.where(ranked_indices[:k] == correct_idx)[0][0]
return 1.0 / np.log2(rank + 2)
return 0.0
def evaluate_all_languages(
self, max_queries_per_lang: int = 100, languages: list[str] | None = None
) -> dict[str, Any]:
"""Evaluate on all specified languages."""
eval_languages = languages or languages_config.all
logger.info(f"π Starting evaluation on {len(eval_languages)} languages")
logger.info(f"π Model: {self.model_name}")
logger.info(f"π’ Max queries per language: {max_queries_per_lang}")
start_time = time.time()
results = {
"model_name": self.model_name,
"model_path": self.model_path,
"languages": {},
"overall": {},
"evaluation_time_seconds": 0,
}
languages_dict: dict[str, Any] = {}
# Evaluate each language
for language in eval_languages:
logger.info(f"\n{'=' * 50}")
logger.info(f"π Evaluating {language}")
logger.info(f"{'=' * 50}")
lang_results = self.evaluate_language(language, max_queries_per_lang)
if lang_results:
languages_dict[language] = lang_results
results["languages"] = languages_dict
# Compute overall metrics
if languages_dict:
overall_metrics = {}
metric_names = list(next(iter(languages_dict.values()))["metrics"].keys())
for metric in metric_names:
values = [languages_dict[lang]["metrics"][metric] for lang in languages_dict]
overall_metrics[metric] = np.mean(values)
results["overall"] = overall_metrics
total_time = time.time() - start_time
results["evaluation_time_seconds"] = total_time
logger.info(f"Evaluation completed in {total_time:.2f} seconds")
return results
class ComprehensiveModelEvaluator:
"""Combined evaluator for both task performance and operational benchmarks."""
def __init__(self, model_path: str, model_name: str | None = None) -> None:
"""Initialize the comprehensive evaluator with a model."""
self.model_path = model_path
self.model_name = model_name or Path(model_path).name
# Initialize sub-evaluators
self.codesearch_evaluator = CodeSearchNetEvaluator(model_path, model_name)
self.performance_benchmarker = PerformanceBenchmark(model_path, model_name)
self.results: dict[str, Any] = {}
def run_comprehensive_evaluation(
self,
max_queries_per_lang: int = 100,
languages: list[str] | None = None,
skip_benchmark: bool = False,
) -> dict[str, Any]:
"""Run both CodeSearchNet evaluation and performance benchmarking."""
logger.info(f"π Starting comprehensive evaluation for {self.model_name}")
start_time = time.time()
# Initialize results structure
self.results = {
"model_name": self.model_name,
"model_path": self.model_path,
"timestamp": time.strftime("%Y-%m-%d %H:%M:%S"),
"evaluation_time_seconds": 0,
}
try:
# 1. Run CodeSearchNet evaluation
logger.info("π Running CodeSearchNet task evaluation...")
codesearch_results = self.codesearch_evaluator.evaluate_all_languages(max_queries_per_lang, languages)
# Extract CodeSearchNet metrics
self.results.update(
{
"codesearch_languages": codesearch_results.get("languages", {}),
"codesearch_overall": codesearch_results.get("overall", {}),
}
)
# 2. Run performance benchmarking (unless skipped)
if not skip_benchmark:
logger.info("β‘ Running operational performance benchmarking...")
benchmark_results = self.performance_benchmarker.run_comprehensive_benchmark()
# Extract benchmark metrics
self.results.update(
{
"size_metrics": benchmark_results.get("size_metrics", {}),
"speed_benchmarks": benchmark_results.get("speed_benchmarks", {}),
"memory_benchmarks": benchmark_results.get("memory_benchmarks", {}),
"cpu_vs_gpu": benchmark_results.get("cpu_vs_gpu", {}),
}
)
else:
logger.info("βοΈ Skipping performance benchmarking")
self.results["benchmark_skipped"] = True
except Exception as e:
logger.exception(f"β Comprehensive evaluation failed for {self.model_name}")
self.results["error"] = str(e)
# Calculate total time
total_time = time.time() - start_time
self.results["evaluation_time_seconds"] = total_time
logger.info(f"β
Comprehensive evaluation completed in {total_time:.2f} seconds")
return self.results
def print_summary(self) -> None:
"""Print a comprehensive summary of all results."""
logger.info(f"\n{'=' * 60}")
logger.info(f"π COMPREHENSIVE EVALUATION RESULTS: {self.model_name}")
logger.info(f"{'=' * 60}")
# CodeSearchNet results
overall = self.results.get("codesearch_overall", {})
if overall:
logger.info("π CodeSearchNet Performance:")
for metric, value in overall.items():
logger.info(f" π― {metric.upper()}: {value:.4f}")
# Benchmark results
if not self.results.get("benchmark_skipped", False):
size_metrics = self.results.get("size_metrics", {})
if size_metrics:
logger.info(f"\nπ Model Size: {size_metrics.get('disk_size_mb', 0):.1f}MB")
if "parameters_millions" in size_metrics:
logger.info(f"π’ Parameters: {size_metrics['parameters_millions']:.1f}M")
speed_benchmarks = self.results.get("speed_benchmarks", {})
if "medium" in speed_benchmarks and "batch_32" in speed_benchmarks["medium"]:
batch_32 = speed_benchmarks["medium"]["batch_32"]
logger.info(f"β‘ Throughput (batch 32): {batch_32.get('texts_per_second', 0):.1f} texts/sec")
cpu_vs_gpu = self.results.get("cpu_vs_gpu", {})
if "gpu_speedup" in cpu_vs_gpu:
speedup = cpu_vs_gpu["gpu_speedup"]
logger.info(f"π GPU speedup: {speedup:.1f}x")
# Language breakdown
languages = self.results.get("codesearch_languages", {})
if languages:
logger.info("\nπ Language Breakdown:")
for lang, lang_results in languages.items():
metrics = lang_results.get("metrics", {})
ndcg10 = metrics.get("ndcg@10", 0)
mrr = metrics.get("mrr", 0)
logger.info(f" {lang}: NDCG@10={ndcg10:.4f}, MRR={mrr:.4f}")
# =============================================================================
# UTILITY FUNCTIONS
# =============================================================================
def check_existing_results(model_name: str, local_dir: str = LOCAL_EVALUATION_DIR) -> dict[str, Any] | None:
"""Check if comprehensive evaluation results already exist for a model."""
local_path = Path(local_dir)
safe_model_name = get_safe_model_name(model_name)
# Check for new comprehensive format first
comprehensive_file = local_path / f"comprehensive_eval_{safe_model_name}.json"
if comprehensive_file.exists():
try:
with comprehensive_file.open("r") as f:
results = json.load(f)
logger.info(f"β
Found existing comprehensive results for {model_name}")
return results
except Exception as e:
logger.warning(f"β οΈ Could not load existing comprehensive results for {model_name}: {e}")
# Fallback to legacy codesearchnet format for backward compatibility
legacy_file = local_path / f"codesearchnet_eval_{safe_model_name}.json"
if legacy_file.exists():
try:
with legacy_file.open("r") as f:
results = json.load(f)
logger.info(f"β
Found existing legacy results for {model_name}")
return results
except Exception as e:
logger.warning(f"β οΈ Could not load existing legacy results for {model_name}: {e}")
return None
def save_evaluation_results(results: dict[str, Any], local_dir: str = LOCAL_EVALUATION_DIR) -> bool:
"""Save comprehensive evaluation results to local directory as a single JSON file."""
try:
local_path = Path(local_dir)
local_path.mkdir(parents=True, exist_ok=True)
model_name = results.get("model_name", "unknown")
safe_model_name = get_safe_model_name(model_name)
# Save single comprehensive results file (CodeSearchNet + Benchmark combined)
result_file = local_path / f"comprehensive_eval_{safe_model_name}.json"
with result_file.open("w") as f:
json.dump(results, f, indent=2, default=str)
logger.info(f"πΎ Saved comprehensive evaluation results for {model_name}")
return True
except Exception:
logger.exception("β Error saving evaluation results")
return False
def discover_local_models(models_dir: str = LOCAL_MODELS_DIR) -> list[str]:
"""Discover models in the local models directory."""
models_path = Path(models_dir)
discovered_models = []
if models_path.exists():
for model_dir in models_path.iterdir():
if model_dir.is_dir() and (
any(model_dir.glob("*.json")) or any(model_dir.glob("*.bin")) or any(model_dir.glob("*.safetensors"))
):
discovered_models.append(str(model_dir))
logger.info(f"π Found local model: {model_dir.name}")
return discovered_models
def print_results_summary(results: dict[str, Any]) -> None:
"""Print a formatted summary of comprehensive evaluation results."""
logger.info(f"\n{'=' * 60}")
logger.info(f"π COMPREHENSIVE EVALUATION: {results.get('model_name', 'Unknown')}")
logger.info(f"{'=' * 60}")
# CodeSearchNet results
overall = results.get("codesearch_overall", {})
if overall:
logger.info("π CodeSearchNet Performance:")
for metric, value in overall.items():
logger.info(f" π― {metric.upper()}: {value:.4f}")
# Benchmark results
if not results.get("benchmark_skipped", False):
size_metrics = results.get("size_metrics", {})
if size_metrics:
logger.info(f"\nπ Model Size: {size_metrics.get('disk_size_mb', 0):.1f}MB")
if "parameters_millions" in size_metrics:
logger.info(f"π’ Parameters: {size_metrics['parameters_millions']:.1f}M")
speed_benchmarks = results.get("speed_benchmarks", {})
if "medium" in speed_benchmarks and "batch_32" in speed_benchmarks["medium"]:
batch_32 = speed_benchmarks["medium"]["batch_32"]
logger.info(f"β‘ Throughput (batch 32): {batch_32.get('texts_per_second', 0):.1f} texts/sec")
# Language breakdown
languages = results.get("codesearch_languages", {})
if languages:
logger.info("\nπ Language Breakdown:")
for lang, lang_results in languages.items():
metrics = lang_results.get("metrics", {})
ndcg10 = metrics.get("ndcg@10", 0)
mrr = metrics.get("mrr", 0)
logger.info(f" {lang}: NDCG@10={ndcg10:.4f}, MRR={mrr:.4f}")
def create_comparison_report(all_results: list[dict[str, Any]], output_dir: str = LOCAL_EVALUATION_DIR) -> None:
"""Create a comprehensive comparison report with both CodeSearchNet and benchmark data."""
if not all_results:
return
logger.info("π Creating comprehensive comparison report...")
# Create evaluation comparison dataframe
evaluation_data = []
benchmark_data = []
for result in all_results:
model_name = result.get("model_name", "Unknown")
# CodeSearchNet data
overall = result.get("codesearch_overall", {})
eval_row = {"model_name": model_name}
eval_row.update(overall)
evaluation_data.append(eval_row)
# Benchmark data (if available)
if not result.get("benchmark_skipped", False):
benchmark_row = {"model_name": model_name}
size_metrics = result.get("size_metrics", {})
speed_benchmarks = result.get("speed_benchmarks", {})
benchmark_row.update(size_metrics)
if "medium" in speed_benchmarks and "batch_32" in speed_benchmarks["medium"]:
batch_32 = speed_benchmarks["medium"]["batch_32"]
benchmark_row["best_throughput"] = batch_32.get("texts_per_second", 0)
benchmark_data.append(benchmark_row)
# Save comparison results
output_path = Path(output_dir)
output_path.mkdir(parents=True, exist_ok=True)
# Combined evaluation comparison CSV (includes both CodeSearchNet and key benchmark metrics)
if evaluation_data and benchmark_data:
# Merge evaluation and benchmark data
combined_data = []
benchmark_dict = {row["model_name"]: row for row in benchmark_data}
for eval_row in evaluation_data:
model_name = eval_row["model_name"]
combined_row = eval_row.copy()
# Add benchmark metrics if available
if model_name in benchmark_dict:
benchmark_row = benchmark_dict[model_name]
combined_row.update(
{
"disk_size_mb": benchmark_row.get("disk_size_mb", 0),
"parameters_millions": benchmark_row.get("parameters_millions", 0),
"best_throughput": benchmark_row.get("best_throughput", 0),
}
)
combined_data.append(combined_row)
combined_df = pd.DataFrame(combined_data)
combined_csv = output_path / "comprehensive_comparison.csv"
combined_df.to_csv(combined_csv, index=False)
logger.info(f"π Comprehensive comparison CSV saved: {combined_csv}")
# Detailed JSON export
json_path = output_path / "comprehensive_evaluation.json"
with json_path.open("w") as f:
json.dump(all_results, f, indent=2, default=str)
logger.info(f"π Comprehensive results JSON saved: {json_path}")
# =============================================================================
# MAIN EVALUATION FUNCTIONS
# =============================================================================
def run_evaluation(
models: list[str],
max_queries: int = 100,
languages: list[str] | None = None,
use_beam: bool = False,
skip_benchmark: bool = False,
) -> list[dict[str, Any]]:
"""Main evaluation function that handles both local and Beam execution."""
logger.info(f"π Starting comprehensive evaluation ({'Beam' if use_beam else 'Local'})")
logger.info(f"π Evaluating {len(models)} models on {len(languages or languages_config.all)} languages")
logger.info(f"β‘ Benchmarking: {'Disabled' if skip_benchmark else 'Enabled'}")
# Check for existing results and skip already evaluated models
models_to_evaluate = []
skipped_models = []
all_results = []
for model_path in models:
model_name = Path(model_path).name
existing_results = check_existing_results(model_name)
if existing_results:
logger.info(f"β
Model {model_name} already evaluated, skipping")
all_results.append(existing_results)
skipped_models.append(model_name)
else:
models_to_evaluate.append(model_path)
if not models_to_evaluate:
logger.info("π All models already evaluated!")
return all_results
logger.info(f"π Need to evaluate {len(models_to_evaluate)} models")
if use_beam:
# Run on Beam
new_results = _run_beam_evaluation(models_to_evaluate, max_queries, languages, skip_benchmark)
else:
# Run locally
new_results = _run_local_evaluation(models_to_evaluate, max_queries, languages, skip_benchmark)
all_results.extend(new_results)
# Create comparison report
if len(all_results) > 1:
create_comparison_report(all_results)
# Print summary
newly_evaluated = len(new_results)
logger.info(f"\n{'=' * 60}")
logger.info("π EVALUATION SUMMARY")
logger.info(f"{'=' * 60}")
logger.info(f"π Total models: {len(models)}")
logger.info(f"β
Newly evaluated: {newly_evaluated}")
logger.info(f"βοΈ Skipped (already done): {len(skipped_models)}")
logger.info(f"π― Total results: {len(all_results)}")
logger.info(f"β‘ Benchmarking: {'Disabled' if skip_benchmark else 'Enabled'}")
return all_results
def _run_local_evaluation(
models: list[str],
max_queries: int = 100,
languages: list[str] | None = None,
skip_benchmark: bool = False,
) -> list[dict[str, Any]]:
"""Run comprehensive evaluation locally."""
logger.info("π₯οΈ Running local comprehensive evaluation")
results = []
for model_path in models:
model_name = Path(model_path).name
logger.info(f"\n{'=' * 60}")
logger.info(f"π Evaluating model: {model_name}")
logger.info(f"{'=' * 60}")
try:
evaluator = ComprehensiveModelEvaluator(model_path, model_name)
result = evaluator.run_comprehensive_evaluation(max_queries, languages, skip_benchmark)
# Save results locally
save_evaluation_results(result)
print_results_summary(result)
results.append(result)
except Exception:
logger.exception(f"β Failed to evaluate {model_name}")
continue
return results
@function(**get_evaluation_function_kwargs())
def _beam_evaluate_single_model(
model_path: str,
max_queries: int = 100,
languages: list[str] | None = None,
skip_benchmark: bool = False,
) -> dict[str, Any]:
"""Beam function to comprehensively evaluate a single model."""
# Set CUDA memory settings BEFORE any CUDA operations
import os
os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "expandable_segments:True"
model_name = Path(model_path).name
logger.info(f"π Beam comprehensive evaluation starting for {model_name}")
# Clear CUDA cache if available
try:
import torch
if torch.cuda.is_available():
torch.cuda.empty_cache()
torch.cuda.reset_peak_memory_stats()
logger.info(
f"π§Ή Cleared CUDA cache. Available memory: {torch.cuda.get_device_properties(0).total_memory // (1024**3)} GB"
)
except Exception as e:
logger.warning(f"β οΈ Could not clear CUDA cache: {e}")
try:
logger.info("π§ Creating ComprehensiveModelEvaluator...")
evaluator = ComprehensiveModelEvaluator(model_path, model_name)
logger.info("β
ComprehensiveModelEvaluator created successfully")
logger.info("π Starting comprehensive evaluation...")
results = evaluator.run_comprehensive_evaluation(max_queries, languages, skip_benchmark)
logger.info("β
Comprehensive evaluation completed")
# Validate results
if not results or "model_name" not in results:
logger.error(f"β Invalid evaluation results for {model_name}: {results}")
return {"error": "Invalid evaluation results", "model_name": model_name}
# Save to Beam volume as single comprehensive file
logger.info("πΎ Saving results to Beam volume...")
volume_results_dir = Path(VOLUME_CONFIG.mount_path) / "evaluation_results"
volume_results_dir.mkdir(parents=True, exist_ok=True)
safe_model_name = get_safe_model_name(model_name)
result_file = volume_results_dir / f"comprehensive_eval_{safe_model_name}.json"
with result_file.open("w") as f:
json.dump(results, f, indent=2, default=str)
logger.info(f"πΎ Saved Beam comprehensive evaluation results for {model_name} to {result_file}")
logger.info(f"π― Final results summary: {len(results.get('codesearch_languages', {}))} languages evaluated")
return results
except (torch.OutOfMemoryError, RuntimeError, AssertionError) as e:
# Handle CUDA errors including OOM, driver errors, and PyTorch internal assertion failures
is_oom = isinstance(e, torch.OutOfMemoryError) or "out of memory" in str(e).lower()
is_cuda_error = is_oom or "cuda" in str(e).lower() or "INTERNAL ASSERT FAILED" in str(e) or "handles_" in str(e)
if is_cuda_error:
error_type = "CUDA OOM" if is_oom else "CUDA Error"
logger.exception(f"β {error_type} during evaluation of {model_name}")
# Try to clear memory and reset CUDA state more aggressively
with contextlib.suppress(Exception):
reset_cuda_state()
return {
"error": f"{error_type}: {e!s}",
"model_name": model_name,
"timestamp": time.strftime("%Y-%m-%d %H:%M:%S"),
"evaluation_failed": True,
"oom": is_oom,
"cuda_error": True,
}
# Re-raise if not a CUDA-related error
raise
except Exception as e:
logger.exception(f"β Beam comprehensive evaluation failed for {model_name}")
# Return error info in a structured way
error_result = {
"error": str(e),
"model_name": model_name,
"timestamp": time.strftime("%Y-%m-%d %H:%M:%S"),
"evaluation_failed": True,
}
# Try to save error result to volume
try:
volume_results_dir = Path(VOLUME_CONFIG.mount_path) / "evaluation_results"
volume_results_dir.mkdir(parents=True, exist_ok=True)
safe_model_name = get_safe_model_name(model_name)
error_file = volume_results_dir / f"error_eval_{safe_model_name}.json"
with error_file.open("w") as f:
json.dump(error_result, f, indent=2, default=str)
logger.info(f"πΎ Saved error info to {error_file}")
except Exception:
logger.exception("β Could not save error info")
return error_result
def _run_beam_evaluation(
models: list[str],
max_queries: int = 100,
languages: list[str] | None = None,
skip_benchmark: bool = False,
) -> list[dict[str, Any]]:
"""Run comprehensive evaluation on Beam and download results."""
logger.info("βοΈ Running Beam comprehensive evaluation")
results = []
for model_path in models:
model_name = Path(model_path).name
logger.info(f"π Starting Beam comprehensive evaluation for {model_name}")
try:
# Run evaluation on Beam
result = _beam_evaluate_single_model.remote(model_path, max_queries, languages, skip_benchmark)
if result:
# Check if this is an error result
if result.get("evaluation_failed", False):
logger.error(f"β Beam evaluation failed for {model_name}: {result.get('error', 'Unknown error')}")
if result.get("oom", False):
logger.error("π₯ Out of memory error - model may be too large for available GPU")
continue
# Download the comprehensive result file from Beam
success = download_specific_evaluation_file(
VOLUME_CONFIG.name,
model_name,
"evaluation_results",
LOCAL_EVALUATION_DIR,
file_prefix="comprehensive_eval",
)
if success:
logger.info(f"π₯ Downloaded comprehensive results for {model_name}")
print_results_summary(result)
results.append(result)
else:
logger.warning(f"β οΈ Could not download results for {model_name}")
else:
logger.warning(f"β οΈ No result returned for {model_name}")
except Exception:
logger.exception(f"β Beam comprehensive evaluation failed for {model_name}")
continue
return results
# =============================================================================
# CLI INTERFACE
# =============================================================================
def main(
use_beam: bool = typer.Option(default=False, help="Use Beam for evaluation"),
skip_third_party: bool = typer.Option(default=False, help="Skip third-party models"),
skip_benchmark: bool = typer.Option(default=False, help="Skip performance benchmarking"),
max_queries: int = typer.Option(default=100, help="Maximum queries per language"),
) -> None:
"""Main comprehensive evaluation function."""
logger.info("π Starting comprehensive model evaluation (CodeSearchNet + Performance)")
# Build model list
models = []
# Add third-party models if not skipped
if not skip_third_party:
logger.info("π Including third-party peer models for comparison")
models.extend(DEFAULT_EVALUATION_MODELS)
else:
logger.info("βοΈ Skipping third-party models")
# Discover local models from code_model2vec/final
logger.info("π Discovering local distillation models...")
local_models = discover_local_models()
if local_models:
logger.info(f"β
Found {len(local_models)} local models:")
for model_path in local_models:
models.append(model_path)
logger.info(f" π {Path(model_path).name}")
else:
logger.warning("β οΈ No local distillation models found")
if skip_third_party:
logger.error("β No models to evaluate!")
return
if not models:
logger.error("β No models to evaluate!")
return
logger.info(f"π Will evaluate {len(models)} models:")
for i, model in enumerate(models, 1):
logger.info(f" {i}. {Path(model).name}")
# Run evaluation
results = run_evaluation(
models=models,
max_queries=max_queries,
languages=languages_config.all,
use_beam=use_beam,
skip_benchmark=skip_benchmark,
)
logger.info("π Comprehensive evaluation workflow completed!")
logger.info(f"π Successfully evaluated {len(results)} models")
logger.info(f"πΎ Results saved to: {LOCAL_EVALUATION_DIR}")
logger.info("π Format: Single comprehensive JSON per model (CodeSearchNet + Benchmarks)")
if __name__ == "__main__":
typer.run(main)
|