File size: 67,142 Bytes
167596f | 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 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 | # """
# Query functionality for RAGAnything
# Contains all query-related methods for both text and multimodal queries
# """
# import json
# import hashlib
# import re
# from typing import Dict, List, Any
# from pathlib import Path
# from lightrag import QueryParam
# from lightrag.utils import always_get_an_event_loop
# from raganything.prompt import PROMPTS
# from raganything.utils import (
# get_processor_for_type,
# encode_image_to_base64,
# validate_image_file,
# )
# # Add these imports
# from raganything.query_improvement import QueryImprovementMixin
# from raganything.verification import DualLLMVerificationMixin
# class QueryMixin(QueryImprovementMixin, DualLLMVerificationMixin):
# """QueryMixin class containing query functionality for RAGAnything"""
# def _generate_multimodal_cache_key(
# self, query: str, multimodal_content: List[Dict[str, Any]], mode: str, **kwargs
# ) -> str:
# """
# Generate cache key for multimodal query
# Args:
# query: Base query text
# multimodal_content: List of multimodal content
# mode: Query mode
# **kwargs: Additional parameters
# Returns:
# str: Cache key hash
# """
# # Create a normalized representation of the query parameters
# cache_data = {
# "query": query.strip(),
# "mode": mode,
# }
# # Normalize multimodal content for stable caching
# normalized_content = []
# if multimodal_content:
# for item in multimodal_content:
# if isinstance(item, dict):
# normalized_item = {}
# for key, value in item.items():
# # For file paths, use basename to make cache more portable
# if key in [
# "img_path",
# "image_path",
# "file_path",
# ] and isinstance(value, str):
# normalized_item[key] = Path(value).name
# # For large content, create a hash instead of storing directly
# elif (
# key in ["table_data", "table_body"]
# and isinstance(value, str)
# and len(value) > 200
# ):
# normalized_item[f"{key}_hash"] = hashlib.md5(
# value.encode()
# ).hexdigest()
# else:
# normalized_item[key] = value
# normalized_content.append(normalized_item)
# else:
# normalized_content.append(item)
# cache_data["multimodal_content"] = normalized_content
# # Add relevant kwargs to cache data
# relevant_kwargs = {
# k: v
# for k, v in kwargs.items()
# if k
# in [
# "stream",
# "response_type",
# "top_k",
# "max_tokens",
# "temperature",
# # "only_need_context",
# # "only_need_prompt",
# ]
# }
# cache_data.update(relevant_kwargs)
# # Generate hash from the cache data
# cache_str = json.dumps(cache_data, sort_keys=True, ensure_ascii=False)
# cache_hash = hashlib.md5(cache_str.encode()).hexdigest()
# return f"multimodal_query:{cache_hash}"
# # async def aquery(self, query: str, mode: str = "mix", **kwargs) -> str:
# # """
# # Pure text query - directly calls LightRAG's query functionality
# # Args:
# # query: Query text
# # mode: Query mode ("local", "global", "hybrid", "naive", "mix", "bypass")
# # **kwargs: Other query parameters, will be passed to QueryParam
# # - vlm_enhanced: bool, default True when vision_model_func is available.
# # If True, will parse image paths in retrieved context and replace them
# # with base64 encoded images for VLM processing.
# # Returns:
# # str: Query result
# # """
# # if self.lightrag is None:
# # raise ValueError(
# # "No LightRAG instance available. Please process documents first or provide a pre-initialized LightRAG instance."
# # )
# # # Check if VLM enhanced query should be used
# # vlm_enhanced = kwargs.pop("vlm_enhanced", None)
# # # Auto-determine VLM enhanced based on availability
# # if vlm_enhanced is None:
# # vlm_enhanced = (
# # hasattr(self, "vision_model_func")
# # and self.vision_model_func is not None
# # )
# # # Use VLM enhanced query if enabled and available
# # if (
# # vlm_enhanced
# # and hasattr(self, "vision_model_func")
# # and self.vision_model_func
# # ):
# # return await self.aquery_vlm_enhanced(query, mode=mode, **kwargs)
# # elif vlm_enhanced and (
# # not hasattr(self, "vision_model_func") or not self.vision_model_func
# # ):
# # self.logger.warning(
# # "VLM enhanced query requested but vision_model_func is not available, falling back to normal query"
# # )
# # # Create query parameters
# # query_param = QueryParam(mode=mode, **kwargs)
# # self.logger.info(f"Executing text query: {query[:100]}...")
# # self.logger.info(f"Query mode: {mode}")
# # # Call LightRAG's query method
# # result = await self.lightrag.aquery(query, param=query_param)
# # self.logger.info("Text query completed")
# # return result
# # async def aquery_with_multimodal(
# # self,
# # query: str,
# # multimodal_content: List[Dict[str, Any]] = None,
# # mode: str = "mix",
# # **kwargs,
# # ) -> str:
# # """
# # Multimodal query - combines text and multimodal content for querying
# # Args:
# # query: Base query text
# # multimodal_content: List of multimodal content, each element contains:
# # - type: Content type ("image", "table", "equation", etc.)
# # - Other fields depend on type (e.g., img_path, table_data, latex, etc.)
# # mode: Query mode ("local", "global", "hybrid", "naive", "mix", "bypass")
# # **kwargs: Other query parameters, will be passed to QueryParam
# # Returns:
# # str: Query result
# # Examples:
# # # Pure text query
# # result = await rag.query_with_multimodal("What is machine learning?")
# # # Image query
# # result = await rag.query_with_multimodal(
# # "Analyze the content in this image",
# # multimodal_content=[{
# # "type": "image",
# # "img_path": "./image.jpg"
# # }]
# # )
# # # Table query
# # result = await rag.query_with_multimodal(
# # "Analyze the data trends in this table",
# # multimodal_content=[{
# # "type": "table",
# # "table_data": "Name,Age\nAlice,25\nBob,30"
# # }]
# # )
# # """
# # # Ensure LightRAG is initialized
# # await self._ensure_lightrag_initialized()
# # self.logger.info(f"Executing multimodal query: {query[:100]}...")
# # self.logger.info(f"Query mode: {mode}")
# # # If no multimodal content, fallback to pure text query
# # if not multimodal_content:
# # self.logger.info("No multimodal content provided, executing text query")
# # return await self.aquery(query, mode=mode, **kwargs)
# # # Generate cache key for multimodal query
# # cache_key = self._generate_multimodal_cache_key(
# # query, multimodal_content, mode, **kwargs
# # )
# # # Check cache if available and enabled
# # cached_result = None
# # if (
# # hasattr(self, "lightrag")
# # and self.lightrag
# # and hasattr(self.lightrag, "llm_response_cache")
# # and self.lightrag.llm_response_cache
# # ):
# # if self.lightrag.llm_response_cache.global_config.get(
# # "enable_llm_cache", True
# # ):
# # try:
# # cached_result = await self.lightrag.llm_response_cache.get_by_id(
# # cache_key
# # )
# # if cached_result and isinstance(cached_result, dict):
# # result_content = cached_result.get("return")
# # if result_content:
# # self.logger.info(
# # f"Multimodal query cache hit: {cache_key[:16]}..."
# # )
# # return result_content
# # except Exception as e:
# # self.logger.debug(f"Error accessing multimodal query cache: {e}")
# # # Process multimodal content to generate enhanced query text
# # enhanced_query = await self._process_multimodal_query_content(
# # query, multimodal_content
# # )
# # self.logger.info(
# # f"Generated enhanced query length: {len(enhanced_query)} characters"
# # )
# # # Execute enhanced query
# # result = await self.aquery(enhanced_query, mode=mode, **kwargs)
# # # Save to cache if available and enabled
# # if (
# # hasattr(self, "lightrag")
# # and self.lightrag
# # and hasattr(self.lightrag, "llm_response_cache")
# # and self.lightrag.llm_response_cache
# # ):
# # if self.lightrag.llm_response_cache.global_config.get(
# # "enable_llm_cache", True
# # ):
# # try:
# # # Create cache entry for multimodal query
# # cache_entry = {
# # "return": result,
# # "cache_type": "multimodal_query",
# # "original_query": query,
# # "multimodal_content_count": len(multimodal_content),
# # "mode": mode,
# # }
# # await self.lightrag.llm_response_cache.upsert(
# # {cache_key: cache_entry}
# # )
# # self.logger.info(
# # f"Saved multimodal query result to cache: {cache_key[:16]}..."
# # )
# # except Exception as e:
# # self.logger.debug(f"Error saving multimodal query to cache: {e}")
# # # Ensure cache is persisted to disk
# # if (
# # hasattr(self, "lightrag")
# # and self.lightrag
# # and hasattr(self.lightrag, "llm_response_cache")
# # and self.lightrag.llm_response_cache
# # ):
# # try:
# # await self.lightrag.llm_response_cache.index_done_callback()
# # except Exception as e:
# # self.logger.debug(f"Error persisting multimodal query cache: {e}")
# # self.logger.info("Multimodal query completed")
# # return result
# async def aquery(self, query: str, mode: str = "mix", **kwargs) -> str:
# """
# Pure text query with optional query improvement and verification
# Args:
# query: Query text
# mode: Query mode ("local", "global", "hybrid", "naive", "mix", "bypass")
# **kwargs: Other query parameters
# - enable_query_improvement: bool, override config setting
# - enable_verification: bool, override config setting
# - return_verification_info: bool, return detailed verification info
# Returns:
# str: Query result (or dict if return_verification_info=True)
# """
# if self.lightrag is None:
# raise ValueError(
# "No LightRAG instance available. Please process documents first or provide a pre-initialized LightRAG instance."
# )
# # Check override flags
# use_query_improvement = kwargs.pop('enable_query_improvement',
# getattr(self.config, 'enable_query_improvement', False))
# use_verification = kwargs.pop('enable_verification',
# getattr(self.config, 'enable_dual_llm_verification', False))
# return_verification_info = kwargs.pop('return_verification_info', False)
# original_query = query
# query_improvement_result = None
# # Step 1: Apply query improvement if enabled
# if use_query_improvement and hasattr(self, 'query_improver') and self.query_improver:
# self.logger.info("Applying query improvement...")
# query_improvement_result = await self._apply_query_improvement(query)
# query = query_improvement_result["improved_query"]
# self.logger.info(f"Query improved: '{original_query[:50]}...' -> '{query[:50]}...'")
# # Step 2: Check VLM enhanced query
# vlm_enhanced = kwargs.pop("vlm_enhanced", None)
# if vlm_enhanced is None:
# vlm_enhanced = (
# hasattr(self, "vision_model_func") and self.vision_model_func is not None
# )
# # If using VLM enhanced or verification is disabled, use existing flow
# if vlm_enhanced or not use_verification:
# if vlm_enhanced and hasattr(self, "vision_model_func") and self.vision_model_func:
# result = await self.aquery_vlm_enhanced(query, mode=mode, **kwargs)
# else:
# from lightrag import QueryParam
# query_param = QueryParam(mode=mode, **kwargs)
# result = await self.lightrag.aquery(query, param=query_param)
# if return_verification_info:
# return {
# "answer": result,
# "original_query": original_query,
# "improved_query": query if query_improvement_result else original_query,
# "query_improvement": query_improvement_result,
# "verification_passed": True,
# "verification_score": 10.0
# }
# return result
# # Step 3: Generate with verification
# if use_verification and hasattr(self, 'answer_verifier') and self.answer_verifier:
# self.logger.info("Using dual-LLM verification...")
# # Get context without final answer
# from lightrag import QueryParam
# query_param = QueryParam(mode=mode, only_need_context=True, **kwargs)
# context = await self.lightrag.aquery(query, param=query_param)
# # Generate with verification
# verification_result = await self._generate_with_verification(
# query=query,
# context=context,
# original_query=original_query
# )
# if return_verification_info:
# return {
# "answer": verification_result["answer"],
# "original_query": original_query,
# "improved_query": query if query_improvement_result else original_query,
# "query_improvement": query_improvement_result,
# "verification_passed": verification_result["verification_passed"],
# "verification_score": verification_result["verification_score"],
# "modification_attempts": verification_result["modification_attempts"],
# "verification_history": verification_result.get("verification_history", [])
# }
# return verification_result["answer"]
# # Fallback to normal query
# from lightrag import QueryParam
# query_param = QueryParam(mode=mode, **kwargs)
# result = await self.lightrag.aquery(query, param=query_param)
# if return_verification_info:
# return {
# "answer": result,
# "original_query": original_query,
# "improved_query": query if query_improvement_result else original_query,
# "query_improvement": query_improvement_result
# }
# return result
# async def aquery_vlm_enhanced(self, query: str, mode: str = "mix", **kwargs) -> str:
# """
# VLM enhanced query - replaces image paths in retrieved context with base64 encoded images for VLM processing
# Args:
# query: User query
# mode: Underlying LightRAG query mode
# **kwargs: Other query parameters
# Returns:
# str: VLM query result
# """
# # Ensure VLM is available
# if not hasattr(self, "vision_model_func") or not self.vision_model_func:
# raise ValueError(
# "VLM enhanced query requires vision_model_func. "
# "Please provide a vision model function when initializing RAGAnything."
# )
# # Ensure LightRAG is initialized
# await self._ensure_lightrag_initialized()
# self.logger.info(f"Executing VLM enhanced query: {query[:100]}...")
# # Clear previous image cache
# if hasattr(self, "_current_images_base64"):
# delattr(self, "_current_images_base64")
# # 1. Get original retrieval prompt (without generating final answer)
# query_param = QueryParam(mode=mode, only_need_prompt=True, **kwargs)
# raw_prompt = await self.lightrag.aquery(query, param=query_param)
# self.logger.debug("Retrieved raw prompt from LightRAG")
# # 2. Extract and process image paths
# enhanced_prompt, images_found = await self._process_image_paths_for_vlm(
# raw_prompt
# )
# if not images_found:
# self.logger.info("No valid images found, falling back to normal query")
# # Fallback to normal query
# query_param = QueryParam(mode=mode, **kwargs)
# return await self.lightrag.aquery(query, param=query_param)
# self.logger.info(f"Processed {images_found} images for VLM")
# # 3. Build VLM message format
# messages = self._build_vlm_messages_with_images(enhanced_prompt, query)
# # 4. Call VLM for question answering
# result = await self._call_vlm_with_multimodal_content(messages)
# self.logger.info("VLM enhanced query completed")
# return result
# async def _process_multimodal_query_content(
# self, base_query: str, multimodal_content: List[Dict[str, Any]]
# ) -> str:
# """
# Process multimodal query content to generate enhanced query text
# Args:
# base_query: Base query text
# multimodal_content: List of multimodal content
# Returns:
# str: Enhanced query text
# """
# self.logger.info("Starting multimodal query content processing...")
# enhanced_parts = [f"User query: {base_query}"]
# for i, content in enumerate(multimodal_content):
# content_type = content.get("type", "unknown")
# self.logger.info(
# f"Processing {i+1}/{len(multimodal_content)} multimodal content: {content_type}"
# )
# try:
# # Get appropriate processor
# processor = get_processor_for_type(self.modal_processors, content_type)
# if processor:
# # Generate content description
# description = await self._generate_query_content_description(
# processor, content, content_type
# )
# enhanced_parts.append(
# f"\nRelated {content_type} content: {description}"
# )
# else:
# # If no appropriate processor, use basic description
# basic_desc = str(content)[:200]
# enhanced_parts.append(
# f"\nRelated {content_type} content: {basic_desc}"
# )
# except Exception as e:
# self.logger.error(f"Error processing multimodal content: {str(e)}")
# # Continue processing other content
# continue
# enhanced_query = "\n".join(enhanced_parts)
# enhanced_query += PROMPTS["QUERY_ENHANCEMENT_SUFFIX"]
# self.logger.info("Multimodal query content processing completed")
# return enhanced_query
# async def _generate_query_content_description(
# self, processor, content: Dict[str, Any], content_type: str
# ) -> str:
# """
# Generate content description for query
# Args:
# processor: Multimodal processor
# content: Content data
# content_type: Content type
# Returns:
# str: Content description
# """
# try:
# if content_type == "image":
# return await self._describe_image_for_query(processor, content)
# elif content_type == "table":
# return await self._describe_table_for_query(processor, content)
# elif content_type == "equation":
# return await self._describe_equation_for_query(processor, content)
# else:
# return await self._describe_generic_for_query(
# processor, content, content_type
# )
# except Exception as e:
# self.logger.error(f"Error generating {content_type} description: {str(e)}")
# return f"{content_type} content: {str(content)[:100]}"
# async def _describe_image_for_query(
# self, processor, content: Dict[str, Any]
# ) -> str:
# """Generate image description for query"""
# image_path = content.get("img_path")
# captions = content.get("image_caption", content.get("img_caption", []))
# footnotes = content.get("image_footnote", content.get("img_footnote", []))
# if image_path and Path(image_path).exists():
# # If image exists, use vision model to generate description
# image_base64 = processor._encode_image_to_base64(image_path)
# if image_base64:
# prompt = PROMPTS["QUERY_IMAGE_DESCRIPTION"]
# description = await processor.modal_caption_func(
# prompt,
# image_data=image_base64,
# system_prompt=PROMPTS["QUERY_IMAGE_ANALYST_SYSTEM"],
# )
# return description
# # If image doesn't exist or processing failed, use existing information
# parts = []
# if image_path:
# parts.append(f"Image path: {image_path}")
# if captions:
# parts.append(f"Image captions: {', '.join(captions)}")
# if footnotes:
# parts.append(f"Image footnotes: {', '.join(footnotes)}")
# return "; ".join(parts) if parts else "Image content information incomplete"
# async def _describe_table_for_query(
# self, processor, content: Dict[str, Any]
# ) -> str:
# """Generate table description for query"""
# table_data = content.get("table_data", "")
# table_caption = content.get("table_caption", "")
# prompt = PROMPTS["QUERY_TABLE_ANALYSIS"].format(
# table_data=table_data, table_caption=table_caption
# )
# description = await processor.modal_caption_func(
# prompt, system_prompt=PROMPTS["QUERY_TABLE_ANALYST_SYSTEM"]
# )
# return description
# async def _describe_equation_for_query(
# self, processor, content: Dict[str, Any]
# ) -> str:
# """Generate equation description for query"""
# latex = content.get("latex", "")
# equation_caption = content.get("equation_caption", "")
# prompt = PROMPTS["QUERY_EQUATION_ANALYSIS"].format(
# latex=latex, equation_caption=equation_caption
# )
# description = await processor.modal_caption_func(
# prompt, system_prompt=PROMPTS["QUERY_EQUATION_ANALYST_SYSTEM"]
# )
# return description
# async def _describe_generic_for_query(
# self, processor, content: Dict[str, Any], content_type: str
# ) -> str:
# """Generate generic content description for query"""
# content_str = str(content)
# prompt = PROMPTS["QUERY_GENERIC_ANALYSIS"].format(
# content_type=content_type, content_str=content_str
# )
# description = await processor.modal_caption_func(
# prompt,
# system_prompt=PROMPTS["QUERY_GENERIC_ANALYST_SYSTEM"].format(
# content_type=content_type
# ),
# )
# return description
# async def _process_image_paths_for_vlm(self, prompt: str) -> tuple[str, int]:
# """
# Process image paths in prompt, keeping original paths and adding VLM markers
# Args:
# prompt: Original prompt
# Returns:
# tuple: (processed prompt, image count)
# """
# enhanced_prompt = prompt
# images_processed = 0
# # Initialize image cache
# self._current_images_base64 = []
# # Enhanced regex pattern for matching image paths
# # Matches only the path ending with image file extensions
# image_path_pattern = (
# r"Image Path:\s*([^\r\n]*?\.(?:jpg|jpeg|png|gif|bmp|webp|tiff|tif))"
# )
# # First, let's see what matches we find
# matches = re.findall(image_path_pattern, prompt)
# self.logger.info(f"Found {len(matches)} image path matches in prompt")
# def replace_image_path(match):
# nonlocal images_processed
# image_path = match.group(1).strip()
# self.logger.debug(f"Processing image path: '{image_path}'")
# # Validate path format (basic check)
# if not image_path or len(image_path) < 3:
# self.logger.warning(f"Invalid image path format: {image_path}")
# return match.group(0) # Keep original
# # Use utility function to validate image file
# self.logger.debug(f"Calling validate_image_file for: {image_path}")
# is_valid = validate_image_file(image_path)
# self.logger.debug(f"Validation result for {image_path}: {is_valid}")
# if not is_valid:
# self.logger.warning(f"Image validation failed for: {image_path}")
# return match.group(0) # Keep original if validation fails
# try:
# # Encode image to base64 using utility function
# self.logger.debug(f"Attempting to encode image: {image_path}")
# image_base64 = encode_image_to_base64(image_path)
# if image_base64:
# images_processed += 1
# # Save base64 to instance variable for later use
# self._current_images_base64.append(image_base64)
# # Keep original path info and add VLM marker
# result = f"Image Path: {image_path}\n[VLM_IMAGE_{images_processed}]"
# self.logger.debug(
# f"Successfully processed image {images_processed}: {image_path}"
# )
# return result
# else:
# self.logger.error(f"Failed to encode image: {image_path}")
# return match.group(0) # Keep original if encoding failed
# except Exception as e:
# self.logger.error(f"Failed to process image {image_path}: {e}")
# return match.group(0) # Keep original
# # Execute replacement
# enhanced_prompt = re.sub(
# image_path_pattern, replace_image_path, enhanced_prompt
# )
# return enhanced_prompt, images_processed
# def _build_vlm_messages_with_images(
# self, enhanced_prompt: str, user_query: str
# ) -> List[Dict]:
# """
# Build VLM message format, using markers to correspond images with text positions
# Args:
# enhanced_prompt: Enhanced prompt with image markers
# user_query: User query
# Returns:
# List[Dict]: VLM message format
# """
# images_base64 = getattr(self, "_current_images_base64", [])
# if not images_base64:
# # Pure text mode
# return [
# {
# "role": "user",
# "content": f"Context:\n{enhanced_prompt}\n\nUser Question: {user_query}",
# }
# ]
# # Build multimodal content
# content_parts = []
# # Split text at image markers and insert images
# text_parts = enhanced_prompt.split("[VLM_IMAGE_")
# for i, text_part in enumerate(text_parts):
# if i == 0:
# # First text part
# if text_part.strip():
# content_parts.append({"type": "text", "text": text_part})
# else:
# # Find marker number and insert corresponding image
# marker_match = re.match(r"(\d+)\](.*)", text_part, re.DOTALL)
# if marker_match:
# image_num = (
# int(marker_match.group(1)) - 1
# ) # Convert to 0-based index
# remaining_text = marker_match.group(2)
# # Insert corresponding image
# if 0 <= image_num < len(images_base64):
# content_parts.append(
# {
# "type": "image_url",
# "image_url": {
# "url": f"data:image/jpeg;base64,{images_base64[image_num]}"
# },
# }
# )
# # Insert remaining text
# if remaining_text.strip():
# content_parts.append({"type": "text", "text": remaining_text})
# # Add user question
# content_parts.append(
# {
# "type": "text",
# "text": f"\n\nUser Question: {user_query}\n\nPlease answer based on the context and images provided.",
# }
# )
# return [
# {
# "role": "system",
# "content": "You are a helpful assistant that can analyze both text and image content to provide comprehensive answers.",
# },
# {"role": "user", "content": content_parts},
# ]
# async def _call_vlm_with_multimodal_content(self, messages: List[Dict]) -> str:
# """
# Call VLM to process multimodal content
# Args:
# messages: VLM message format
# Returns:
# str: VLM response result
# """
# try:
# user_message = messages[1]
# content = user_message["content"]
# system_prompt = messages[0]["content"]
# if isinstance(content, str):
# # Pure text mode
# result = await self.vision_model_func(
# content, system_prompt=system_prompt
# )
# else:
# # Multimodal mode - pass complete messages directly to VLM
# result = await self.vision_model_func(
# "", # Empty prompt since we're using messages format
# messages=messages,
# )
# return result
# except Exception as e:
# self.logger.error(f"VLM call failed: {e}")
# raise
# # Synchronous versions of query methods
# def query(self, query: str, mode: str = "mix", **kwargs) -> str:
# """
# Synchronous version of pure text query
# Args:
# query: Query text
# mode: Query mode ("local", "global", "hybrid", "naive", "mix", "bypass")
# **kwargs: Other query parameters, will be passed to QueryParam
# - vlm_enhanced: bool, default True when vision_model_func is available.
# If True, will parse image paths in retrieved context and replace them
# with base64 encoded images for VLM processing.
# Returns:
# str: Query result
# """
# loop = always_get_an_event_loop()
# return loop.run_until_complete(self.aquery(query, mode=mode, **kwargs))
# def query_with_multimodal(
# self,
# query: str,
# multimodal_content: List[Dict[str, Any]] = None,
# mode: str = "mix",
# **kwargs,
# ) -> str:
# """
# Synchronous version of multimodal query
# Args:
# query: Base query text
# multimodal_content: List of multimodal content, each element contains:
# - type: Content type ("image", "table", "equation", etc.)
# - Other fields depend on type (e.g., img_path, table_data, latex, etc.)
# mode: Query mode ("local", "global", "hybrid", "naive", "mix", "bypass")
# **kwargs: Other query parameters, will be passed to QueryParam
# Returns:
# str: Query result
# """
# loop = always_get_an_event_loop()
# return loop.run_until_complete(
# self.aquery_with_multimodal(query, multimodal_content, mode=mode, **kwargs)
# )
"""
Query functionality for RAGAnything - ENHANCED VERSION
Contains all query-related methods for text and multimodal queries,
plus query improvement and dual-LLM verification capabilities.
"""
import json
import hashlib
import re
import asyncio
from typing import Dict, List, Any
from pathlib import Path
from lightrag import QueryParam
from lightrag.utils import always_get_an_event_loop
from raganything.prompt import PROMPTS
from raganything.utils import (
get_processor_for_type,
encode_image_to_base64,
validate_image_file,
)
# Import new enhancement modules
from raganything.query_improvement import QueryImprovementMixin
from raganything.verification import DualLLMVerificationMixin
from raganything.streaming import StreamingQueryMixin
class QueryMixin(QueryImprovementMixin, DualLLMVerificationMixin, StreamingQueryMixin):
"""
QueryMixin class containing query functionality for RAGAnything
Enhanced with:
- Query improvement (rewriting, expansion, decomposition)
- Dual-LLM verification system
- Answer modification based on feedback
- Real-time streaming with verification support
"""
def _generate_multimodal_cache_key(
self, query: str, multimodal_content: List[Dict[str, Any]], mode: str, **kwargs
) -> str:
"""
Generate cache key for multimodal query
Args:
query: Base query text
multimodal_content: List of multimodal content
mode: Query mode
**kwargs: Additional parameters
Returns:
str: Cache key hash
"""
# Create a normalized representation of the query parameters
cache_data = {
"query": query.strip(),
"mode": mode,
}
# Normalize multimodal content for stable caching
normalized_content = []
if multimodal_content:
for item in multimodal_content:
if isinstance(item, dict):
normalized_item = {}
for key, value in item.items():
# For file paths, use basename to make cache more portable
if key in [
"img_path",
"image_path",
"file_path",
] and isinstance(value, str):
normalized_item[key] = Path(value).name
# For large content, create a hash instead of storing directly
elif (
key in ["table_data", "table_body"]
and isinstance(value, str)
and len(value) > 200
):
normalized_item[f"{key}_hash"] = hashlib.md5(
value.encode()
).hexdigest()
else:
normalized_item[key] = value
normalized_content.append(normalized_item)
else:
normalized_content.append(item)
cache_data["multimodal_content"] = normalized_content
# Add relevant kwargs to cache data
relevant_kwargs = {
k: v
for k, v in kwargs.items()
if k
in [
"stream",
"response_type",
"top_k",
"max_tokens",
"temperature",
]
}
cache_data.update(relevant_kwargs)
# Generate hash from the cache data
cache_str = json.dumps(cache_data, sort_keys=True, ensure_ascii=False)
cache_hash = hashlib.md5(cache_str.encode()).hexdigest()
return f"multimodal_query:{cache_hash}"
async def aquery(self, query: str, mode: str = "mix", **kwargs) -> str:
"""
Pure text query with optional query improvement and verification
Args:
query: Query text
mode: Query mode ("local", "global", "hybrid", "naive", "mix", "bypass")
**kwargs: Other query parameters
- vlm_enhanced: bool, default True when vision_model_func is available
- enable_query_improvement: bool, override config setting
- enable_verification: bool, override config setting
- return_verification_info: bool, return detailed verification info
Returns:
str: Query result (or dict if return_verification_info=True)
"""
if self.lightrag is None:
raise ValueError(
"No LightRAG instance available. Please process documents first or provide a pre-initialized LightRAG instance."
)
# Check override flags
use_query_improvement = kwargs.pop('enable_query_improvement',
getattr(self.config, 'enable_query_improvement', False))
use_verification = kwargs.pop('enable_verification',
getattr(self.config, 'enable_dual_llm_verification', False))
return_verification_info = kwargs.pop('return_verification_info', False)
original_query = query
query_improvement_result = None
# Step 1: Apply query improvement if enabled
if use_query_improvement and hasattr(self, 'query_improver') and self.query_improver:
self.logger.info("Applying query improvement...")
query_improvement_result = await self._apply_query_improvement(query)
if not query_improvement_result["improved_query"]:
self.logger.warning("Query improvement resulted in an empty query, using original query.")
query = original_query
else:
query = query_improvement_result["improved_query"]
self.logger.info(f"Query improved: '{original_query[:50]}...' -> '{query[:50]}...'")
# Check if VLM enhanced query should be used
vlm_enhanced = kwargs.pop("vlm_enhanced", None)
# Auto-determine VLM enhanced based on availability
if vlm_enhanced is None:
vlm_enhanced = (
hasattr(self, "vision_model_func")
and self.vision_model_func is not None
)
# If using VLM enhanced or verification is disabled, use existing flow
if vlm_enhanced or not use_verification:
# Use VLM enhanced query if enabled and available
if (
vlm_enhanced
and hasattr(self, "vision_model_func")
and self.vision_model_func
):
result = await self.aquery_vlm_enhanced(query, mode=mode, **kwargs)
elif vlm_enhanced and (
not hasattr(self, "vision_model_func") or not self.vision_model_func
):
self.logger.warning(
"VLM enhanced query requested but vision_model_func is not available, falling back to normal query"
)
# Create query parameters
query_param = QueryParam(mode=mode, **kwargs)
# Call LightRAG's query method
result = await self.lightrag.aquery(query, param=query_param)
else:
# Create query parameters
query_param = QueryParam(mode=mode, **kwargs)
# Call LightRAG's query method
result = await self.lightrag.aquery(query, param=query_param)
# Handle None result from LightRAG
if result is None:
result = "I couldn't find any relevant information in the knowledge base to answer your question."
# Return with verification info if requested
if return_verification_info:
return {
"answer": result,
"original_query": original_query,
"improved_query": query if query_improvement_result else original_query,
"query_improvement": query_improvement_result,
"verification_passed": True,
"verification_score": 10.0,
"modification_attempts": 0
}
self.logger.info("Query completed")
return result
# Step 2: Generate with verification if enabled
if use_verification and hasattr(self, 'answer_verifier') and self.answer_verifier:
self.logger.info("Using dual-LLM verification...")
# Get context without final answer
query_param = QueryParam(mode=mode, only_need_context=True, **kwargs)
context = await self.lightrag.aquery(query, param=query_param)
# Check if context is None or empty
if context is None or (isinstance(context, str) and not context.strip()):
self.logger.warning("No context retrieved from knowledge base")
no_context_answer = "I couldn't find any relevant information in the knowledge base to answer your question."
if return_verification_info:
return {
"answer": no_context_answer,
"original_query": original_query,
"improved_query": query if query_improvement_result else original_query,
"query_improvement": query_improvement_result,
"verification_passed": False,
"verification_score": 0.0,
"modification_attempts": 0,
"verification_history": []
}
return no_context_answer
# Generate with verification
verification_result = await self._generate_with_verification(
query=query,
context=context,
original_query=original_query
)
if return_verification_info:
return {
"answer": verification_result["answer"],
"original_query": original_query,
"improved_query": query if query_improvement_result else original_query,
"query_improvement": query_improvement_result,
"verification_passed": verification_result["verification_passed"],
"verification_score": verification_result["verification_score"],
"modification_attempts": verification_result["modification_attempts"],
"verification_history": verification_result.get("verification_history", [])
}
self.logger.info("Verified query completed")
return verification_result["answer"]
# Fallback to normal query
query_param = QueryParam(mode=mode, **kwargs)
result = await self.lightrag.aquery(query, param=query_param)
# Handle None result from LightRAG
if result is None:
result = "I couldn't find any relevant information in the knowledge base to answer your question."
if return_verification_info:
return {
"answer": result,
"original_query": original_query,
"improved_query": query if query_improvement_result else original_query,
"query_improvement": query_improvement_result,
"verification_passed": True,
"verification_score": 10.0,
"modification_attempts": 0
}
self.logger.info("Query completed")
return result
async def aquery_with_multimodal(
self,
query: str,
multimodal_content: List[Dict[str, Any]] = None,
mode: str = "mix",
**kwargs,
) -> str:
"""
Multimodal query - combines text and multimodal content for querying
Args:
query: Base query text
multimodal_content: List of multimodal content
mode: Query mode
**kwargs: Other query parameters
Returns:
str: Query result
"""
# Ensure LightRAG is initialized
await self._ensure_lightrag_initialized()
self.logger.info(f"Executing multimodal query: {query[:100]}...")
self.logger.info(f"Query mode: {mode}")
# If no multimodal content, fallback to pure text query
if not multimodal_content:
self.logger.info("No multimodal content provided, executing text query")
return await self.aquery(query, mode=mode, **kwargs)
# Generate cache key for multimodal query
cache_key = self._generate_multimodal_cache_key(
query, multimodal_content, mode, **kwargs
)
# Check cache if available and enabled
cached_result = None
if (
hasattr(self, "lightrag")
and self.lightrag
and hasattr(self.lightrag, "llm_response_cache")
and self.lightrag.llm_response_cache
):
if self.lightrag.llm_response_cache.global_config.get(
"enable_llm_cache", True
):
try:
cached_result = await self.lightrag.llm_response_cache.get_by_id(
cache_key
)
if cached_result and isinstance(cached_result, dict):
result_content = cached_result.get("return")
if result_content:
self.logger.info(
f"Multimodal query cache hit: {cache_key[:16]}..."
)
return result_content
except Exception as e:
self.logger.debug(f"Error accessing multimodal query cache: {e}")
# Process multimodal content to generate enhanced query text
enhanced_query = await self._process_multimodal_query_content(
query, multimodal_content
)
self.logger.info(
f"Generated enhanced query length: {len(enhanced_query)} characters"
)
# Execute enhanced query
result = await self.aquery(enhanced_query, mode=mode, **kwargs)
# Save to cache if available and enabled
if (
hasattr(self, "lightrag")
and self.lightrag
and hasattr(self.lightrag, "llm_response_cache")
and self.lightrag.llm_response_cache
):
if self.lightrag.llm_response_cache.global_config.get(
"enable_llm_cache", True
):
try:
# Create cache entry for multimodal query
cache_entry = {
"return": result,
"cache_type": "multimodal_query",
"original_query": query,
"multimodal_content_count": len(multimodal_content),
"mode": mode,
}
await self.lightrag.llm_response_cache.upsert(
{cache_key: cache_entry}
)
self.logger.info(
f"Saved multimodal query result to cache: {cache_key[:16]}..."
)
except Exception as e:
self.logger.debug(f"Error saving multimodal query to cache: {e}")
# Ensure cache is persisted to disk
if (
hasattr(self, "lightrag")
and self.lightrag
and hasattr(self.lightrag, "llm_response_cache")
and self.lightrag.llm_response_cache
):
try:
await self.lightrag.llm_response_cache.index_done_callback()
except Exception as e:
self.logger.debug(f"Error persisting multimodal query cache: {e}")
self.logger.info("Multimodal query completed")
return result
async def aquery_vlm_enhanced(self, query: str, mode: str = "mix", **kwargs) -> str:
"""
VLM enhanced query - replaces image paths in retrieved context with base64 encoded images
Args:
query: User query
mode: Underlying LightRAG query mode
**kwargs: Other query parameters
Returns:
str: VLM query result
"""
# Ensure VLM is available
if not hasattr(self, "vision_model_func") or not self.vision_model_func:
raise ValueError(
"VLM enhanced query requires vision_model_func. "
"Please provide a vision model function when initializing RAGAnything."
)
# Ensure LightRAG is initialized
await self._ensure_lightrag_initialized()
self.logger.info(f"Executing VLM enhanced query: {query[:100]}...")
# Clear previous image cache
if hasattr(self, "_current_images_base64"):
delattr(self, "_current_images_base64")
# 1. Get original retrieval prompt (without generating final answer)
self.logger.info(f"Getting raw prompt for query: {query[:100]}...")
query_param = QueryParam(mode=mode, only_need_prompt=True, **kwargs)
try:
raw_prompt = await self.lightrag.aquery(query, param=query_param)
except Exception as e:
self.logger.error(f"Error in self.lightrag.aquery: {e}", exc_info=True)
raw_prompt = None
self.logger.info(f"Retrieved raw prompt: {str(raw_prompt)[:200]}...")
if raw_prompt is None:
self.logger.warning("raw_prompt is None, falling back to normal query (single pass)")
query_param = QueryParam(mode=mode, **kwargs)
return await self.lightrag.aquery(query, param=query_param)
self.logger.debug("Retrieved raw prompt from LightRAG")
# 2. Extract and process image paths
enhanced_prompt, images_found = await self._process_image_paths_for_vlm(
raw_prompt
)
if not images_found:
self.logger.info("No valid images found, falling back to normal query WITHOUT re-retrieval")
# OPTIMIZATION: Reuse the already-retrieved context instead of querying again
# The raw_prompt already contains the full RAG context, so we can use it directly
# Try to use the existing model function if available
if hasattr(self.lightrag, 'llm_model_func') and self.lightrag.llm_model_func:
try:
# Generate answer using the already-retrieved context
self.logger.info("Generating answer from cached context (avoiding re-query)")
# Call the LLM with the raw prompt directly
if asyncio.iscoroutinefunction(self.lightrag.llm_model_func):
result = await self.lightrag.llm_model_func(raw_prompt)
else:
result = self.lightrag.llm_model_func(raw_prompt)
self.logger.info("Successfully generated answer from cached context (no re-query)")
return result
except Exception as e:
self.logger.warning(f"Failed to use cached context, falling back to re-query: {e}")
# Fall back to re-query if direct generation fails
query_param = QueryParam(mode=mode, **kwargs)
return await self.lightrag.aquery(query, param=query_param)
else:
# No model_func available, must re-query (original behavior)
# This maintains backward compatibility
self.logger.debug("llm_model_func not available, using standard re-query")
query_param = QueryParam(mode=mode, **kwargs)
return await self.lightrag.aquery(query, param=query_param)
self.logger.info(f"Processed {images_found} images for VLM")
# 3. Build VLM message format
messages = self._build_vlm_messages_with_images(enhanced_prompt, query)
# 4. Call VLM for question answering
result = await self._call_vlm_with_multimodal_content(messages)
self.logger.info("VLM enhanced query completed")
return result
# ... (rest of the existing methods remain the same) ...
async def _process_multimodal_query_content(
self, base_query: str, multimodal_content: List[Dict[str, Any]]
) -> str:
"""Process multimodal query content to generate enhanced query text"""
self.logger.info("Starting multimodal query content processing...")
enhanced_parts = [f"User query: {base_query}"]
for i, content in enumerate(multimodal_content):
content_type = content.get("type", "unknown")
self.logger.info(
f"Processing {i+1}/{len(multimodal_content)} multimodal content: {content_type}"
)
try:
# Get appropriate processor
processor = get_processor_for_type(self.modal_processors, content_type)
if processor:
# Generate content description
description = await self._generate_query_content_description(
processor, content, content_type
)
enhanced_parts.append(
f"\nRelated {content_type} content: {description}"
)
else:
# If no appropriate processor, use basic description
basic_desc = str(content)[:200]
enhanced_parts.append(
f"\nRelated {content_type} content: {basic_desc}"
)
except Exception as e:
self.logger.error(f"Error processing multimodal content: {str(e)}")
continue
enhanced_query = "\n".join(enhanced_parts)
enhanced_query += PROMPTS["QUERY_ENHANCEMENT_SUFFIX"]
self.logger.info("Multimodal query content processing completed")
return enhanced_query
async def _generate_query_content_description(
self, processor, content: Dict[str, Any], content_type: str
) -> str:
"""Generate content description for query"""
try:
if content_type == "image":
return await self._describe_image_for_query(processor, content)
elif content_type == "table":
return await self._describe_table_for_query(processor, content)
elif content_type == "equation":
return await self._describe_equation_for_query(processor, content)
else:
return await self._describe_generic_for_query(
processor, content, content_type
)
except Exception as e:
self.logger.error(f"Error generating {content_type} description: {str(e)}")
return f"{content_type} content: {str(content)[:100]}"
async def _describe_image_for_query(
self, processor, content: Dict[str, Any]
) -> str:
"""Generate image description for query"""
image_path = content.get("img_path")
captions = content.get("image_caption", content.get("img_caption", []))
footnotes = content.get("image_footnote", content.get("img_footnote", []))
if image_path and Path(image_path).exists():
image_base64 = processor._encode_image_to_base64(image_path)
if image_base64:
prompt = PROMPTS["QUERY_IMAGE_DESCRIPTION"]
description = await processor.modal_caption_func(
prompt,
image_data=image_base64,
system_prompt=PROMPTS["QUERY_IMAGE_ANALYST_SYSTEM"],
)
return description
parts = []
if image_path:
parts.append(f"Image path: {image_path}")
if captions:
parts.append(f"Image captions: {', '.join(captions)}")
if footnotes:
parts.append(f"Image footnotes: {', '.join(footnotes)}")
return "; ".join(parts) if parts else "Image content information incomplete"
async def _describe_table_for_query(
self, processor, content: Dict[str, Any]
) -> str:
"""Generate table description for query"""
table_data = content.get("table_data", "")
table_caption = content.get("table_caption", "")
prompt = PROMPTS["QUERY_TABLE_ANALYSIS"].format(
table_data=table_data, table_caption=table_caption
)
description = await processor.modal_caption_func(
prompt, system_prompt=PROMPTS["QUERY_TABLE_ANALYST_SYSTEM"]
)
return description
async def _describe_equation_for_query(
self, processor, content: Dict[str, Any]
) -> str:
"""Generate equation description for query"""
latex = content.get("latex", "")
equation_caption = content.get("equation_caption", "")
prompt = PROMPTS["QUERY_EQUATION_ANALYSIS"].format(
latex=latex, equation_caption=equation_caption
)
description = await processor.modal_caption_func(
prompt, system_prompt=PROMPTS["QUERY_EQUATION_ANALYST_SYSTEM"]
)
return description
async def _describe_generic_for_query(
self, processor, content: Dict[str, Any], content_type: str
) -> str:
"""Generate generic content description for query"""
content_str = str(content)
prompt = PROMPTS["QUERY_GENERIC_ANALYSIS"].format(
content_type=content_type, content_str=content_str
)
description = await processor.modal_caption_func(
prompt,
system_prompt=PROMPTS["QUERY_GENERIC_ANALYST_SYSTEM"].format(
content_type=content_type
),
)
return description
async def _process_image_paths_for_vlm(self, prompt: str) -> tuple[str, int]:
"""Process image paths in prompt, keeping original paths and adding VLM markers"""
if prompt is None:
self.logger.warning("prompt is None in _process_image_paths_for_vlm, returning as is")
return prompt, 0
enhanced_prompt = prompt
images_processed = 0
self._current_images_base64 = []
image_path_pattern = (
r"Image Path:\s*([^\r\n]*?\.(?:jpg|jpeg|png|gif|bmp|webp|tiff|tif))"
)
matches = re.findall(image_path_pattern, prompt)
self.logger.info(f"Found {len(matches)} image path matches in prompt")
def replace_image_path(match):
nonlocal images_processed
image_path = match.group(1).strip()
self.logger.debug(f"Processing image path: '{image_path}'")
if not image_path or len(image_path) < 3:
self.logger.warning(f"Invalid image path format: {image_path}")
return match.group(0)
self.logger.debug(f"Calling validate_image_file for: {image_path}")
is_valid = validate_image_file(image_path)
self.logger.debug(f"Validation result for {image_path}: {is_valid}")
if not is_valid:
self.logger.warning(f"Image validation failed for: {image_path}")
return match.group(0)
try:
self.logger.debug(f"Attempting to encode image: {image_path}")
image_base64 = encode_image_to_base64(image_path)
if image_base64:
images_processed += 1
self._current_images_base64.append(image_base64)
result = f"Image Path: {image_path}\n[VLM_IMAGE_{images_processed}]"
self.logger.debug(
f"Successfully processed image {images_processed}: {image_path}"
)
return result
else:
self.logger.error(f"Failed to encode image: {image_path}")
return match.group(0)
except Exception as e:
self.logger.error(f"Failed to process image {image_path}: {e}")
return match.group(0)
enhanced_prompt = re.sub(
image_path_pattern, replace_image_path, enhanced_prompt
)
return enhanced_prompt, images_processed
def _build_vlm_messages_with_images(
self, enhanced_prompt: str, user_query: str
) -> List[Dict]:
"""Build VLM message format, using markers to correspond images with text positions"""
images_base64 = getattr(self, "_current_images_base64", [])
if not images_base64:
return [
{
"role": "user",
"content": f"Context:\n{enhanced_prompt}\n\nUser Question: {user_query}",
}
]
content_parts = []
text_parts = enhanced_prompt.split("[VLM_IMAGE_")
for i, text_part in enumerate(text_parts):
if i == 0:
if text_part.strip():
content_parts.append({"type": "text", "text": text_part})
else:
marker_match = re.match(r"(\d+)\](.*)", text_part, re.DOTALL)
if marker_match:
image_num = int(marker_match.group(1)) - 1
remaining_text = marker_match.group(2)
if 0 <= image_num < len(images_base64):
content_parts.append(
{
"type": "image_url",
"image_url": {
"url": f"data:image/jpeg;base64,{images_base64[image_num]}"
},
}
)
if remaining_text.strip():
content_parts.append({"type": "text", "text": remaining_text})
content_parts.append(
{
"type": "text",
"text": f"\n\nUser Question: {user_query}\n\nPlease answer based on the context and images provided.",
}
)
return [
{
"role": "system",
"content": "You are a helpful assistant that can analyze both text and image content to provide comprehensive answers.",
},
{"role": "user", "content": content_parts},
]
async def _call_vlm_with_multimodal_content(self, messages: List[Dict]) -> str:
"""Call VLM to process multimodal content"""
try:
user_message = messages[1]
content = user_message["content"]
system_prompt = messages[0]["content"]
if isinstance(content, str):
result = await self.vision_model_func(
content, system_prompt=system_prompt
)
else:
result = await self.vision_model_func(
"",
messages=messages,
)
return result
except Exception as e:
self.logger.error(f"VLM call failed: {e}")
raise
# Synchronous versions of query methods
def query(self, query: str, mode: str = "mix", **kwargs) -> str:
"""Synchronous version of pure text query"""
loop = always_get_an_event_loop()
return loop.run_until_complete(self.aquery(query, mode=mode, **kwargs))
def query_with_multimodal(
self,
query: str,
multimodal_content: List[Dict[str, Any]] = None,
mode: str = "mix",
**kwargs,
) -> str:
"""Synchronous version of multimodal query"""
loop = always_get_an_event_loop()
return loop.run_until_complete(
self.aquery_with_multimodal(query, multimodal_content, mode=mode, **kwargs)
) |