File size: 92,809 Bytes
5c18e4c |
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 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 |
import argparse
import math
import os
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.utils.data import DataLoader
import copy
from torch.optim.lr_scheduler import CosineAnnealingLR
from torch.cuda.amp import autocast, GradScaler
from datasets import load_dataset
from transformers import AutoTokenizer
from typing import List, Tuple
import sys
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
def parse_args():
parser = argparse.ArgumentParser(description='Train or Inference with World Model and Tree of Thought.')
parser.add_argument('--model_name', type=str, default='gpt2', help='Pretrained model name or path')
parser.add_argument('--dataset_name', type=str, default='wikitext', help='Dataset name from HuggingFace Datasets')
parser.add_argument('--dataset_config', type=str, default='wikitext-2-raw-v1', help='Dataset configuration name')
parser.add_argument('--batch_size', type=int, default=4, help='Batch size')
parser.add_argument('--num_epochs', type=int, default=3, help='Number of epochs')
parser.add_argument('--max_length', type=int, default=128, help='Maximum sequence length')
parser.add_argument('--mcts_iterations', type=int, default=3, help='Number of MCTS Iterations')
parser.add_argument('--mcts_exploration_constant', type=float, default=1.414, help='Exploration constant for MCTS')
parser.add_argument('--accumulation_steps', type=int, default=4, help='Gradient accumulation steps')
parser.add_argument('--learning_rate', type=float, default=1e-4, help='Learning rate')
parser.add_argument('--weight_decay', type=float, default=1e-2, help='Weight decay')
parser.add_argument('--alpha', type=float, default=0.1, help='Entropy regularization weight')
parser.add_argument('--beta', type=float, default=0.1, help='Variance regularization weight')
parser.add_argument('--max_grad_norm', type=float, default=1.0, help='Max gradient norm for clipping')
parser.add_argument('--save_dir', type=str, default='./models', help='Directory to save the models')
parser.add_argument('--temperature', type=float, default=1.0, help='Temperature parameter for entropy and variance')
parser.add_argument('--mode', type=str, choices=['train', 'inference'], default='train', help='Mode: train or inference')
parser.add_argument('--inference_mode', type=str, choices=['world_model', 'without_world_model', 'world_model_tree_of_thought'], default='world_model_tree_of_thought', help='Inference mode')
parser.add_argument('--query', type=str, default='', help='Input query for inference')
parser.add_argument('--train_mode', type=str, choices=['world_model', 'language_model'], default='language_model', help='Train world model or language model only')
parser.add_argument('--beam_size', type=int, default=5, help='Beam size for beam search')
parser.add_argument('--n_tokens_predict', type=int, default=3, help='Number of tokens to predict at each step')
parser.add_argument('--load_model', type=str, default=None,
help='Path to load saved model. If not provided, a new model will be initialized.')
parser.add_argument('--use_custom_data', action='store_true', help='Use custom data for training')
# Determine the base directory
if hasattr(sys, 'frozen') and hasattr(sys, '_MEIPASS'):
# PyInstaller creates a temp folder and stores path in _MEIPASS
base_dir = sys._MEIPASS
elif '__file__' in globals():
# Running as a script
base_dir = os.path.dirname(os.path.abspath(__file__))
else:
# Running in an interactive environment (e.g., Jupyter, Colab)
base_dir = os.getcwd()
default_paths = [
'/content/drive/MyDrive/lightbulb/knowledge_base.json',
'/content/drive/MyDrive/lightbulb/rag_cache.json',
'/content/drive/MyDrive/lightbulb/llm_training_data/llm_training_data.jsonl'
]
parser.add_argument('--custom_data_paths', nargs='+', default=default_paths,
help='Paths to custom data files (relative to the script location or current working directory)')
args, unknown = parser.parse_known_args()
# Convert relative paths to absolute paths
args.custom_data_paths = [os.path.abspath(os.path.join(base_dir, path)) for path in args.custom_data_paths]
return args
import json
import jsonlines
def load_custom_data_from_files(file_paths):
custom_data = []
for file_path in file_paths:
if file_path.endswith('.json'):
with open(file_path, 'r') as f:
data = json.load(f)
if isinstance(data, list):
custom_data.extend(data)
else:
custom_data.append(data)
elif file_path.endswith('.jsonl'):
with jsonlines.open(file_path) as reader:
custom_data.extend(reader)
return custom_data
def preprocess_custom_data(data_list):
processed_data = []
for item in data_list:
# Check if the item is a string (JSON)
if isinstance(item, str):
try:
item = json.loads(item)
except json.JSONDecodeError:
print(f"Failed to parse JSON: {item[:100]}...") # Print first 100 chars for debugging
continue # Skip this item if it's not valid JSON
# Process query and content
query = item.get('query', '')
content = item.get('content', '')
if content == "RAG response generation failed.":
content = ""
# Combine query and content
combined_text = f"Query: {query} Content: {content}"
# Process numerical data (assuming these are available in the item dict)
episode_reward = item.get('episode_reward', 0)
loss = item.get('loss', 0)
cosine_similarity = item.get('cosine_similarity', 0)
rag_performance = item.get('rag_performance', 0)
ranking_model_performance = item.get('ranking_model_performance', 0)
# Create a dictionary with processed data
processed_item = {
'text': combined_text,
'episode_reward': episode_reward,
'loss': loss,
'cosine_similarity': cosine_similarity,
'rag_performance': rag_performance,
'ranking_model_performance': ranking_model_performance
}
processed_data.append(processed_item)
return processed_data
def load_custom_data(args, tokenizer, custom_data):
# Preprocess the custom data
processed_data = preprocess_custom_data(custom_data)
# Create a custom dataset
class CustomDataset(torch.utils.data.Dataset):
def __init__(self, data, tokenizer, max_length):
self.data = data
self.tokenizer = tokenizer
self.max_length = max_length
def __len__(self):
return len(self.data)
def __getitem__(self, idx):
item = self.data[idx]
encoded = self.tokenizer.encode_plus(
item['text'],
max_length=self.max_length,
padding='max_length',
truncation=True,
return_tensors='pt'
)
return {
'input_ids': encoded['input_ids'].squeeze(),
'attention_mask': encoded['attention_mask'].squeeze(),
'episode_reward': torch.tensor(item['episode_reward'], dtype=torch.float),
'loss': torch.tensor(item['loss'], dtype=torch.float),
'cosine_similarity': torch.tensor(item['cosine_similarity'], dtype=torch.float),
'rag_performance': torch.tensor(item['rag_performance'], dtype=torch.float),
'ranking_model_performance': torch.tensor(item['ranking_model_performance'], dtype=torch.float)
}
# Create dataset and dataloader
dataset = CustomDataset(processed_data, tokenizer, args.max_length)
# Split the dataset into train and eval
train_size = int(0.8 * len(dataset))
eval_size = len(dataset) - train_size
train_dataset, eval_dataset = torch.utils.data.random_split(dataset, [train_size, eval_size])
train_loader = DataLoader(
train_dataset,
batch_size=args.batch_size,
shuffle=True,
num_workers=4
)
eval_loader = DataLoader(
eval_dataset,
batch_size=args.batch_size,
shuffle=False,
num_workers=4
)
return train_loader, eval_loader
def load_data(args, tokenizer):
# Load the dataset
dataset = load_dataset(args.dataset_name, args.dataset_config)
# Ensure the tokenizer has a padding token
if tokenizer.pad_token is None:
tokenizer.pad_token = tokenizer.eos_token
def tokenize_function(examples):
return tokenizer(examples['text'], truncation=True, max_length=args.max_length)
tokenized_datasets = dataset.map(
tokenize_function,
batched=True,
num_proc=4,
remove_columns=dataset['train'].column_names,
)
# Build inputs and labels for language modeling
block_size = args.max_length
def group_texts(examples):
# Concatenate all texts
concatenated_examples = {k: sum(examples[k], []) for k in examples.keys()}
total_length = len(concatenated_examples['input_ids'])
# We drop the small remainder
total_length = (total_length // block_size) * block_size
# Split by chunks of block_size
result = {
k: [t[i : i + block_size] for i in range(0, total_length, block_size)]
for k, t in concatenated_examples.items()
}
result['labels'] = result['input_ids'].copy()
return result
lm_datasets = tokenized_datasets.map(
group_texts,
batched=True,
num_proc=4,
)
# Create DataLoader
train_dataset = lm_datasets['train']
eval_dataset = lm_datasets['validation'] if 'validation' in lm_datasets else lm_datasets['test']
def data_collator(data):
return {
'input_ids': torch.tensor([f['input_ids'] for f in data], dtype=torch.long),
'labels': torch.tensor([f['labels'] for f in data], dtype=torch.long)
}
train_loader = DataLoader(
train_dataset,
shuffle=True,
batch_size=args.batch_size,
collate_fn=data_collator,
pin_memory=True, # Speeds up transfer to GPU
num_workers=4
)
eval_loader = DataLoader(
eval_dataset,
shuffle=False,
batch_size=args.batch_size,
collate_fn=data_collator,
pin_memory=True,
num_workers=4
)
return train_loader, eval_loader
def save_all_models(transformer_model, representation_network, dynamics_network, prediction_network, action_encoder, save_dir, epoch):
"""
Save all models to the specified directory.
Args:
transformer_model (nn.Module): Transformer model.
representation_network (nn.Module): Representation network.
dynamics_network (nn.Module): Dynamics network.
prediction_network (nn.Module): Prediction network.
action_encoder (nn.Module): Action encoder.
save_dir (str): Directory to save the models.
epoch (int): Current epoch number.
"""
os.makedirs(save_dir, exist_ok=True)
torch.save(transformer_model.state_dict(), os.path.join(save_dir, f'transformer_model_epoch_{epoch}.pt'))
torch.save(representation_network.state_dict(), os.path.join(save_dir, f'representation_network_epoch_{epoch}.pt'))
torch.save(dynamics_network.state_dict(), os.path.join(save_dir, f'dynamics_network_epoch_{epoch}.pt'))
torch.save(prediction_network.state_dict(), os.path.join(save_dir, f'prediction_network_epoch_{epoch}.pt'))
torch.save(action_encoder.state_dict(), os.path.join(save_dir, f'action_encoder_epoch_{epoch}.pt'))
print(f"All models saved for epoch {epoch}.")
class RotaryPositionalEncoding(nn.Module):
def __init__(self, d_model):
super(RotaryPositionalEncoding, self).__init__()
inv_freq = 1.0 / (10000 ** (torch.arange(0, d_model, 2).float() / d_model))
self.register_buffer('inv_freq', inv_freq)
def forward(self, x):
seq_len, batch_size, _ = x.size()
t = torch.arange(seq_len, device=x.device).type_as(self.inv_freq)
sinusoid_inp = torch.einsum("i,j->ij", t, self.inv_freq)
sin = sinusoid_inp.sin().unsqueeze(1) # (seq_len, 1, d_model/2)
cos = sinusoid_inp.cos().unsqueeze(1) # (seq_len, 1, d_model/2)
x1 = x[..., 0::2]
x2 = x[..., 1::2]
# Apply rotation
x_rotated = torch.zeros_like(x)
x_rotated[..., 0::2] = x1 * cos - x2 * sin
x_rotated[..., 1::2] = x1 * sin + x2 * cos
return x_rotated
class MultiHeadAttention(nn.Module):
def __init__(self, d_model, num_heads):
super(MultiHeadAttention, self).__init__()
assert d_model % num_heads == 0, "d_model must be divisible by num_heads"
self.d_k = d_model // num_heads
self.num_heads = num_heads
self.linear_q = nn.Linear(d_model, d_model)
self.linear_k = nn.Linear(d_model, d_model)
self.linear_v = nn.Linear(d_model, d_model)
self.linear_out = nn.Linear(d_model, d_model)
def forward(self, query, key, value, mask=None):
batch_size = query.size(0)
query = self.linear_q(query).view(batch_size, -1, self.num_heads, self.d_k).transpose(1, 2)
key = self.linear_k(key).view(batch_size, -1, self.num_heads, self.d_k).transpose(1, 2)
value = self.linear_v(value).view(batch_size, -1, self.num_heads, self.d_k).transpose(1, 2)
scores = torch.matmul(query, key.transpose(-2, -1)) / math.sqrt(self.d_k)
if mask is not None:
scores = scores.masked_fill(mask == 0, -1e4)
attn = F.softmax(scores, dim=-1)
output = torch.matmul(attn, value)
output = output.transpose(1, 2).contiguous().view(batch_size, -1, self.num_heads * self.d_k)
return self.linear_out(output)
class MoE(nn.Module):
def __init__(self, d_model, num_experts, d_ff, top_k=2, dropout=0.1):
super(MoE, self).__init__()
self.num_experts = num_experts
self.top_k = top_k
self.experts = nn.ModuleList([
nn.Sequential(
nn.Linear(d_model, d_ff),
nn.GELU() if i % 2 == 0 else nn.SiLU(),
nn.Linear(d_ff, d_model)
)
for i in range(num_experts)
])
self.gate = nn.Linear(d_model, num_experts)
self.dropout = nn.Dropout(dropout)
def forward(self, x):
batch_size, seq_len, d_model = x.size()
# Compute gating scores
gate_scores = self.gate(x) # (batch_size, seq_len, num_experts)
top_k_scores, top_k_indices = torch.topk(gate_scores, self.top_k, dim=-1) # (batch_size, seq_len, top_k)
top_k_scores = F.softmax(top_k_scores, dim=-1) # (batch_size, seq_len, top_k)
# Initialize output
output = torch.zeros_like(x)
# Flatten batch and sequence dimensions
x_flat = x.view(-1, d_model) # (batch_size * seq_len, d_model)
output_flat = output.view(-1, d_model)
top_k_indices_flat = top_k_indices.view(-1, self.top_k) # (batch_size * seq_len, top_k)
top_k_scores_flat = top_k_scores.view(-1, self.top_k) # (batch_size * seq_len, top_k)
for k in range(self.top_k):
expert_idx_flat = top_k_indices_flat[:, k] # (batch_size * seq_len)
expert_scores_flat = top_k_scores_flat[:, k] # (batch_size * seq_len)
for e in range(self.num_experts):
mask = (expert_idx_flat == e) # Boolean mask
if mask.any():
x_masked = x_flat[mask] # Select tokens for expert e
expert_output = self.experts[e](x_masked) # Apply expert e
output_flat[mask] += expert_scores_flat[mask].unsqueeze(-1) * expert_output
output = output_flat.view(batch_size, seq_len, d_model)
return self.dropout(output)
class TransformerBlock(nn.Module):
def __init__(self, d_model, num_heads, d_ff, num_experts, dropout=0.1, top_k=2):
super(TransformerBlock, self).__init__()
self.self_attention = MultiHeadAttention(d_model, num_heads)
self.norm1 = nn.LayerNorm(d_model)
self.cross_attention = MultiHeadAttention(d_model, num_heads)
self.norm2 = nn.LayerNorm(d_model)
self.moe = MoE(d_model, num_experts, d_ff, top_k, dropout)
self.norm3 = nn.LayerNorm(d_model)
def forward(self, x, mask=None, enc_output=None, enc_mask=None):
# Self-attention
attn_output = self.self_attention(x, x, x, mask)
x = self.norm1(x + attn_output)
# Cross-attention (only in decoder)
if enc_output is not None:
cross_attn_output = self.cross_attention(x, enc_output, enc_output, enc_mask)
x = self.norm2(x + cross_attn_output)
# Feedforward/MoE
moe_output = self.moe(x)
return self.norm3(x + moe_output)
class Transformer(nn.Module):
def __init__(self, input_dim, d_model, num_heads, num_layers, d_ff, num_experts, output_dim, dropout=0.1, top_k=2):
super(Transformer, self).__init__()
self.embedding = nn.Embedding(input_dim, d_model, padding_idx=input_dim - 1)
self.rotary_positional_encoding = RotaryPositionalEncoding(d_model)
self.encoder_layers = nn.ModuleList(
[TransformerBlock(d_model, num_heads, d_ff, num_experts, dropout, top_k) for _ in range(num_layers)]
)
self.decoder_layers = nn.ModuleList(
[TransformerBlock(d_model, num_heads, d_ff, num_experts, dropout, top_k) for _ in range(num_layers)]
)
self.output_layer = nn.Linear(d_model, output_dim)
self.d_model = d_model
def forward(self, src, tgt, src_mask=None, tgt_mask=None):
# Encoder
src = self.embedding(src) * math.sqrt(self.d_model)
src = src.transpose(0, 1) # (batch_size, seq_len, d_model) -> (seq_len, batch_size, d_model)
src = self.rotary_positional_encoding(src)
src = src.transpose(0, 1) # (seq_len, batch_size, d_model) -> (batch_size, seq_len, d_model)
for layer in self.encoder_layers:
src = layer(src, src_mask)
# Decoder
tgt = self.embedding(tgt) * math.sqrt(self.d_model)
tgt = tgt.transpose(0, 1)
tgt = self.rotary_positional_encoding(tgt)
tgt = tgt.transpose(0, 1)
for layer in self.decoder_layers:
tgt = layer(tgt, tgt_mask, src, src_mask)
output = self.output_layer(tgt)
return output
def generate_with_beam_search(self, src, tokenizer, beam_size=5, max_length=20, n_tokens_predict=3, temperature=1.0):
"""
Generate sequences using beam search with multi-token prediction.
Args:
src (torch.Tensor): Source input tensor of shape (batch_size, seq_len)
tokenizer: Tokenizer to access special tokens
beam_size (int): Size of the beam for beam search
max_length (int): Maximum length of the generated sequence
n_tokens_predict (int): Number of tokens to predict at each step
temperature (float): Temperature parameter for softmax
Returns:
List[Tuple[torch.Tensor, float]]: List of (sequence, score) tuples
"""
batch_size = src.size(0)
device = src.device
vocab_size = self.output_layer.out_features
# Encode the source
src_enc = self.encode(src)
# Initialize beam
beam = [(torch.full((batch_size, 1), tokenizer.bos_token_id, dtype=torch.long, device=device),
0.0, # log probability
torch.zeros(batch_size, device=device), # cumulative entropy
torch.zeros(batch_size, device=device))] # cumulative variance
for _ in range(max_length // n_tokens_predict):
all_candidates = []
for seq, score, cum_entropy, cum_variance in beam:
if seq[:, -1].item() == tokenizer.eos_token_id:
all_candidates.append((seq, score, cum_entropy, cum_variance))
continue
# Predict next n tokens
logits = self.predict_next_n_tokens(src_enc, seq, n_tokens_predict)
# Calculate probabilities, entropy, and variance
probs = F.softmax(logits / temperature, dim=-1)
entropy = -torch.sum(probs * torch.log(probs + 1e-9), dim=-1)
variance = torch.var(probs, dim=-1)
# Sample top-k tokens for each position
topk_probs, topk_indices = torch.topk(probs, k=beam_size, dim=-1)
# Generate all possible continuations
for i in range(beam_size ** n_tokens_predict):
indices = [i // (beam_size ** j) % beam_size for j in range(n_tokens_predict)]
new_tokens = topk_indices[:, range(n_tokens_predict), indices]
new_seq = torch.cat([seq, new_tokens], dim=-1)
new_score = score + torch.sum(torch.log(topk_probs[:, range(n_tokens_predict), indices]))
new_entropy = cum_entropy + torch.sum(entropy[:, indices])
new_variance = cum_variance + torch.sum(variance[:, indices])
all_candidates.append((new_seq, new_score, new_entropy, new_variance))
# Select top beam_size candidates
beam = sorted(all_candidates, key=lambda x: x[1] - 0.1 * x[2] + 0.05 * x[3], reverse=True)[:beam_size]
# Stop if all beams have ended
if all(seq[:, -1].item() == tokenizer.eos_token_id for seq, _, _, _ in beam):
break
return [(seq, score) for seq, score, _, _ in beam]
def encode(self, src):
src_emb = self.embedding(src) * math.sqrt(self.d_model)
src_emb = src_emb.transpose(0, 1)
src_emb = self.rotary_positional_encoding(src_emb)
src_emb = src_emb.transpose(0, 1)
src_enc = src_emb
for layer in self.encoder_layers:
src_enc = layer(src_enc)
return src_enc
def predict_next_n_tokens(self, src_enc, tgt_seq, n_tokens):
tgt_emb = self.embedding(tgt_seq) * math.sqrt(self.d_model)
tgt_emb = tgt_emb.transpose(0, 1)
tgt_emb = self.rotary_positional_encoding(tgt_emb)
tgt_emb = tgt_emb.transpose(0, 1)
tgt_dec = tgt_emb
for layer in self.decoder_layers:
tgt_dec = layer(tgt_dec, None, src_enc, None)
output = self.output_layer(tgt_dec[:, -1:])
return output.repeat(1, n_tokens, 1)
# Objective Functions
class InfoNCE_Loss(nn.Module):
def __init__(self, temperature=0.07):
super(InfoNCE_Loss, self).__init__()
self.temperature = temperature
self.cross_entropy = nn.CrossEntropyLoss()
def forward(self, z_i, z_j):
"""
Args:
z_i (torch.Tensor): Flattened representations from view i, shape (2n, embed_dim)
z_j (torch.Tensor): Flattened representations from view j, shape (2n, embed_dim)
Returns:
torch.Tensor: InfoNCE loss
"""
n = z_i.size(0)
z = torch.cat([z_i, z_j], dim=0) # Shape: (2n, embed_dim)
z = F.normalize(z, dim=1)
similarity_matrix = torch.matmul(z, z.T) # Shape: (2n, 2n)
# Create a mask to exclude self-similarity
mask = torch.eye(2 * n, device=z.device, dtype=torch.bool)
similarity_matrix = similarity_matrix.masked_fill(mask, -1e4) # Use a manageable negative value
# Create labels for contrastive learning
labels = torch.arange(n, device=z.device)
labels = torch.cat([labels + n, labels], dim=0) # Shape: (2n,)
# Apply temperature scaling
similarity_matrix /= self.temperature
# Compute cross-entropy loss
loss = self.cross_entropy(similarity_matrix, labels)
return loss
class CovarianceRegularization(nn.Module):
def __init__(self, lambda_reg=1e-3):
super(CovarianceRegularization, self).__init__()
self.lambda_reg = lambda_reg
def forward(self, embeddings):
"""
Args:
embeddings (torch.Tensor): Embedding tensor, shape (batch_size, embed_dim)
Returns:
torch.Tensor: Covariance regularization loss
"""
batch_size, embed_dim = embeddings.size()
mean = embeddings.mean(dim=0)
embeddings_centered = embeddings - mean
cov = (embeddings_centered.T @ embeddings_centered) / (batch_size - 1)
cov_loss = torch.sum(cov ** 2) - torch.sum(torch.diag(cov) ** 2)
return self.lambda_reg * cov_loss
class DynamicsPerformanceLoss(nn.Module):
def __init__(self, lambda_var=1e-3):
super(DynamicsPerformanceLoss, self).__init__()
self.lambda_var = lambda_var
def forward(self, true_next_state, predicted_next_state):
"""
Args:
true_next_state (torch.Tensor): Ground truth next state, shape (batch_size, state_dim)
predicted_next_state (torch.Tensor): Predicted next state, shape (batch_size, state_dim)
Returns:
torch.Tensor: Dynamics performance loss
"""
mse_loss = F.mse_loss(predicted_next_state, true_next_state)
variance_loss = torch.var(predicted_next_state, dim=0).mean()
return mse_loss + self.lambda_var * variance_loss
class ThoughtConsistencyLoss(nn.Module):
def __init__(self):
super(ThoughtConsistencyLoss, self).__init__()
def forward(self, true_next_state, perturbed_next_state):
"""
Args:
true_next_state (torch.Tensor): Ground truth next state, shape (batch_size, state_dim)
perturbed_next_state (torch.Tensor): Perturbed next state, shape (batch_size, state_dim)
Returns:
torch.Tensor: Thought-consistency loss
"""
return F.mse_loss(true_next_state, perturbed_next_state)
class PolicyValueJointLoss(nn.Module):
def __init__(self, lambda_value=0.5):
super(PolicyValueJointLoss, self).__init__()
self.lambda_value = lambda_value
self.cross_entropy = nn.CrossEntropyLoss()
self.mse_loss = nn.MSELoss()
def forward(self, policy_logits, true_policy, value_pred, true_value):
"""
Args:
policy_logits (torch.Tensor): Logits from the policy network, shape (batch_size * seq_len, num_actions)
true_policy (torch.Tensor): Ground truth policy, shape (batch_size * seq_len, num_actions)
value_pred (torch.Tensor): Predicted values, shape (batch_size * seq_len)
true_value (torch.Tensor): Ground truth values, shape (batch_size * seq_len)
Returns:
torch.Tensor: Combined policy and value loss
"""
policy_logits = policy_logits.reshape(-1, policy_logits.size(-1))
true_policy = true_policy.reshape(-1, true_policy.size(-1))
value_pred = value_pred.reshape(-1)
true_value = true_value.reshape(-1)
policy_loss = self.cross_entropy(policy_logits, true_policy.argmax(dim=1))
value_loss = self.mse_loss(value_pred, true_value)
return policy_loss + self.lambda_value * value_loss
class ActionDiversityReward(nn.Module):
def __init__(self, lambda_div=1e-3):
super(ActionDiversityReward, self).__init__()
self.lambda_div = lambda_div
def forward(self, action_embeddings):
"""
Args:
action_embeddings (torch.Tensor): Embeddings of actions, shape (batch_size, embed_dim)
Returns:
torch.Tensor: Action diversity loss
"""
similarity_matrix = F.cosine_similarity(action_embeddings.unsqueeze(1), action_embeddings.unsqueeze(0), dim=2)
# Zero out self-similarity
similarity_matrix = similarity_matrix - torch.eye(similarity_matrix.size(0)).to(action_embeddings.device)
diversity_loss = torch.sum(similarity_matrix ** 2)
return self.lambda_div * diversity_loss
class ExpectedThoughtValueLoss(nn.Module):
def __init__(self):
super(ExpectedThoughtValueLoss, self).__init__()
def forward(self, mcts_best_values):
"""
Args:
mcts_best_values (torch.Tensor): Best values from MCTS, shape (batch_size)
Returns:
torch.Tensor: ETV loss
"""
return -mcts_best_values.mean()
class ExplorationRegularization(nn.Module):
def __init__(self, lambda_expl=1e-3):
super(ExplorationRegularization, self).__init__()
self.lambda_expl = lambda_expl
def forward(self, visit_counts):
"""
Args:
visit_counts (torch.Tensor): Visit counts for actions, shape (batch_size, num_actions)
Returns:
torch.Tensor: Exploration regularization loss
"""
reward = torch.sum(1.0 / (visit_counts + 1), dim=-1)
return self.lambda_expl * reward.mean()
class KL_DivergenceLoss(nn.Module):
def __init__(self):
super(KL_DivergenceLoss, self).__init__()
def forward(self, old_policy, new_policy):
"""
Args:
old_policy (torch.Tensor): Old policy probabilities, shape (batch_size, num_actions)
new_policy (torch.Tensor): New policy probabilities, shape (batch_size, num_actions)
Returns:
torch.Tensor: KL divergence loss
"""
kl_div = F.kl_div(new_policy.log(), old_policy, reduction='batchmean')
return kl_div
# MuZero Components
class ActionEncoder(nn.Module):
def __init__(self, action_vocab_size, embed_dim):
super(ActionEncoder, self).__init__()
self.embedding = nn.Embedding(action_vocab_size, embed_dim)
def forward(self, action_indices):
"""
Args:
action_indices (torch.Tensor): Tensor of shape (batch_size, seq_len)
Returns:
torch.Tensor: Encoded actions of shape (batch_size, seq_len, embed_dim)
"""
return self.embedding(action_indices)
class RepresentationNetwork(nn.Module):
def __init__(self, vocab_dim, d_model, state_dim):
super(RepresentationNetwork, self).__init__()
self.proj = nn.Linear(vocab_dim, d_model) # Project from vocab_dim to d_model
self.linear = nn.Linear(d_model, state_dim) # Project from d_model to state_dim
self.norm = nn.LayerNorm(state_dim)
def forward(self, transformer_output):
"""
Args:
transformer_output (torch.Tensor): Shape (batch_size, seq_len, vocab_dim)
Returns:
torch.Tensor: Encoded state of shape (batch_size, seq_len, state_dim)
"""
# First project down from vocab_dim to d_model
projected_output = self.proj(transformer_output) # Shape: (batch_size, seq_len, d_model)
# Then project down from d_model to state_dim
state = self.linear(projected_output) # Shape: (batch_size, seq_len, state_dim)
state = self.norm(state) # Shape: (batch_size, seq_len, state_dim)
return state
class DynamicsNetwork(nn.Module):
def __init__(self, state_dim, action_dim, hidden_dim):
super(DynamicsNetwork, self).__init__()
self.rms_norm = nn.LayerNorm(state_dim)
self.fc1 = nn.Linear(state_dim + action_dim, hidden_dim)
self.activation = nn.GELU()
self.fc2 = nn.Linear(hidden_dim, state_dim)
def forward(self, state, action):
"""
Args:
state (torch.Tensor): Current state, shape (batch_size, seq_len, state_dim)
action (torch.Tensor): Action embedding, shape (batch_size, seq_len, action_dim)
Returns:
torch.Tensor: Predicted next state, shape (batch_size, seq_len, state_dim)
"""
norm_state = self.rms_norm(state)
combined = torch.cat([norm_state, action], dim=-1)
hidden = self.activation(self.fc1(combined))
next_state = self.fc2(hidden)
return next_state
class PredictionNetwork(nn.Module):
def __init__(self, state_dim, action_vocab_size, value_dim):
super(PredictionNetwork, self).__init__()
self.state_dim = state_dim
self.rms_norm = nn.LayerNorm(state_dim)
self.policy_head = nn.Linear(state_dim, action_vocab_size) # Output size is action_vocab_size
self.value_head = nn.Linear(state_dim, value_dim)
def forward(self, state):
"""
Args:
state (torch.Tensor): State representation, shape (batch_size, state_dim)
Returns:
Tuple[torch.Tensor, torch.Tensor]: Policy logits and value estimates
"""
norm_state = self.rms_norm(state)
policy_logits = self.policy_head(norm_state) # Shape: (batch_size, action_vocab_size)
value_estimates = self.value_head(norm_state).squeeze(-1) # Shape: (batch_size)
return policy_logits, value_estimates
class MCTSNode:
__slots__ = [
'state',
'parent',
'action',
'children',
'visit_count',
'value_sum',
'prior',
'cached_policy',
'cached_value',
'thought_node',
'entropy',
'variance'
]
def __init__(self, state, thought_node, parent=None, action=None):
self.state = state
self.thought_node = thought_node
self.parent = parent
self.action = action
self.children = {}
self.visit_count = 0
self.value_sum = 0.0
self.prior = 0.0
self.cached_policy = None
self.cached_value = None
self.entropy = 0.0
self.variance = 0.0
def expand(self, priors):
for child_thought_node in self.thought_node.children:
action = child_thought_node.name
if action not in self.children:
child_state = self.state.apply_action(action)
child_node = MCTSNode(
state=child_state,
thought_node=child_thought_node,
parent=self,
action=action
)
child_node.prior = priors.get(action, 1.0 / len(self.thought_node.children))
self.children[action] = child_node
def is_leaf(self):
return len(self.children) == 0
def ucb_score(self, total_visits, exploration_constant=math.sqrt(2)):
if self.visit_count == 0:
return float('inf') # Ensure unvisited nodes are selected first
avg_value = self.value_sum / self.visit_count
exploration_term = exploration_constant * self.prior * math.sqrt(total_visits) / (1 + self.visit_count)
entropy_term = -0.1 * self.entropy # Slightly prefer lower entropy
variance_term = 0.05 * self.variance # Slightly prefer higher variance
return avg_value + exploration_term + entropy_term + variance_term
class MCTS:
def __init__(self, prediction_network, dynamics_network, action_encoder, num_iterations=10, exploration_constant=math.sqrt(2), beam_size=5, n_tokens_predict=3):
self.prediction_network = prediction_network
self.dynamics_network = dynamics_network
self.action_encoder = action_encoder
self.num_iterations = num_iterations
self.exploration_constant = exploration_constant
self.beam_size = beam_size
self.n_tokens_predict = n_tokens_predict
self.cache = {}
def search_with_beam(self, root_state):
root_node = MCTSNode(state=root_state, thought_node=root_state.thought_node)
# Evaluate the root node and backpropagate
value_estimate = self.evaluate(root_node) # Evaluate and expand root_node
self.backpropagate(root_node, value_estimate) # Backpropagate the value
beam = [(root_node, 0.0, 0.0, 0.0, [])] # (node, score, cum_entropy, cum_variance, action_sequence)
for iteration in range(self.num_iterations):
all_candidates = []
for node, score, cum_entropy, cum_variance, action_sequence in beam:
if node.is_leaf():
value_estimate = self.evaluate(node)
self.backpropagate(node, value_estimate) # Backpropagate after evaluation
if len(node.children) == 0:
continue # No children to expand
total_visits = sum(child.visit_count for child in node.children.values())
# Select top actions based on UCB score
sorted_children = sorted(
node.children.items(),
key=lambda item: item[1].ucb_score(total_visits, self.exploration_constant),
reverse=True
)[:self.beam_size]
for selected_action, selected_node in sorted_children:
current_node = selected_node
current_sequence = action_sequence + [selected_action]
current_score = score
current_entropy = cum_entropy + selected_node.entropy
current_variance = cum_variance + selected_node.variance
# Predict n_tokens_predict actions
for _ in range(self.n_tokens_predict):
if current_node.is_leaf():
value_estimate = self.evaluate(current_node)
self.backpropagate(current_node, value_estimate) # Backpropagate after evaluation
if len(current_node.children) == 0:
break # No more actions
total_visits = sum(child.visit_count for child in current_node.children.values())
next_action, next_node = max(
current_node.children.items(),
key=lambda item: item[1].ucb_score(total_visits, self.exploration_constant)
)
current_sequence.append(next_action)
# Prevent division by zero by ensuring visit_count > 0
if next_node.visit_count > 0:
current_score += next_node.value_sum / next_node.visit_count
else:
# Assign a default value or handle the zero division case
current_score += 0.0 # Alternatively, use a small epsilon or skip
current_entropy += next_node.entropy
current_variance += next_node.variance
current_node = next_node
all_candidates.append((current_node, current_score, current_entropy, current_variance, current_sequence))
if not all_candidates:
break # No more candidates to expand
# Select top beam_size candidates
beam = sorted(all_candidates, key=lambda x: x[1] - 0.1 * x[2] + 0.05 * x[3], reverse=True)[:self.beam_size]
print(f"Iteration {iteration + 1}: Beam size after sorting: {len(beam)}") # Debug
if beam:
best_sequence = beam[0][4]
return best_sequence
else:
return []
def search(self, root_state):
root_node = MCTSNode(state=root_state, thought_node=root_state.thought_node)
for _ in range(self.num_iterations):
node = self.select(root_node)
value = self.evaluate(node)
self.backpropagate(node, value)
return self.best_action_sequence(root_node)
def select(self, node):
while not node.is_leaf():
total_visits = sum(child.visit_count for child in node.children.values())
_, node = max(
node.children.items(),
key=lambda item: item[1].ucb_score(total_visits, self.exploration_constant)
)
return node
def evaluate(self, node):
# Extract the last time step
state_representation = node.state.representation[:, -1, :] # Shape: (batch_size=1, state_dim)
print(f"Evaluating node with state_representation shape: {state_representation.shape}") # Debug
policy_logits, value_estimate = self.prediction_network(state_representation)
print(f"Policy logits shape: {policy_logits.shape}, Value estimate shape: {value_estimate.shape}") # Debug
value_estimate = value_estimate.item() # Now safe as batch_size=1
policy_probs = F.softmax(policy_logits, dim=-1).squeeze(0) # Shape: (action_vocab_size,)
print(f"Policy probabilities shape: {policy_probs.shape}") # Debug
priors = {}
for child in node.thought_node.children:
action_name = child.name
action_idx = action_to_index.get(action_name, None)
if action_idx is not None and action_idx < policy_probs.size(0):
priors[action_name] = policy_probs[action_idx].item()
else:
priors[action_name] = 1.0 / len(node.thought_node.children)
node.expand(priors)
# Calculate entropy and variance
entropy = -torch.sum(policy_probs * torch.log(policy_probs + 1e-9))
variance = torch.var(policy_probs)
node.entropy = entropy.item()
node.variance = variance.item()
print(f"Node entropy: {node.entropy}, variance: {node.variance}") # Debug
return value_estimate # Return the value estimate for backpropagation
def backpropagate(self, node, value):
while node is not None:
node.visit_count += 1
node.value_sum += value
node = node.parent
def best_action_sequence(self, root_node):
sequences = []
self._generate_sequences(root_node, [], sequences)
# Score sequences based on visit counts, entropy, and variance
scored_sequences = []
for seq in sequences:
score = sum(node.visit_count for node in seq)
entropy = sum(node.entropy for node in seq)
variance = sum(node.variance for node in seq)
adjusted_score = score - 0.1 * entropy + 0.05 * variance
scored_sequences.append((seq, adjusted_score))
# Sort sequences by adjusted score and select top beam_size
best_sequences = sorted(scored_sequences, key=lambda x: x[1], reverse=True)[:self.beam_size]
# Return the actions of the best sequence
best_sequence = best_sequences[0][0]
return [node.action for node in best_sequence[1:self.n_tokens_predict+1]] # Exclude root node
def _generate_sequences(self, node, current_sequence, sequences):
current_sequence.append(node)
if len(current_sequence) > self.n_tokens_predict or not node.children:
sequences.append(current_sequence)
else:
for child in node.children.values():
self._generate_sequences(child, current_sequence.copy(), sequences)
class State:
def __init__(self, representation, dynamics_network, action_encoder, thought_node):
self.representation = representation
self.dynamics_network = dynamics_network
self.action_encoder = action_encoder
self.thought_node = thought_node
def apply_action(self, action):
next_thought_node = None
for child in self.thought_node.children:
if child.name == action:
next_thought_node = child
break
if next_thought_node is None:
raise ValueError(f"Action '{action}' is not valid from the current thought node.")
# Adjust action_index and action_embedding shapes
action_index = torch.tensor([action_to_index[action]], device=self.representation.device)
action_embedding = self.action_encoder(action_index) # Shape: (batch_size=1, action_dim)
# Extract the last time step of the state
state = self.representation[:, -1, :] # Shape: (batch_size, state_dim)
# Ensure action_embedding matches the state dimension
next_state_representation = self.dynamics_network(state, action_embedding) # Shape: (batch_size, state_dim)
# Append the new state to the representation history
new_representation = torch.cat([self.representation, next_state_representation.unsqueeze(1)], dim=1) # Shape: (batch_size, seq_len+1, state_dim)
return State(
representation=new_representation,
dynamics_network=self.dynamics_network,
action_encoder=self.action_encoder,
thought_node=next_thought_node
)
class PPOAgent:
def __init__(self, policy_network, optimizer, clip_epsilon=0.2, entropy_coef=0.01, value_coef=0.5):
self.policy_network = policy_network
self.optimizer = optimizer
self.clip_epsilon = clip_epsilon
self.entropy_coef = entropy_coef
self.value_coef = value_coef
def compute_loss(self, states, old_log_probs, actions, returns, advantages):
# Get policy logits and value estimates
policy_logits, value_estimates = self.policy_network(states)
# Flatten all tensors
policy_logits = policy_logits.reshape(-1, policy_logits.size(-1))
value_estimates = value_estimates.reshape(-1)
actions = actions.reshape(-1)
old_log_probs = old_log_probs.reshape(-1)
returns = returns.reshape(-1)
advantages = advantages.reshape(-1)
# Ensure all tensors have the same first dimension
assert policy_logits.size(0) == value_estimates.size(0) == actions.size(0) == old_log_probs.size(0) == returns.size(0) == advantages.size(0), "Tensor sizes mismatch"
# Compute new log probabilities
new_log_probs_all = F.log_softmax(policy_logits, dim=-1)
new_log_probs = new_log_probs_all.gather(1, actions.unsqueeze(-1)).squeeze(-1)
# Compute ratios
ratios = torch.exp(new_log_probs - old_log_probs)
# PPO surrogate loss
surr1 = ratios * advantages
surr2 = torch.clamp(ratios, 1 - self.clip_epsilon, 1 + self.clip_epsilon) * advantages
policy_loss = -torch.min(surr1, surr2).mean()
# Value loss
value_loss = F.mse_loss(value_estimates, returns)
# Entropy loss
entropy = -(new_log_probs * torch.exp(new_log_probs)).mean()
# Total loss
total_loss = policy_loss + self.value_coef * value_loss - self.entropy_coef * entropy
return total_loss
# Tree of Thought Components
class ThoughtNode:
def __init__(self, name):
self.name = name
self.children = []
self.parent = None
def add_child(self, child_node):
child_node.parent = self
self.children.append(child_node)
# Function to build the Tree of Thought from your detailed structure
def build_tree_of_thought():
# Create the root node
root = ThoughtNode('Problem-Solving Process')
# Level 1 nodes
problem_identification = ThoughtNode('Problem Identification')
problem_analysis = ThoughtNode('Problem Analysis')
solution_generation = ThoughtNode('Solution Generation')
implementation = ThoughtNode('Implementation')
evaluation_adjustment = ThoughtNode('Evaluation and Adjustment')
root.add_child(problem_identification)
root.add_child(problem_analysis)
root.add_child(solution_generation)
root.add_child(implementation)
root.add_child(evaluation_adjustment)
# Problem Identification children
B1 = ThoughtNode('Define the Problem')
B2 = ThoughtNode('Identify Stakeholders')
B3 = ThoughtNode('Determine Constraints')
B4 = ThoughtNode('Recognize Problem Type')
B5 = ThoughtNode('Historical Context')
problem_identification.add_child(B1)
problem_identification.add_child(B2)
problem_identification.add_child(B3)
problem_identification.add_child(B4)
problem_identification.add_child(B5)
# Define the Problem children
B1a = ThoughtNode('Problem Statement Formulation')
B1b = ThoughtNode('Scope Definition')
B1c = ThoughtNode('Objective Setting')
B1.add_child(B1a)
B1.add_child(B1b)
B1.add_child(B1c)
# Identify Stakeholders children
B2a = ThoughtNode('Stakeholder Mapping')
B2b = ThoughtNode('Interest and Influence Analysis')
B2c = ThoughtNode('Engagement Strategy')
B2.add_child(B2a)
B2.add_child(B2b)
B2.add_child(B2c)
# Determine Constraints children
B3a = ThoughtNode('Resource Limitations')
B3b = ThoughtNode('Time Constraints')
B3c = ThoughtNode('Legal and Regulatory Constraints')
B3.add_child(B3a)
B3.add_child(B3b)
B3.add_child(B3c)
# Recognize Problem Type children
B4a = ThoughtNode('Simple vs Complex')
B4b = ThoughtNode('Known vs Unknown')
B4c = ThoughtNode('Tame vs Wicked Problems')
B4.add_child(B4a)
B4.add_child(B4b)
B4.add_child(B4c)
# Historical Context children
B5a = ThoughtNode('Previous Attempts')
B5b = ThoughtNode('Lessons Learned')
B5c = ThoughtNode('Environmental Factors')
B5.add_child(B5a)
B5.add_child(B5b)
B5.add_child(B5c)
# Problem Analysis children
C1 = ThoughtNode('Root Cause Analysis')
C2 = ThoughtNode('System Mapping')
C3 = ThoughtNode('Data Collection')
C4 = ThoughtNode('Impact Assessment')
C5 = ThoughtNode('Theoretical Framework')
problem_analysis.add_child(C1)
problem_analysis.add_child(C2)
problem_analysis.add_child(C3)
problem_analysis.add_child(C4)
problem_analysis.add_child(C5)
# Root Cause Analysis children
C1a = ThoughtNode('5 Whys Technique')
C1b = ThoughtNode('Fishbone Diagram')
C1c = ThoughtNode('Pareto Analysis')
C1.add_child(C1a)
C1.add_child(C1b)
C1.add_child(C1c)
# System Mapping children
C2a = ThoughtNode('Causal Loop Diagrams')
C2b = ThoughtNode('Stock and Flow Models')
C2c = ThoughtNode('Network Analysis')
C2.add_child(C2a)
C2.add_child(C2b)
C2.add_child(C2c)
# Data Collection children
C3a = ThoughtNode('Quantitative Data')
C3b = ThoughtNode('Qualitative Data')
C3c = ThoughtNode('Data Validation')
C3.add_child(C3a)
C3.add_child(C3b)
C3.add_child(C3c)
# Quantitative Data children
C3a1 = ThoughtNode('Surveys and Questionnaires')
C3a2 = ThoughtNode('Experimental Data')
C3a3 = ThoughtNode('Big Data Analytics')
C3a.add_child(C3a1)
C3a.add_child(C3a2)
C3a.add_child(C3a3)
# Qualitative Data children
C3b1 = ThoughtNode('Interviews')
C3b2 = ThoughtNode('Focus Groups')
C3b3 = ThoughtNode('Observational Studies')
C3b.add_child(C3b1)
C3b.add_child(C3b2)
C3b.add_child(C3b3)
# Data Validation children
C3c1 = ThoughtNode('Statistical Validation')
C3c2 = ThoughtNode('Cross-Validation')
C3c3 = ThoughtNode('Expert Review')
C3c.add_child(C3c1)
C3c.add_child(C3c2)
C3c.add_child(C3c3)
# Impact Assessment children
C4a = ThoughtNode('Environmental Impact')
C4b = ThoughtNode('Social Impact')
C4c = ThoughtNode('Economic Impact')
C4.add_child(C4a)
C4.add_child(C4b)
C4.add_child(C4c)
# Theoretical Framework children
C5a = ThoughtNode('Literature Review')
C5b = ThoughtNode('Conceptual Modeling')
C5c = ThoughtNode('Hypothesis Formation')
C5.add_child(C5a)
C5.add_child(C5b)
C5.add_child(C5c)
# Solution Generation children
D1 = ThoughtNode('Creative Problem Solving')
D2 = ThoughtNode('Analytical Approach')
D3 = ThoughtNode('Mathematical Computation')
D4 = ThoughtNode('Decision Making')
solution_generation.add_child(D1)
solution_generation.add_child(D2)
solution_generation.add_child(D3)
solution_generation.add_child(D4)
# Action Planning, Resource Allocation, Change Management children (implementation phase)
E1 = ThoughtNode('Action Planning')
E2 = ThoughtNode('Resource Allocation')
E3 = ThoughtNode('Change Management')
implementation.add_child(E1)
implementation.add_child(E2)
implementation.add_child(E3)
# Verification, Performance Metrics, Feedback Loops, Continuous Improvement children (evaluation phase)
F1 = ThoughtNode('Verification')
F2 = ThoughtNode('Performance Metrics')
F3 = ThoughtNode('Feedback Loops')
F4 = ThoughtNode('Continuous Improvement')
evaluation_adjustment.add_child(F1)
evaluation_adjustment.add_child(F2)
evaluation_adjustment.add_child(F3)
evaluation_adjustment.add_child(F4)
# Cross-Cutting Considerations children
G = ThoughtNode('Cross-Cutting Considerations')
root.add_child(G)
# Cross-Cutting Considerations children
G1 = ThoughtNode('Ethical Framework')
G2 = ThoughtNode('Stakeholder Management')
G3 = ThoughtNode('Interdisciplinary Connections')
G4 = ThoughtNode('Technological Integration')
G5 = ThoughtNode('Emotional Intelligence')
G6 = ThoughtNode('Collaborative Problem Solving')
G7 = ThoughtNode('Computational Considerations') # Assuming H was intended as G7
G8 = ThoughtNode('Order of Operations') # Assuming I was intended as G8
G9 = ThoughtNode('Critical Thinking') # Assuming J was intended as G9
G10 = ThoughtNode('Future Perspective') # Assuming K was intended as G10
G11 = ThoughtNode('Learning and Adaptation') # Assuming L was intended as G11
G.add_child(G1)
G.add_child(G2)
G.add_child(G3)
G.add_child(G4)
G.add_child(G5)
G.add_child(G6)
G.add_child(G7)
G.add_child(G8)
G.add_child(G9)
G.add_child(G10)
G.add_child(G11)
# Ethical Framework children
G1a = ThoughtNode('Value-based Decision Making')
G1b = ThoughtNode('Long-term Consequences')
G1.add_child(G1a)
G1.add_child(G1b)
# Value-based Decision Making children
G1a1 = ThoughtNode('Ethical Theories Application')
G1a2 = ThoughtNode('Moral Dilemma Resolution')
G1a.add_child(G1a1)
G1a.add_child(G1a2)
# Long-term Consequences children
G1b1 = ThoughtNode('Sustainability Assessment')
G1b2 = ThoughtNode('Intergenerational Impact')
G1b.add_child(G1b1)
G1b.add_child(G1b2)
# Stakeholder Management children
G2a = ThoughtNode('Direct Stakeholders')
G2b = ThoughtNode('Indirect Stakeholders')
G2c = ThoughtNode('Conflicting Interests')
G2.add_child(G2a)
G2.add_child(G2b)
G2.add_child(G2c)
# Conflicting Interests children
G2c1 = ThoughtNode('Negotiation Strategies')
G2c2 = ThoughtNode('Conflict Resolution Techniques')
G2c.add_child(G2c1)
G2c.add_child(G2c2)
# Interdisciplinary Connections children
G3a = ThoughtNode('Related Fields')
G3b = ThoughtNode('Cross-disciplinary Impact')
G3.add_child(G3a)
G3.add_child(G3b)
# Related Fields children
G3a1 = ThoughtNode('Cross-domain Knowledge Transfer')
G3a2 = ThoughtNode('Interdisciplinary Collaboration')
G3a.add_child(G3a1)
G3a.add_child(G3a2)
# Cross-disciplinary Impact children
G3b1 = ThoughtNode('Synergy Identification')
G3b2 = ThoughtNode('Holistic Impact Assessment')
G3b.add_child(G3b1)
G3b.add_child(G3b2)
# Technological Integration children
G4a = ThoughtNode('AI-assisted Problem Solving')
G4b = ThoughtNode('Data-driven Insights')
G4c = ThoughtNode('Digital Collaboration Tools')
G4.add_child(G4a)
G4.add_child(G4b)
G4.add_child(G4c)
# AI-assisted Problem Solving children
G4a1 = ThoughtNode('Machine Learning Models')
G4a2 = ThoughtNode('Natural Language Processing')
G4a.add_child(G4a1)
G4a.add_child(G4a2)
# Data-driven Insights children
G4b1 = ThoughtNode('Big Data Analytics')
G4b2 = ThoughtNode('Predictive Modeling')
G4b.add_child(G4b1)
G4b.add_child(G4b2)
# Digital Collaboration Tools children
G4c1 = ThoughtNode('Project Management Platforms')
G4c2 = ThoughtNode('Virtual Reality Collaboration')
G4c.add_child(G4c1)
G4c.add_child(G4c2)
# Emotional Intelligence children
G5a = ThoughtNode('Self-Awareness')
G5b = ThoughtNode('Empathy')
G5c = ThoughtNode('Stress Management')
G5.add_child(G5a)
G5.add_child(G5b)
G5.add_child(G5c)
# Self-Awareness children
G5a1 = ThoughtNode('Emotional Recognition')
G5a2 = ThoughtNode('Personal Bias Identification')
G5a.add_child(G5a1)
G5a.add_child(G5a2)
# Empathy children
G5b1 = ThoughtNode('Perspective Taking')
G5b2 = ThoughtNode('Active Listening')
G5b.add_child(G5b1)
G5b.add_child(G5b2)
# Stress Management children
G5c1 = ThoughtNode('Mindfulness Techniques')
G5c2 = ThoughtNode('Resilience Building')
G5c.add_child(G5c1)
G5c.add_child(G5c2)
# Collaborative Problem Solving children
G6a = ThoughtNode('Team Dynamics')
G6b = ThoughtNode('Communication Strategies')
G6c = ThoughtNode('Conflict Resolution')
G6.add_child(G6a)
G6.add_child(G6b)
G6.add_child(G6c)
# Team Dynamics children
G6a1 = ThoughtNode('Team Formation Strategies')
G6a2 = ThoughtNode('Role Assignment')
G6a.add_child(G6a1)
G6a.add_child(G6a2)
# Communication Strategies children
G6b1 = ThoughtNode('Clear Messaging')
G6b2 = ThoughtNode('Feedback Mechanisms')
G6b.add_child(G6b1)
G6b.add_child(G6b2)
# Conflict Resolution children
G6c1 = ThoughtNode('Mediation Techniques')
G6c2 = ThoughtNode('Consensus Building')
G6c.add_child(G6c1)
G6c.add_child(G6c2)
# Computational Considerations children
G7a = ThoughtNode('CPU Operations')
G7b = ThoughtNode('GPU Parallelization')
G7c = ThoughtNode('Floating-Point Precision')
G7.add_child(G7a)
G7.add_child(G7b)
G7.add_child(G7c)
# CPU Operations children
G7a1 = ThoughtNode('Instruction Set Architecture')
G7a2 = ThoughtNode('Pipelining and Parallelism')
G7a.add_child(G7a1)
G7a.add_child(G7a2)
# GPU Parallelization children
G7b1 = ThoughtNode('CUDA Programming')
G7b2 = ThoughtNode('OpenCL Framework')
G7b.add_child(G7b1)
G7b.add_child(G7b2)
# Floating-Point Precision children
G7c1 = ThoughtNode('IEEE 754 Standard')
G7c2 = ThoughtNode('Error Propagation Analysis')
G7c.add_child(G7c1)
G7c.add_child(G7c2)
# Order of Operations children
G8a = ThoughtNode('Parentheses')
G8b = ThoughtNode('Exponents')
G8c = ThoughtNode('Multiplication and Division')
G8d = ThoughtNode('Addition and Subtraction')
G8.add_child(G8a)
G8.add_child(G8b)
G8.add_child(G8c)
G8.add_child(G8d)
# Critical Thinking children
G9a = ThoughtNode('Assumptions Questioning')
G9b = ThoughtNode('Bias Recognition')
G9.add_child(G9a)
G9.add_child(G9b)
# Assumptions Questioning children
G9a1 = ThoughtNode('Socratic Questioning')
G9a2 = ThoughtNode('Devil\'s Advocate Approach')
G9a.add_child(G9a1)
G9a.add_child(G9a2)
# Bias Recognition children
G9b1 = ThoughtNode('Cognitive Bias Identification')
G9b2 = ThoughtNode('Debiasing Techniques')
G9b.add_child(G9b1)
G9b.add_child(G9b2)
# Future Perspective children
G10a = ThoughtNode('Short-term Projections')
G10b = ThoughtNode('Long-term Scenarios')
G10c = ThoughtNode('Potential Impacts')
G10.add_child(G10a)
G10.add_child(G10b)
G10.add_child(G10c)
# Short-term Projections children
G10a1 = ThoughtNode('Trend Analysis')
G10a2 = ThoughtNode('Scenario Planning')
G10a.add_child(G10a1)
G10a.add_child(G10a2)
# Long-term Scenarios children
G10b1 = ThoughtNode('Futures Wheel')
G10b2 = ThoughtNode('Backcasting')
G10b.add_child(G10b1)
G10b.add_child(G10b2)
# Potential Impacts children
G10c1 = ThoughtNode('Risk Assessment')
G10c2 = ThoughtNode('Opportunity Identification')
G10c.add_child(G10c1)
G10c.add_child(G10c2)
# Learning and Adaptation children
G11a = ThoughtNode('Reflective Practice')
G11b = ThoughtNode('Knowledge Transfer')
G11c = ThoughtNode('Adaptive Problem Solving')
G11.add_child(G11a)
G11.add_child(G11b)
G11.add_child(G11c)
# Reflective Practice children
G11a1 = ThoughtNode('After Action Review')
G11a2 = ThoughtNode('Learning Journals')
G11a.add_child(G11a1)
G11a.add_child(G11a2)
# Knowledge Transfer children
G11b1 = ThoughtNode('Best Practice Documentation')
G11b2 = ThoughtNode('Mentoring Programs')
G11b.add_child(G11b1)
G11b.add_child(G11b2)
# Adaptive Problem Solving children
G11c1 = ThoughtNode('Iterative Approaches')
G11c2 = ThoughtNode('Flexibility in Methodology')
G11c.add_child(G11c1)
G11c.add_child(G11c2)
return root
def traverse_tree(node, action_list):
if node.name not in action_list:
action_list.append(node.name)
for child in node.children:
traverse_tree(child, action_list)
def infer(query, world_model_components, root_thought_node, tokenizer, max_length=2000, inference_mode='world_model', beam_size=5, n_tokens_predict=3, mcts_iterations=10, exploration_constant=1.414):
"""
Perform inference given a query, utilizing the Tree of Thought and MCTS with multi-token beam search.
Args:
query (str): The input query or prompt.
world_model_components (tuple): Tuple containing the model components.
root_thought_node (ThoughtNode): The root node of the Tree of Thought.
tokenizer (transformers.PreTrainedTokenizer): The tokenizer used.
max_length (int): Maximum length for the generated sequence.
inference_mode (str): Inference mode ('world_model', 'without_world_model', 'world_model_tree_of_thought')
beam_size (int): Size of the beam for beam search
n_tokens_predict (int): Number of tokens to predict at each step
Returns:
List[str] or str: The sequence of actions (thoughts) selected or generated text.
"""
representation_network, dynamics_network, prediction_network, action_encoder, ppo_agent, model_transformer = world_model_components
# Tokenize and encode the query
input_ids = tokenizer.encode(query, return_tensors='pt').to(device)
attention_mask = (input_ids != tokenizer.pad_token_id).long()
if inference_mode == 'without_world_model':
# Directly use the transformer model to generate text with beam search
with torch.no_grad():
generated_sequences = model_transformer.generate_with_beam_search(
src=input_ids,
tokenizer=tokenizer,
beam_size=beam_size,
max_length=max_length,
n_tokens_predict=n_tokens_predict,
temperature=args.temperature
)
best_sequence, best_score = generated_sequences[0]
generated_text = tokenizer.decode(best_sequence[0], skip_special_tokens=True)
return generated_text
else:
# Use the world model components
with torch.no_grad():
transformer_output = model_transformer(input_ids, input_ids)
# Get the initial state representation
initial_representation = representation_network(transformer_output) # Shape: (batch_size=1, seq_len, state_dim)
initial_representation = initial_representation[:, -1, :].unsqueeze(1) # Shape: (batch_size=1, 1, state_dim)
initial_state = State(
representation=initial_representation,
dynamics_network=dynamics_network,
action_encoder=action_encoder,
thought_node=root_thought_node
)
if inference_mode == 'world_model_tree_of_thought':
# Use MCTS with Tree of Thought and multi-token beam search
mcts = MCTS(prediction_network, dynamics_network, action_encoder, num_iterations=mcts_iterations, exploration_constant=exploration_constant)
current_state = initial_state
thought_sequence = []
for _ in range(max_length // n_tokens_predict):
best_actions = mcts.search_with_beam(current_state)
thought_sequence.extend(best_actions)
# Apply the best actions to get the next state
for action in best_actions:
current_state = current_state.apply_action(action)
# Check if we've reached a leaf node (no further actions)
if len(current_state.thought_node.children) == 0:
break
return thought_sequence
else:
# Use the world model without Tree of Thought, but with multi-token beam search
beam = [(initial_state, 0.0, torch.zeros(1, device=device), torch.zeros(1, device=device))] # (state, score, cum_entropy, cum_variance)
for _ in range(max_length // n_tokens_predict):
all_candidates = []
for state, score, cum_entropy, cum_variance in beam:
policy_logits, _ = prediction_network(state.representation)
probs = F.softmax(policy_logits / args.temperature, dim=-1)
entropy = -torch.sum(probs * torch.log(probs + 1e-9), dim=-1)
variance = torch.var(probs, dim=-1)
topk_probs, topk_indices = torch.topk(probs, k=beam_size, dim=-1)
for i in range(beam_size ** n_tokens_predict):
indices = [i // (beam_size ** j) % beam_size for j in range(n_tokens_predict)]
new_actions = [index_to_action[topk_indices[0, j, indices[j]].item()] for j in range(n_tokens_predict)]
new_score = score + torch.sum(torch.log(topk_probs[0, range(n_tokens_predict), indices]))
new_entropy = cum_entropy + torch.sum(entropy[0, indices])
new_variance = cum_variance + torch.sum(variance[0, indices])
new_state = state
for action in new_actions:
new_state = new_state.apply_action(action)
all_candidates.append((new_state, new_score, new_entropy, new_variance, new_actions))
# Select top beam_size candidates
beam = sorted(all_candidates, key=lambda x: x[1] - 0.1 * x[2] + 0.05 * x[3], reverse=True)[:beam_size]
# Accumulate actions
if not thought_sequence:
thought_sequence = [b[4] for b in beam]
else:
for i, b in enumerate(beam):
thought_sequence[i].extend(b[4])
# Return the top sequence
return thought_sequence[0]
def train_epoch_world_model(world_model_components, train_loader, optimizer, scheduler, scaler, args, model_transformer, state_dim, embed_dim, input_dim):
representation_network, dynamics_network, prediction_network, action_encoder, ppo_agent, _ = world_model_components
representation_network.train()
dynamics_network.train()
prediction_network.train()
action_encoder.train()
ppo_agent.policy_network.train()
total_loss = 0.0
optimizer.zero_grad()
print(f"Starting World Model training epoch with {len(train_loader)} batches...")
for i, batch in enumerate(train_loader):
print(f"Processing batch {i+1}/{len(train_loader)}...")
# Move batches to the device
src_batch = batch['input_ids'].to(device)
tgt_batch = batch['labels'].to(device)
with torch.amp.autocast(device_type='cuda'):
print("Forward pass through Transformer (frozen)...")
with torch.no_grad():
transformer_output = model_transformer(src_batch, tgt_batch[:, :-1])
# World Model - Representation
state_representation = representation_network(transformer_output)
# For simplicity, let's assume true actions are provided (e.g., next tokens)
true_actions = tgt_batch[:, :-1]
action_sequences = true_actions
# Get action embeddings
action_embeddings = action_encoder(action_sequences)
# Apply dynamics network
predicted_next_state_batch = dynamics_network(state_representation, action_embeddings)
# Prediction Network - Policy logits and value
policy_logits, value_estimates = prediction_network(predicted_next_state_batch)
# Define true_policy and true_value as placeholders on the GPU
true_policy = F.one_hot(true_actions, num_classes=input_dim).float()
true_value = torch.zeros_like(value_estimates).to(device)
# Compute individual losses
ppo_loss = ppo_agent.compute_loss(
state_representation,
torch.zeros_like(true_actions, dtype=torch.float32).to(device),
true_actions,
torch.zeros_like(value_estimates, dtype=torch.float32).to(device),
torch.zeros_like(value_estimates, dtype=torch.float32).to(device)
)
info_nce = InfoNCE_Loss()(
state_representation.reshape(-1, state_dim),
F.dropout(state_representation.reshape(-1, state_dim), p=0.1, training=True)
)
covariance = CovarianceRegularization()(predicted_next_state_batch.view(-1, predicted_next_state_batch.size(-1)))
dynamics_loss = DynamicsPerformanceLoss()(state_representation, predicted_next_state_batch)
perturbed_next_state = predicted_next_state_batch + torch.randn_like(predicted_next_state_batch) * 0.01
thought_loss = ThoughtConsistencyLoss()(predicted_next_state_batch, perturbed_next_state)
pv_loss = PolicyValueJointLoss()(policy_logits, true_policy, value_estimates.squeeze(-1), true_value.squeeze(-1))
action_diversity = ActionDiversityReward()(action_embeddings.view(-1, embed_dim))
mcts_best_values = torch.zeros(true_actions.size(0)).to(device)
etv = ExpectedThoughtValueLoss()(mcts_best_values)
visit_counts = torch.ones(true_actions.size(0), policy_logits.size(-1)).to(device)
exploration = ExplorationRegularization()(visit_counts)
old_policy = F.softmax(policy_logits.detach(), dim=-1)
new_policy = F.softmax(policy_logits, dim=-1)
kl_loss = KL_DivergenceLoss()(old_policy, new_policy)
# Total Loss
loss = (
ppo_loss +
info_nce +
covariance +
dynamics_loss +
thought_loss +
pv_loss +
action_diversity +
etv +
exploration +
kl_loss
)
loss = loss / args.accumulation_steps
print("Backward pass...")
scaler.scale(loss).backward()
if (i + 1) % args.accumulation_steps == 0 or (i + 1) == len(train_loader):
print("Gradient clipping...")
scaler.unscale_(optimizer)
torch.nn.utils.clip_grad_norm_(
[param for group in optimizer.param_groups for param in group['params']],
args.max_grad_norm
)
print("Optimizer step...")
scaler.step(optimizer)
scaler.update()
print("Zeroing gradients...")
optimizer.zero_grad()
print("Updating learning rate...")
scheduler.step()
total_loss += loss.item() * args.accumulation_steps
# Print individual losses and total loss for this batch
print(f"Batch {i+1} completed. Losses:")
print(f" PPO Loss: {ppo_loss.item():.4f}")
print(f" InfoNCE Loss: {info_nce.item():.4f}")
print(f" Covariance Loss: {covariance.item():.4f}")
print(f" Dynamics Loss: {dynamics_loss.item():.4f}")
print(f" Thought Consistency Loss: {thought_loss.item():.4f}")
print(f" Policy-Value Loss: {pv_loss.item():.4f}")
print(f" Action Diversity Loss: {action_diversity.item():.4f}")
print(f" Expected Thought Value Loss: {etv.item():.4f}")
print(f" Exploration Loss: {exploration.item():.4f}")
print(f" KL Divergence Loss: {kl_loss.item():.4f}")
print(f" Total Loss: {loss.item():.4f}")
avg_loss = total_loss / len(train_loader)
print(f"World Model training epoch completed. Average loss: {avg_loss:.4f}")
return avg_loss
def train_epoch_language_model(model, train_loader, optimizer, scheduler, scaler, args):
model.train()
total_loss = 0.0
optimizer.zero_grad()
print(f"Starting Language Model training epoch with {len(train_loader)} batches...")
for i, batch in enumerate(train_loader):
input_ids = batch['input_ids'].to(device)
labels = batch['labels'].to(device)
with autocast():
outputs = model(input_ids, input_ids)
logits = outputs.view(-1, outputs.size(-1))
labels = labels.view(-1)
loss = F.cross_entropy(logits, labels, ignore_index=model.embedding.padding_idx)
loss = loss / args.accumulation_steps
scaler.scale(loss).backward()
if (i + 1) % args.accumulation_steps == 0 or (i + 1) == len(train_loader):
scaler.unscale_(optimizer)
torch.nn.utils.clip_grad_norm_(
[param for group in optimizer.param_groups for param in group['params']],
args.max_grad_norm
)
scaler.step(optimizer)
scaler.update()
optimizer.zero_grad()
scheduler.step()
total_loss += loss.item() * args.accumulation_steps
print(f"Batch {i + 1} completed. Current loss: {loss.item():.4f}")
avg_loss = total_loss / len(train_loader)
print(f"Language Model training epoch completed. Average loss: {avg_loss:.4f}")
return avg_loss
def train_custom_data_epoch_world_model(world_model_components, train_loader, optimizer, scheduler, scaler, args, model_transformer, state_dim, embed_dim, input_dim):
representation_network, dynamics_network, prediction_network, action_encoder, ppo_agent, _ = world_model_components
representation_network.train()
dynamics_network.train()
prediction_network.train()
action_encoder.train()
ppo_agent.policy_network.train()
total_loss = 0.0
optimizer.zero_grad()
print(f"Starting World Model training epoch with {len(train_loader)} batches...")
for i, batch in enumerate(train_loader):
print(f"Processing batch {i+1}/{len(train_loader)}...")
# Move batches to the device
input_ids = batch['input_ids'].to(device)
attention_mask = batch['attention_mask'].to(device)
episode_reward = batch['episode_reward'].to(device)
loss_value = batch['loss'].to(device)
cosine_similarity = batch['cosine_similarity'].to(device)
rag_performance = batch['rag_performance'].to(device)
ranking_model_performance = batch['ranking_model_performance'].to(device)
with torch.amp.autocast(device_type='cuda'):
print("Forward pass through Transformer (frozen)...")
with torch.no_grad():
transformer_output = model_transformer(input_ids, input_ids)
# World Model - Representation
state_representation = representation_network(transformer_output)
print(f"State representation shape: {state_representation.shape}")
# For simplicity, let's assume true actions are provided (e.g., next tokens)
true_actions = input_ids[:, 1:] # Shift input_ids by 1 to get next tokens
print(f"True actions shape: {true_actions.shape}")
action_sequences = true_actions
# Get action embeddings
action_embeddings = action_encoder(action_sequences)
print(f"Action embeddings shape: {action_embeddings.shape}")
# Ensure state_representation and action_embeddings have the same sequence length
min_seq_len = min(state_representation.size(1), action_embeddings.size(1))
state_representation = state_representation[:, :min_seq_len, :]
action_embeddings = action_embeddings[:, :min_seq_len, :]
print(f"Adjusted state representation shape: {state_representation.shape}")
print(f"Adjusted action embeddings shape: {action_embeddings.shape}")
# Apply dynamics network
predicted_next_state_batch = dynamics_network(state_representation, action_embeddings)
print(f"Predicted next state batch shape: {predicted_next_state_batch.shape}")
# Prediction Network - Policy logits and value
policy_logits, value_estimates = prediction_network(predicted_next_state_batch)
# Adjust true_actions to match the sequence length
true_actions = true_actions[:, :min_seq_len]
# Define true_policy and true_value
true_policy = F.one_hot(true_actions, num_classes=input_dim).float()
true_value = episode_reward.unsqueeze(1).expand(-1, min_seq_len) # Expand to match sequence length
# Compute individual losses
info_nce = InfoNCE_Loss()(
state_representation.reshape(-1, state_dim),
F.dropout(state_representation.reshape(-1, state_dim), p=0.1, training=True)
)
covariance = CovarianceRegularization()(predicted_next_state_batch.view(-1, predicted_next_state_batch.size(-1)))
dynamics_loss = DynamicsPerformanceLoss()(state_representation, predicted_next_state_batch)
perturbed_next_state = predicted_next_state_batch + torch.randn_like(predicted_next_state_batch) * 0.01
thought_loss = ThoughtConsistencyLoss()(predicted_next_state_batch, perturbed_next_state)
pv_loss = PolicyValueJointLoss()(policy_logits, true_policy, value_estimates.squeeze(-1), true_value.squeeze(-1))
action_diversity = ActionDiversityReward()(action_embeddings.view(-1, embed_dim))
mcts_best_values = torch.zeros(true_actions.size(0)).to(device)
etv = ExpectedThoughtValueLoss()(mcts_best_values)
visit_counts = torch.ones(true_actions.size(0), policy_logits.size(-1)).to(device)
exploration = ExplorationRegularization()(visit_counts)
old_policy = F.softmax(policy_logits.detach(), dim=-1)
new_policy = F.softmax(policy_logits, dim=-1)
kl_loss = KL_DivergenceLoss()(old_policy, new_policy)
# Compute mean value estimates over the sequence length
value_estimates_mean = value_estimates.squeeze(-1).mean(dim=1) # Shape: [batch_size]
# Add new loss components
rag_loss = F.mse_loss(value_estimates_mean, rag_performance)
ranking_loss = F.mse_loss(value_estimates_mean, ranking_model_performance)
cosine_similarity_loss = 1 - cosine_similarity.mean() # Maximize cosine similarity
# Total Loss
loss = (
info_nce +
covariance +
dynamics_loss +
thought_loss +
pv_loss +
action_diversity +
etv +
exploration +
kl_loss +
rag_loss +
ranking_loss +
cosine_similarity_loss
)
loss = loss / args.accumulation_steps
print("Backward pass...")
scaler.scale(loss).backward()
if (i + 1) % args.accumulation_steps == 0 or (i + 1) == len(train_loader):
print("Gradient clipping...")
scaler.unscale_(optimizer)
torch.nn.utils.clip_grad_norm_(
[param for group in optimizer.param_groups for param in group['params']],
args.max_grad_norm
)
print("Optimizer step...")
scaler.step(optimizer)
scaler.update()
print("Zeroing gradients...")
optimizer.zero_grad()
print("Updating learning rate...")
scheduler.step()
# Print individual losses and total loss for this batch
print(f"Batch {i+1} completed. Losses:")
print(f" InfoNCE Loss: {info_nce.item():.4f}")
print(f" Covariance Loss: {covariance.item():.4f}")
print(f" Dynamics Loss: {dynamics_loss.item():.4f}")
print(f" Thought Consistency Loss: {thought_loss.item():.4f}")
print(f" Policy-Value Loss: {pv_loss.item():.4f}")
print(f" Action Diversity Loss: {action_diversity.item():.4f}")
print(f" Expected Thought Value Loss: {etv.item():.4f}")
print(f" Exploration Loss: {exploration.item():.4f}")
print(f" KL Divergence Loss: {kl_loss.item():.4f}")
print(f" RAG Loss: {rag_loss.item():.4f}")
print(f" Ranking Loss: {ranking_loss.item():.4f}")
print(f" Cosine Similarity Loss: {cosine_similarity_loss.item():.4f}")
print(f" Total Loss: {loss.item():.4f}")
avg_loss = total_loss / len(train_loader)
print(f"World Model training epoch completed. Average loss: {avg_loss:.4f}")
return avg_loss
def main():
args = parse_args()
print("Arguments parsed successfully.")
# Create save directory
os.makedirs(args.save_dir, exist_ok=True)
print(f"Save directory created: {args.save_dir}")
# Load tokenizer
print("Loading tokenizer...")
tokenizer = AutoTokenizer.from_pretrained(args.model_name)
if tokenizer.pad_token is None:
tokenizer.pad_token = tokenizer.eos_token
print("Tokenizer loaded successfully.")
# Define padding_idx and input dimension based on tokenizer
padding_idx = tokenizer.pad_token_id
input_dim = len(tokenizer)
# Initialize the Transformer model on GPU
print("Initializing Transformer model...")
model_transformer = Transformer(
input_dim=input_dim,
d_model=128,
num_heads=4,
num_layers=4,
d_ff=256,
num_experts=2,
output_dim=input_dim,
dropout=0.1,
top_k=2
).to(device)
model_transformer.train()
print("Transformer model initialized on device.")
# Define model parameters (adjusted for speed)
d_model = 32
state_dim = 32
action_dim = d_model
hidden_dim = 64
vocab_dim = input_dim
embed_dim = d_model
# Define World Model components
representation_network = RepresentationNetwork(vocab_dim, d_model, state_dim).to(device)
dynamics_network = DynamicsNetwork(state_dim, action_dim, hidden_dim).to(device)
prediction_network = PredictionNetwork(state_dim, input_dim, 1).to(device)
action_encoder = ActionEncoder(input_dim, action_dim).to(device)
# Initialize PPO Agent
ppo_agent = PPOAgent(
policy_network=prediction_network,
optimizer=optim.AdamW(prediction_network.parameters(), lr=args.learning_rate),
clip_epsilon=0.2,
entropy_coef=0.01,
value_coef=0.5
)
# Bundle World Model components
world_model_components = (representation_network, dynamics_network, prediction_network, action_encoder, ppo_agent, model_transformer)
print(f"Current mode: {args.mode}")
if args.mode == 'train':
print("Loading and preprocessing data...")
if args.use_custom_data:
custom_data = load_custom_data_from_files(args.custom_data_paths)
processed_data = preprocess_custom_data(custom_data)
train_loader, eval_loader = load_custom_data(args, tokenizer, processed_data)
print("Custom data loaded and preprocessed successfully.")
else:
train_loader, eval_loader = load_data(args, tokenizer)
print("Default data loaded and preprocessed successfully.")
# Optimizer and Scheduler
optimizer = optim.AdamW(
list(representation_network.parameters()) +
list(dynamics_network.parameters()) +
list(prediction_network.parameters()) +
list(action_encoder.parameters()),
lr=args.learning_rate, weight_decay=args.weight_decay
) if args.train_mode == 'world_model' else optim.AdamW(model_transformer.parameters(), lr=args.learning_rate)
scheduler = CosineAnnealingLR(optimizer, T_max=args.num_epochs)
scaler = GradScaler()
print(f"Starting {args.train_mode} training...")
for epoch in range(args.num_epochs):
if args.train_mode == 'world_model':
if args.use_custom_data:
avg_loss = train_custom_data_epoch_world_model(
world_model_components,
train_loader,
optimizer,
scheduler,
scaler,
args,
model_transformer,
state_dim,
embed_dim,
input_dim
)
else:
avg_loss = train_epoch_world_model(
world_model_components,
train_loader,
optimizer,
scheduler,
scaler,
args,
model_transformer,
state_dim,
embed_dim,
input_dim
)
else:
avg_loss = train_epoch_language_model(
model_transformer,
train_loader,
optimizer,
scheduler,
scaler,
args
)
print(f"{args.train_mode.capitalize()} training epoch {epoch + 1} completed. Average loss: {avg_loss:.4f}")
# Save models
if args.train_mode == 'world_model':
save_all_models(model_transformer, representation_network, dynamics_network, prediction_network, action_encoder, args.save_dir, epoch + 1)
print(f"Models saved for epoch {epoch + 1}")
else:
torch.save(model_transformer.state_dict(), os.path.join(args.save_dir, f'language_model_epoch_{epoch + 1}.pt'))
print(f"Language model saved for epoch {epoch + 1}")
print("Training completed.")
elif args.mode == 'inference':
print("Entering inference mode...")
# Build Tree of Thought if needed
print("Building Tree of Thought...")
tree_root = build_tree_of_thought()
print("Tree of Thought built successfully.")
# Generate action list
print("Generating action list...")
action_list = []
traverse_tree(tree_root, action_list)
print(f"Action list generated. Total actions: {len(action_list)}")
# Create mappings
global action_to_index, index_to_action
action_to_index = {action: idx for idx, action in enumerate(action_list)}
index_to_action = {idx: action for action, idx in action_to_index.items()}
action_vocab_size = len(action_list)
print(f"Action mappings created. Vocabulary size: {action_vocab_size}")
# Initialize or load models based on the load_model argument
if args.load_model:
print(f"Loading saved model from {args.load_model}")
# Load the saved models
model_transformer.load_state_dict(torch.load(os.path.join(args.load_model, 'transformer_model.pt')))
representation_network.load_state_dict(torch.load(os.path.join(args.load_model, 'representation_network.pt')))
dynamics_network.load_state_dict(torch.load(os.path.join(args.load_model, 'dynamics_network.pt')))
# Load prediction network and adjust its size if necessary
saved_state_dict = torch.load(os.path.join(args.load_model, 'prediction_network.pt'))
saved_vocab_size = saved_state_dict['policy_head.weight'].size(0)
if saved_vocab_size != action_vocab_size:
print(f"Adjusting prediction network size from {saved_vocab_size} to {action_vocab_size}")
prediction_network = PredictionNetwork(state_dim, saved_vocab_size, 1).to(device)
prediction_network.load_state_dict(saved_state_dict)
prediction_network.policy_head = nn.Linear(prediction_network.state_dim, action_vocab_size).to(device)
else:
prediction_network = PredictionNetwork(state_dim, action_vocab_size, 1).to(device)
prediction_network.load_state_dict(saved_state_dict)
action_encoder.load_state_dict(torch.load(os.path.join(args.load_model, 'action_encoder.pt')))
else:
print("Using newly initialized models")
# Prepare the components
world_model_components = (representation_network, dynamics_network, prediction_network, action_encoder, ppo_agent, model_transformer)
print("Starting inference loop...")
while True:
if args.query:
query = args.query
args.query = None # Reset query for next iteration
else:
query = input("Please enter your query (or type 'exit' to quit): ")
if query.lower() == 'exit':
break
print(f"Processing query: {query}")
result = infer(query, world_model_components, tree_root, tokenizer,
max_length=args.max_length,
inference_mode=args.inference_mode,
beam_size=args.beam_size,
n_tokens_predict=args.n_tokens_predict,
mcts_iterations=args.mcts_iterations,
exploration_constant=args.mcts_exploration_constant)
if args.inference_mode == 'without_world_model':
print("Generated Text:")
print(result)
else:
print("Generated Thought Sequence:")
for thought in result:
print(thought)
print("\n") # Add a newline for better readability between queries
print("Inference completed.")
else:
print(f"Invalid mode: {args.mode}. Please choose 'train' or 'inference'.")
if __name__ == '__main__':
sys.argv = [
'lightbulb_2.py',
'--mode', 'inference',
'--train_mode', 'world_model', # Set 'world_model' or 'language_model' depending on the training mode
'--dataset_name', 'wikitext', # Specify the Hugging Face dataset (e.g., 'wikitext')
'--dataset_config', 'wikitext-2-raw-v1', # Use if you need a specific config of the dataset
'--num_epochs', '10',
'--batch_size', '4',
'--accumulation_steps', '1',
'--max_grad_norm', '1.0',
'--weight_decay', '0.01',
'--learning_rate', '1e-4',
'--max_length', '512',
'--save_dir', './trained_models',
# Uncomment the following line to use custom data instead of a Hugging Face dataset
#'--use_custom_data',
'--custom_data_paths', '/content/drive/MyDrive/lightbulb/knowledge_base.json',
'--custom_data_paths', '/content/drive/MyDrive/lightbulb/rag_cache.json',
'--custom_data_paths', '/content/drive/MyDrive/lightbulb/llm_training_data/llm_training_data.jsonl'
]
# Parse the arguments and run the main training function
args = parse_args()
# Check which data source to use
if args.use_custom_data:
print("Training with custom data from paths:")
for path in args.custom_data_paths:
print(f" - {path}")
else:
print(f"Training with dataset '{args.dataset_name}' from Hugging Face Datasets")
main()
|