API Reference
High Level API
High-level Python bindings for llama.cpp.
llama_cpp.Llama
High-level Python wrapper for a llama.cpp model.
Source code in llama_cpp/llama.py
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 |
|
__init__(model_path, *, n_gpu_layers=0, split_mode=llama_cpp.LLAMA_SPLIT_MODE_LAYER, main_gpu=0, tensor_split=None, vocab_only=False, use_mmap=True, use_mlock=False, kv_overrides=None, seed=llama_cpp.LLAMA_DEFAULT_SEED, n_ctx=512, n_batch=512, n_threads=None, n_threads_batch=None, rope_scaling_type=llama_cpp.LLAMA_ROPE_SCALING_TYPE_UNSPECIFIED, pooling_type=llama_cpp.LLAMA_POOLING_TYPE_UNSPECIFIED, rope_freq_base=0.0, rope_freq_scale=0.0, yarn_ext_factor=-1.0, yarn_attn_factor=1.0, yarn_beta_fast=32.0, yarn_beta_slow=1.0, yarn_orig_ctx=0, logits_all=False, embedding=False, offload_kqv=True, last_n_tokens_size=64, lora_base=None, lora_scale=1.0, lora_path=None, numa=False, chat_format=None, chat_handler=None, draft_model=None, tokenizer=None, type_k=None, type_v=None, verbose=True, **kwargs)
Load a llama.cpp model from model_path
.
Examples:
Basic usage
>>> import llama_cpp
>>> model = llama_cpp.Llama(
... model_path="path/to/model",
... )
>>> print(model("The quick brown fox jumps ", stop=["."])["choices"][0]["text"])
the lazy dog
Loading a chat model
>>> import llama_cpp
>>> model = llama_cpp.Llama(
... model_path="path/to/model",
... chat_format="llama-2",
... )
>>> print(model.create_chat_completion(
... messages=[{
... "role": "user",
... "content": "what is the meaning of life?"
... }]
... ))
Parameters:
-
model_path
(str
) –Path to the model.
-
n_gpu_layers
(int
, default:0
) –Number of layers to offload to GPU (-ngl). If -1, all layers are offloaded.
-
split_mode
(int
, default:LLAMA_SPLIT_MODE_LAYER
) –How to split the model across GPUs. See llama_cpp.LLAMA_SPLIT_* for options.
-
main_gpu
(int
, default:0
) –main_gpu interpretation depends on split_mode: LLAMA_SPLIT_NONE: the GPU that is used for the entire model. LLAMA_SPLIT_ROW: the GPU that is used for small tensors and intermediate results. LLAMA_SPLIT_LAYER: ignored
-
tensor_split
(Optional[List[float]]
, default:None
) –How split tensors should be distributed across GPUs. If None, the model is not split.
-
vocab_only
(bool
, default:False
) –Only load the vocabulary no weights.
-
use_mmap
(bool
, default:True
) –Use mmap if possible.
-
use_mlock
(bool
, default:False
) –Force the system to keep the model in RAM.
-
kv_overrides
(Optional[Dict[str, Union[bool, int, float, str]]]
, default:None
) –Key-value overrides for the model.
-
seed
(int
, default:LLAMA_DEFAULT_SEED
) –RNG seed, -1 for random
-
n_ctx
(int
, default:512
) –Text context, 0 = from model
-
n_batch
(int
, default:512
) –Prompt processing maximum batch size
-
n_threads
(Optional[int]
, default:None
) –Number of threads to use for generation
-
n_threads_batch
(Optional[int]
, default:None
) –Number of threads to use for batch processing
-
rope_scaling_type
(Optional[int]
, default:LLAMA_ROPE_SCALING_TYPE_UNSPECIFIED
) –RoPE scaling type, from
enum llama_rope_scaling_type
. ref: https://github.com/ggerganov/llama.cpp/pull/2054 -
pooling_type
(int
, default:LLAMA_POOLING_TYPE_UNSPECIFIED
) –Pooling type, from
enum llama_pooling_type
. -
rope_freq_base
(float
, default:0.0
) –RoPE base frequency, 0 = from model
-
rope_freq_scale
(float
, default:0.0
) –RoPE frequency scaling factor, 0 = from model
-
yarn_ext_factor
(float
, default:-1.0
) –YaRN extrapolation mix factor, negative = from model
-
yarn_attn_factor
(float
, default:1.0
) –YaRN magnitude scaling factor
-
yarn_beta_fast
(float
, default:32.0
) –YaRN low correction dim
-
yarn_beta_slow
(float
, default:1.0
) –YaRN high correction dim
-
yarn_orig_ctx
(int
, default:0
) –YaRN original context size
-
logits_all
(bool
, default:False
) –Return logits for all tokens, not just the last token. Must be True for completion to return logprobs.
-
embedding
(bool
, default:False
) –Embedding mode only.
-
offload_kqv
(bool
, default:True
) –Offload K, Q, V to GPU.
-
last_n_tokens_size
(int
, default:64
) –Maximum number of tokens to keep in the last_n_tokens deque.
-
lora_base
(Optional[str]
, default:None
) –Optional path to base model, useful if using a quantized base model and you want to apply LoRA to an f16 model.
-
lora_path
(Optional[str]
, default:None
) –Path to a LoRA file to apply to the model.
-
numa
(Union[bool, int]
, default:False
) –numa policy
-
chat_format
(Optional[str]
, default:None
) –String specifying the chat format to use when calling create_chat_completion.
-
chat_handler
(Optional[LlamaChatCompletionHandler]
, default:None
) –Optional chat handler to use when calling create_chat_completion.
-
draft_model
(Optional[LlamaDraftModel]
, default:None
) –Optional draft model to use for speculative decoding.
-
tokenizer
(Optional[BaseLlamaTokenizer]
, default:None
) –Optional tokenizer to override the default tokenizer from llama.cpp.
-
verbose
(bool
, default:True
) –Print verbose output to stderr.
-
type_k
(Optional[int]
, default:None
) –KV cache data type for K (default: f16)
-
type_v
(Optional[int]
, default:None
) –KV cache data type for V (default: f16)
Raises:
-
ValueError
–If the model path does not exist.
Returns:
-
–
A Llama instance.
Source code in llama_cpp/llama.py
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 |
|
tokenize(text, add_bos=True, special=False)
Tokenize a string.
Parameters:
-
text
(bytes
) –The utf-8 encoded string to tokenize.
Raises:
-
RuntimeError
–If the tokenization failed.
Returns:
Source code in llama_cpp/llama.py
detokenize(tokens, prev_tokens=None)
Detokenize a list of tokens.
Parameters:
-
tokens
(List[int]
) –The list of tokens to detokenize.
-
prev_tokens
(Optional[List[int]]
, default:None
) –The list of previous tokens. Offset mapping will be performed if provided
Returns:
-
bytes
–The detokenized string.
Source code in llama_cpp/llama.py
reset()
eval(tokens)
Evaluate a list of tokens.
Parameters:
Source code in llama_cpp/llama.py
sample(top_k=40, top_p=0.95, min_p=0.05, typical_p=1.0, temp=0.8, repeat_penalty=1.1, frequency_penalty=0.0, presence_penalty=0.0, tfs_z=1.0, mirostat_mode=0, mirostat_eta=0.1, mirostat_tau=5.0, penalize_nl=True, logits_processor=None, grammar=None, idx=None)
Sample a token from the model.
Parameters:
-
top_k
(int
, default:40
) –The top-k sampling parameter.
-
top_p
(float
, default:0.95
) –The top-p sampling parameter.
-
temp
(float
, default:0.8
) –The temperature parameter.
-
repeat_penalty
(float
, default:1.1
) –The repeat penalty parameter.
Returns:
-
–
The sampled token.
Source code in llama_cpp/llama.py
generate(tokens, top_k=40, top_p=0.95, min_p=0.05, typical_p=1.0, temp=0.8, repeat_penalty=1.1, reset=True, frequency_penalty=0.0, presence_penalty=0.0, tfs_z=1.0, mirostat_mode=0, mirostat_tau=5.0, mirostat_eta=0.1, penalize_nl=True, logits_processor=None, stopping_criteria=None, grammar=None)
Create a generator of tokens from a prompt.
Examples:
>>> llama = Llama("models/ggml-7b.bin")
>>> tokens = llama.tokenize(b"Hello, world!")
>>> for token in llama.generate(tokens, top_k=40, top_p=0.95, temp=1.0, repeat_penalty=1.1):
... print(llama.detokenize([token]))
Parameters:
-
tokens
(Sequence[int]
) –The prompt tokens.
-
top_k
(int
, default:40
) –The top-k sampling parameter.
-
top_p
(float
, default:0.95
) –The top-p sampling parameter.
-
temp
(float
, default:0.8
) –The temperature parameter.
-
repeat_penalty
(float
, default:1.1
) –The repeat penalty parameter.
-
reset
(bool
, default:True
) –Whether to reset the model state.
Yields:
-
int
–The generated tokens.
Source code in llama_cpp/llama.py
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 |
|
create_embedding(input, model=None)
Embed a string.
Parameters:
Returns:
-
CreateEmbeddingResponse
–An embedding object.
Source code in llama_cpp/llama.py
embed(input, normalize=False, truncate=True, return_count=False)
Embed a string.
Parameters:
Returns:
-
–
A list of embeddings
Source code in llama_cpp/llama.py
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 |
|
create_completion(prompt, suffix=None, max_tokens=16, temperature=0.8, top_p=0.95, min_p=0.05, typical_p=1.0, logprobs=None, echo=False, stop=[], frequency_penalty=0.0, presence_penalty=0.0, repeat_penalty=1.1, top_k=40, stream=False, seed=None, tfs_z=1.0, mirostat_mode=0, mirostat_tau=5.0, mirostat_eta=0.1, model=None, stopping_criteria=None, logits_processor=None, grammar=None, logit_bias=None)
Generate text from a prompt.
Parameters:
-
prompt
(Union[str, List[int]]
) –The prompt to generate text from.
-
suffix
(Optional[str]
, default:None
) –A suffix to append to the generated text. If None, no suffix is appended.
-
max_tokens
(Optional[int]
, default:16
) –The maximum number of tokens to generate. If max_tokens <= 0 or None, the maximum number of tokens to generate is unlimited and depends on n_ctx.
-
temperature
(float
, default:0.8
) –The temperature to use for sampling.
-
top_p
(float
, default:0.95
) –The top-p value to use for nucleus sampling. Nucleus sampling described in academic paper "The Curious Case of Neural Text Degeneration" https://arxiv.org/abs/1904.09751
-
min_p
(float
, default:0.05
) –The min-p value to use for minimum p sampling. Minimum P sampling as described in https://github.com/ggerganov/llama.cpp/pull/3841
-
typical_p
(float
, default:1.0
) –The typical-p value to use for sampling. Locally Typical Sampling implementation described in the paper https://arxiv.org/abs/2202.00666.
-
logprobs
(Optional[int]
, default:None
) –The number of logprobs to return. If None, no logprobs are returned.
-
echo
(bool
, default:False
) –Whether to echo the prompt.
-
stop
(Optional[Union[str, List[str]]]
, default:[]
) –A list of strings to stop generation when encountered.
-
frequency_penalty
(float
, default:0.0
) –The penalty to apply to tokens based on their frequency in the prompt.
-
presence_penalty
(float
, default:0.0
) –The penalty to apply to tokens based on their presence in the prompt.
-
repeat_penalty
(float
, default:1.1
) –The penalty to apply to repeated tokens.
-
top_k
(int
, default:40
) –The top-k value to use for sampling. Top-K sampling described in academic paper "The Curious Case of Neural Text Degeneration" https://arxiv.org/abs/1904.09751
-
stream
(bool
, default:False
) –Whether to stream the results.
-
seed
(Optional[int]
, default:None
) –The seed to use for sampling.
-
tfs_z
(float
, default:1.0
) –The tail-free sampling parameter. Tail Free Sampling described in https://www.trentonbricken.com/Tail-Free-Sampling/.
-
mirostat_mode
(int
, default:0
) –The mirostat sampling mode.
-
mirostat_tau
(float
, default:5.0
) –The target cross-entropy (or surprise) value you want to achieve for the generated text. A higher value corresponds to more surprising or less predictable text, while a lower value corresponds to less surprising or more predictable text.
-
mirostat_eta
(float
, default:0.1
) –The learning rate used to update
mu
based on the error between the target and observed surprisal of the sampled word. A larger learning rate will causemu
to be updated more quickly, while a smaller learning rate will result in slower updates. -
model
(Optional[str]
, default:None
) –The name to use for the model in the completion object.
-
stopping_criteria
(Optional[StoppingCriteriaList]
, default:None
) –A list of stopping criteria to use.
-
logits_processor
(Optional[LogitsProcessorList]
, default:None
) –A list of logits processors to use.
-
grammar
(Optional[LlamaGrammar]
, default:None
) –A grammar to use for constrained sampling.
-
logit_bias
(Optional[Dict[str, float]]
, default:None
) –A logit bias to use.
Raises:
-
ValueError
–If the requested tokens exceed the context window.
-
RuntimeError
–If the prompt fails to tokenize or the model fails to evaluate the prompt.
Returns:
-
Union[CreateCompletionResponse, Iterator[CreateCompletionStreamResponse]]
–Response object containing the generated text.
Source code in llama_cpp/llama.py
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 |
|
__call__(prompt, suffix=None, max_tokens=16, temperature=0.8, top_p=0.95, min_p=0.05, typical_p=1.0, logprobs=None, echo=False, stop=[], frequency_penalty=0.0, presence_penalty=0.0, repeat_penalty=1.1, top_k=40, stream=False, seed=None, tfs_z=1.0, mirostat_mode=0, mirostat_tau=5.0, mirostat_eta=0.1, model=None, stopping_criteria=None, logits_processor=None, grammar=None, logit_bias=None)
Generate text from a prompt.
Parameters:
-
prompt
(str
) –The prompt to generate text from.
-
suffix
(Optional[str]
, default:None
) –A suffix to append to the generated text. If None, no suffix is appended.
-
max_tokens
(Optional[int]
, default:16
) –The maximum number of tokens to generate. If max_tokens <= 0 or None, the maximum number of tokens to generate is unlimited and depends on n_ctx.
-
temperature
(float
, default:0.8
) –The temperature to use for sampling.
-
top_p
(float
, default:0.95
) –The top-p value to use for nucleus sampling. Nucleus sampling described in academic paper "The Curious Case of Neural Text Degeneration" https://arxiv.org/abs/1904.09751
-
min_p
(float
, default:0.05
) –The min-p value to use for minimum p sampling. Minimum P sampling as described in https://github.com/ggerganov/llama.cpp/pull/3841
-
typical_p
(float
, default:1.0
) –The typical-p value to use for sampling. Locally Typical Sampling implementation described in the paper https://arxiv.org/abs/2202.00666.
-
logprobs
(Optional[int]
, default:None
) –The number of logprobs to return. If None, no logprobs are returned.
-
echo
(bool
, default:False
) –Whether to echo the prompt.
-
stop
(Optional[Union[str, List[str]]]
, default:[]
) –A list of strings to stop generation when encountered.
-
frequency_penalty
(float
, default:0.0
) –The penalty to apply to tokens based on their frequency in the prompt.
-
presence_penalty
(float
, default:0.0
) –The penalty to apply to tokens based on their presence in the prompt.
-
repeat_penalty
(float
, default:1.1
) –The penalty to apply to repeated tokens.
-
top_k
(int
, default:40
) –The top-k value to use for sampling. Top-K sampling described in academic paper "The Curious Case of Neural Text Degeneration" https://arxiv.org/abs/1904.09751
-
stream
(bool
, default:False
) –Whether to stream the results.
-
seed
(Optional[int]
, default:None
) –The seed to use for sampling.
-
tfs_z
(float
, default:1.0
) –The tail-free sampling parameter. Tail Free Sampling described in https://www.trentonbricken.com/Tail-Free-Sampling/.
-
mirostat_mode
(int
, default:0
) –The mirostat sampling mode.
-
mirostat_tau
(float
, default:5.0
) –The target cross-entropy (or surprise) value you want to achieve for the generated text. A higher value corresponds to more surprising or less predictable text, while a lower value corresponds to less surprising or more predictable text.
-
mirostat_eta
(float
, default:0.1
) –The learning rate used to update
mu
based on the error between the target and observed surprisal of the sampled word. A larger learning rate will causemu
to be updated more quickly, while a smaller learning rate will result in slower updates. -
model
(Optional[str]
, default:None
) –The name to use for the model in the completion object.
-
stopping_criteria
(Optional[StoppingCriteriaList]
, default:None
) –A list of stopping criteria to use.
-
logits_processor
(Optional[LogitsProcessorList]
, default:None
) –A list of logits processors to use.
-
grammar
(Optional[LlamaGrammar]
, default:None
) –A grammar to use for constrained sampling.
-
logit_bias
(Optional[Dict[str, float]]
, default:None
) –A logit bias to use.
Raises:
-
ValueError
–If the requested tokens exceed the context window.
-
RuntimeError
–If the prompt fails to tokenize or the model fails to evaluate the prompt.
Returns:
-
Union[CreateCompletionResponse, Iterator[CreateCompletionStreamResponse]]
–Response object containing the generated text.
Source code in llama_cpp/llama.py
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 |
|
create_chat_completion(messages, functions=None, function_call=None, tools=None, tool_choice=None, temperature=0.2, top_p=0.95, top_k=40, min_p=0.05, typical_p=1.0, stream=False, stop=[], seed=None, response_format=None, max_tokens=None, presence_penalty=0.0, frequency_penalty=0.0, repeat_penalty=1.1, tfs_z=1.0, mirostat_mode=0, mirostat_tau=5.0, mirostat_eta=0.1, model=None, logits_processor=None, grammar=None, logit_bias=None, logprobs=None, top_logprobs=None)
Generate a chat completion from a list of messages.
Parameters:
-
messages
(List[ChatCompletionRequestMessage]
) –A list of messages to generate a response for.
-
functions
(Optional[List[ChatCompletionFunction]]
, default:None
) –A list of functions to use for the chat completion.
-
function_call
(Optional[ChatCompletionRequestFunctionCall]
, default:None
) –A function call to use for the chat completion.
-
tools
(Optional[List[ChatCompletionTool]]
, default:None
) –A list of tools to use for the chat completion.
-
tool_choice
(Optional[ChatCompletionToolChoiceOption]
, default:None
) –A tool choice to use for the chat completion.
-
temperature
(float
, default:0.2
) –The temperature to use for sampling.
-
top_p
(float
, default:0.95
) –The top-p value to use for nucleus sampling. Nucleus sampling described in academic paper "The Curious Case of Neural Text Degeneration" https://arxiv.org/abs/1904.09751
-
top_k
(int
, default:40
) –The top-k value to use for sampling. Top-K sampling described in academic paper "The Curious Case of Neural Text Degeneration" https://arxiv.org/abs/1904.09751
-
min_p
(float
, default:0.05
) –The min-p value to use for minimum p sampling. Minimum P sampling as described in https://github.com/ggerganov/llama.cpp/pull/3841
-
typical_p
(float
, default:1.0
) –The typical-p value to use for sampling. Locally Typical Sampling implementation described in the paper https://arxiv.org/abs/2202.00666.
-
stream
(bool
, default:False
) –Whether to stream the results.
-
stop
(Optional[Union[str, List[str]]]
, default:[]
) –A list of strings to stop generation when encountered.
-
seed
(Optional[int]
, default:None
) –The seed to use for sampling.
-
response_format
(Optional[ChatCompletionRequestResponseFormat]
, default:None
) –The response format to use for the chat completion. Use { "type": "json_object" } to contstrain output to only valid json.
-
max_tokens
(Optional[int]
, default:None
) –The maximum number of tokens to generate. If max_tokens <= 0 or None, the maximum number of tokens to generate is unlimited and depends on n_ctx.
-
presence_penalty
(float
, default:0.0
) –The penalty to apply to tokens based on their presence in the prompt.
-
frequency_penalty
(float
, default:0.0
) –The penalty to apply to tokens based on their frequency in the prompt.
-
repeat_penalty
(float
, default:1.1
) –The penalty to apply to repeated tokens.
-
tfs_z
(float
, default:1.0
) –The tail-free sampling parameter.
-
mirostat_mode
(int
, default:0
) –The mirostat sampling mode.
-
mirostat_tau
(float
, default:5.0
) –The mirostat sampling tau parameter.
-
mirostat_eta
(float
, default:0.1
) –The mirostat sampling eta parameter.
-
model
(Optional[str]
, default:None
) –The name to use for the model in the completion object.
-
logits_processor
(Optional[LogitsProcessorList]
, default:None
) –A list of logits processors to use.
-
grammar
(Optional[LlamaGrammar]
, default:None
) –A grammar to use.
-
logit_bias
(Optional[Dict[str, float]]
, default:None
) –A logit bias to use.
Returns:
-
Union[CreateChatCompletionResponse, Iterator[CreateChatCompletionStreamResponse]]
–Generated chat completion or a stream of chat completion chunks.
Source code in llama_cpp/llama.py
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 |
|
create_chat_completion_openai_v1(*args, **kwargs)
Generate a chat completion with return type based on the the OpenAI v1 API.
OpenAI python package is required to use this method.
You can install it with pip install openai
.
Parameters:
-
*args
(Any
, default:()
) –Positional arguments to pass to create_chat_completion.
-
**kwargs
(Any
, default:{}
) –Keyword arguments to pass to create_chat_completion.
Returns:
-
–
Generated chat completion or a stream of chat completion chunks.
Source code in llama_cpp/llama.py
set_cache(cache)
save_state()
Source code in llama_cpp/llama.py
load_state(state)
Source code in llama_cpp/llama.py
token_bos()
token_eos()
from_pretrained(repo_id, filename, local_dir=None, local_dir_use_symlinks='auto', cache_dir=None, **kwargs)
classmethod
Create a Llama model from a pretrained model name or path.
This method requires the huggingface-hub package.
You can install it with pip install huggingface-hub
.
Parameters:
-
repo_id
(str
) –The model repo id.
-
filename
(Optional[str]
) –A filename or glob pattern to match the model file in the repo.
-
local_dir
(Optional[Union[str, PathLike[str]]]
, default:None
) –The local directory to save the model to.
-
local_dir_use_symlinks
(Union[bool, Literal['auto']]
, default:'auto'
) –Whether to use symlinks when downloading the model.
-
**kwargs
(Any
, default:{}
) –Additional keyword arguments to pass to the Llama constructor.
Returns:
-
'Llama'
–A Llama model.
Source code in llama_cpp/llama.py
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 |
|
llama_cpp.LlamaGrammar
Keeps reference counts of all the arguments, so that they are not garbage collected by Python.
Source code in llama_cpp/llama_grammar.py
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 |
|
from_string(grammar, verbose=True)
classmethod
Convert a GBNF grammar to a Llama grammar.
Source code in llama_cpp/llama_grammar.py
from_json_schema(json_schema, verbose=True)
classmethod
Convert a JSON schema to a Llama grammar.
llama_cpp.LlamaCache = LlamaRAMCache
module-attribute
llama_cpp.LlamaState
Source code in llama_cpp/llama.py
llama_cpp.LogitsProcessor = Callable[[npt.NDArray[np.intc], npt.NDArray[np.single]], npt.NDArray[np.single]]
module-attribute
llama_cpp.LogitsProcessorList
llama_cpp.StoppingCriteria = Callable[[npt.NDArray[np.intc], npt.NDArray[np.single]], bool]
module-attribute
llama_cpp.StoppingCriteriaList
Low Level API
Low-level Python bindings for llama.cpp using Python's ctypes library.
llama_cpp.llama_cpp
llama_model_p = NewType('llama_model_p', int)
module-attribute
llama_model_p_ctypes = ctypes.c_void_p
module-attribute
llama_context_p = NewType('llama_context_p', int)
module-attribute
llama_context_p_ctypes = ctypes.c_void_p
module-attribute
llama_pos = ctypes.c_int32
module-attribute
llama_token = ctypes.c_int32
module-attribute
llama_token_p = ctypes.POINTER(llama_token)
module-attribute
llama_seq_id = ctypes.c_int32
module-attribute
llama_token_data
Bases: Structure
Used to store token data
Attributes:
-
id
(llama_token
) –token id
-
logit
(float
) –log-odds of the token
-
p
(float
) –probability of the token
Source code in llama_cpp/llama_cpp.py
llama_token_data_p = ctypes.POINTER(llama_token_data)
module-attribute
llama_token_data_array
Bases: Structure
Used to sample tokens given logits
Attributes:
-
data
(Array[llama_token_data]
) –token data
-
size
(int
) –size of the array
-
sorted
(bool
) –whether the array is sorted
Source code in llama_cpp/llama_cpp.py
llama_token_data_array_p = ctypes.POINTER(llama_token_data_array)
module-attribute
llama_progress_callback = ctypes.CFUNCTYPE(ctypes.c_bool, ctypes.c_float, ctypes.c_void_p)
module-attribute
llama_batch
Bases: Structure
Input data for llama_decode
A llama_batch object can contain input about one or many sequences
The provided arrays (i.e. token, embd, pos, etc.) must have size of n_tokens
Attributes:
-
n_tokens
(int
) –number of tokens
-
token
(Array[llama_token]
) –the token ids of the input (used when embd is NULL)
-
embd
(Array[c_float]
) –token embeddings (i.e. float vector of size n_embd) (used when token is NULL)
-
pos
(Array[Array[llama_pos]]
) –the positions of the respective token in the sequence
-
seq_id
(Array[Array[llama_seq_id]]
) –the sequence to which the respective token belongs
-
logits
(Array[c_int8]
) –if zero, the logits for the respective token will not be output
Source code in llama_cpp/llama_cpp.py
llama_model_kv_override_value
llama_model_kv_override
llama_model_params
Bases: Structure
Parameters for llama_model
Attributes:
-
n_gpu_layers
(int
) –number of layers to store in VRAM
-
split_mode
(int
) –how to split the model across multiple GPUs
-
main_gpu
(int
) –the GPU that is used for the entire model. main_gpu interpretation depends on split_mode: LLAMA_SPLIT_NONE: the GPU that is used for the entire model LLAMA_SPLIT_ROW: the GPU that is used for small tensors and intermediate results LLAMA_SPLIT_LAYER: ignored
-
tensor_split
(Array[c_float]
) –proportion of the model (layers or rows) to offload to each GPU, size: llama_max_devices()
-
progress_callback
(llama_progress_callback
) –called with a progress value between 0.0 and 1.0. Pass NULL to disable. If the provided progress_callback returns true, model loading continues. If it returns false, model loading is immediately aborted.
-
progress_callback_user_data
(c_void_p
) –context pointer passed to the progress callback
-
kv_overrides
(Array[llama_model_kv_override]
) –override key-value pairs of the model meta data
-
vocab_only
(bool
) –only load the vocabulary, no weights
-
use_mmap
(bool
) –use mmap if possible
-
use_mlock
(bool
) –force system to keep model in RAM
-
check_tensors
(bool
) –validate model tensor data
Source code in llama_cpp/llama_cpp.py
llama_context_params
Bases: Structure
Parameters for llama_context
Attributes:
-
seed
(int
) –RNG seed, -1 for random
-
n_ctx
(int
) –text context, 0 = from model
-
n_batch
(int
) –logical maximum batch size that can be submitted to llama_decode
-
n_ubatch
(int
) –physical maximum batch size
-
n_seq_max
(int
) –max number of sequences (i.e. distinct states for recurrent models)
-
n_threads
(int
) –number of threads to use for generation
-
n_threads_batch
(int
) –number of threads to use for batch processing
-
rope_scaling_type
(int
) –RoPE scaling type, from
enum llama_rope_scaling_type
-
pooling_type
(int
) –whether to pool (sum) embedding results by sequence id (ignored if no pooling layer)
-
rope_freq_base
(float
) –RoPE base frequency, 0 = from model
-
rope_freq_scale
(float
) –RoPE frequency scaling factor, 0 = from model
-
yarn_ext_factor
(float
) –YaRN extrapolation mix factor, negative = from model
-
yarn_attn_factor
(float
) –YaRN magnitude scaling factor
-
yarn_beta_fast
(float
) –YaRN low correction dim
-
yarn_beta_slow
(float
) –YaRN high correction dim
-
yarn_orig_ctx
(int
) –YaRN original context size
-
defrag_thold
(float
) –defragment the KV cache if holes/size > thold, < 0 disabled (default)
-
cb_eval
(ggml_backend_sched_eval_callback
) –callback for scheduling eval
-
cb_eval_user_data
(c_void_p
) –user data for cb_eval
-
type_k
(int
) –data type for K cache
-
type_v
(int
) –data type for V cache
-
logits_all
(bool
) –the llama_eval() call computes all logits, not just the last one (DEPRECATED - set llama_batch.logits instead)
-
embeddings
(bool
) –if true, extract embeddings (together with logits)
-
offload_kqv
(bool
) –whether to offload the KQV ops (including the KV cache) to GPU
-
abort_callback
(ggml_abort_callback
) –abort callback if it returns true, execution of llama_decode() will be aborted
-
abort_callback_data
(c_void_p
) –data for abort_callback
Source code in llama_cpp/llama_cpp.py
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 |
|
llama_log_callback = ctypes.CFUNCTYPE(None, ctypes.c_int, ctypes.c_char_p, ctypes.c_void_p)
module-attribute
Signature for logging events Note that text includes the new line character at the end for most events. If your logging mechanism cannot handle that, check if the last character is ' ' and strip it if it exists. It might not exist for progress report where '.' is output repeatedly.
llama_model_quantize_params
Bases: Structure
Parameters for llama_model_quantize
Attributes:
-
nthread
(int
) –number of threads to use for quantizing, if <=0 will use std::hardware_concurrency()
-
ftype
(int
) –quantize to this llama_ftype
-
output_tensor_type
(int
) –output tensor type
-
token_embedding_type
(int
) –itoken embeddings tensor type
-
allow_requantize
(bool
) –allow quantizing non-f32/f16 tensors
-
quantize_output_tensor
(bool
) –quantize output.weight
-
only_copy
(bool
) –only copy tensors - ftype, allow_requantize and quantize_output_tensor are ignored
-
pure
(bool
) –quantize all tensors to the default type
-
keep_split
(bool
) –quantize to the same number of shards
-
imatrix
(c_void_p
) –pointer to importance matrix data
-
kv_overrides
(c_void_p
) –pointer to vector containing overrides
Source code in llama_cpp/llama_cpp.py
llama_grammar_p = ctypes.c_void_p
module-attribute
llama_grammar_element
llama_grammar_element_p = ctypes.POINTER(llama_grammar_element)
module-attribute
llama_timings
Bases: Structure
Source code in llama_cpp/llama_cpp.py
llama_chat_message
llama_model_default_params()
llama_context_default_params()
Get default parameters for llama_context
llama_model_quantize_default_params()
Get default parameters for llama_model_quantize
Source code in llama_cpp/llama_cpp.py
llama_backend_init()
Initialize the llama + ggml backend If numa is true, use NUMA optimizations Call once at the start of the program
llama_numa_init(numa)
llama_backend_free()
llama_load_model_from_file(path_model, params)
Source code in llama_cpp/llama_cpp.py
llama_free_model(model)
llama_new_context_with_model(model, params)
Source code in llama_cpp/llama_cpp.py
llama_free(ctx)
llama_time_us()
llama_max_devices()
llama_supports_mmap()
llama_supports_mlock()
llama_supports_gpu_offload()
llama_get_model(ctx)
llama_n_ctx(ctx)
llama_n_batch(ctx)
llama_n_ubatch(ctx)
llama_n_seq_max(ctx)
llama_pooling_type(ctx)
llama_vocab_type(model)
llama_rope_type(model)
llama_n_vocab(model)
llama_n_ctx_train(model)
llama_n_embd(model)
llama_n_layer(model)
llama_rope_freq_scale_train(model)
Get the model's RoPE frequency scaling factor
llama_model_meta_val_str(model, key, buf, buf_size)
Get metadata value as a string by key name
Source code in llama_cpp/llama_cpp.py
llama_model_meta_count(model)
llama_model_meta_key_by_index(model, i, buf, buf_size)
Get metadata key name by index
Source code in llama_cpp/llama_cpp.py
llama_model_meta_val_str_by_index(model, i, buf, buf_size)
Get metadata value as a string by index
Source code in llama_cpp/llama_cpp.py
llama_model_desc(model, buf, buf_size)
Get a string describing the model type
Source code in llama_cpp/llama_cpp.py
llama_model_size(model)
Returns the total size of all the tensors in the model in bytes
llama_model_n_params(model)
Returns the total number of parameters in the model
llama_get_model_tensor(model, name)
Get a llama model tensor
Source code in llama_cpp/llama_cpp.py
llama_model_quantize(fname_inp, fname_out, params)
Returns 0 on success
Source code in llama_cpp/llama_cpp.py
llama_model_apply_lora_from_file(model, path_lora, scale, path_base_model, n_threads)
Apply a LoRA adapter to a loaded model path_base_model is the path to a higher quality model to use as a base for the layers modified by the adapter. Can be NULL to use the current loaded model. The model needs to be reloaded before applying a new adapter, otherwise the adapter will be applied on top of the previous one Returns 0 on success
Source code in llama_cpp/llama_cpp.py
llama_control_vector_apply(lctx, data, len, n_embd, il_start, il_end)
Apply a loaded control vector to a llama_context, or if data is NULL, clear the currently loaded vector. n_embd should be the size of a single layer's control, and data should point to an n_embd x n_layers buffer starting from layer 1. il_start and il_end are the layer range the vector should apply to (both inclusive) See llama_control_vector_load in common to load a control vector.
Source code in llama_cpp/llama_cpp.py
llama_kv_cache_view_cell
Bases: Structure
Information associated with an individual cell in the KV cache view.
Attributes:
-
pos
(llama_pos
) –The position for this cell. Takes KV cache shifts into account. May be negative if the cell is not populated.
Source code in llama_cpp/llama_cpp.py
llama_kv_cache_view
Bases: Structure
Source code in llama_cpp/llama_cpp.py
llama_kv_cache_view_p = ctypes.POINTER(llama_kv_cache_view)
module-attribute
llama_kv_cache_view_init(ctx, n_seq_max)
Create an empty KV cache view. (use only for debugging purposes)
Source code in llama_cpp/llama_cpp.py
llama_kv_cache_view_free(view)
Free a KV cache view. (use only for debugging purposes)
llama_kv_cache_view_update(ctx, view)
Update the KV cache view structure with the current state of the KV cache. (use only for debugging purposes)
Source code in llama_cpp/llama_cpp.py
llama_get_kv_cache_token_count(ctx)
Returns the number of tokens in the KV cache (slow, use only for debug) If a KV cell has multiple sequences assigned to it, it will be counted multiple times
Source code in llama_cpp/llama_cpp.py
llama_get_kv_cache_used_cells(ctx)
Returns the number of used KV cells (i.e. have at least one sequence assigned to them)
Source code in llama_cpp/llama_cpp.py
llama_kv_cache_clear(ctx)
llama_kv_cache_seq_rm(ctx, seq_id, p0, p1)
Removes all tokens that belong to the specified sequence and have positions in [p0, p1)
Returns false if a partial sequence cannot be removed. Removing a whole sequence never fails
seq_id < 0 : match any sequence p0 < 0 : [0, p1] p1 < 0 : [p0, inf)
Source code in llama_cpp/llama_cpp.py
llama_kv_cache_seq_cp(ctx, seq_id_src, seq_id_dst, p0, p1)
Copy all tokens that belong to the specified sequence to another sequence Note that this does not allocate extra KV cache memory - it simply assigns the tokens to the new sequence p0 < 0 : [0, p1] p1 < 0 : [p0, inf)
Source code in llama_cpp/llama_cpp.py
llama_kv_cache_seq_keep(ctx, seq_id)
Removes all tokens that do not belong to the specified sequence
Source code in llama_cpp/llama_cpp.py
llama_kv_cache_seq_add(ctx, seq_id, p0, p1, delta)
Adds relative position "delta" to all tokens that belong to the specified sequence and have positions in [p0, p1) If the KV cache is RoPEd, the KV data is updated accordingly: - lazily on next llama_decode() - explicitly with llama_kv_cache_update() p0 < 0 : [0, p1] p1 < 0 : [p0, inf)
Source code in llama_cpp/llama_cpp.py
llama_kv_cache_seq_div(ctx, seq_id, p0, p1, d)
Integer division of the positions by factor of d > 1
If the KV cache is RoPEd, the KV data is updated accordingly
p0 < 0 : [0, p1]
p1 < 0 : [p0, inf)
Source code in llama_cpp/llama_cpp.py
llama_kv_cache_defrag(ctx)
Defragment the KV cache This will be applied: - lazily on next llama_decode() - explicitly with llama_kv_cache_update()
Source code in llama_cpp/llama_cpp.py
llama_kv_cache_update(ctx)
Apply the KV cache updates (such as K-shifts, defragmentation, etc.)
llama_state_get_size(ctx)
Returns the maximum size in bytes of the state (rng, logits, embedding and kv_cache) - will often be smaller after compacting tokens
Source code in llama_cpp/llama_cpp.py
llama_get_state_size(ctx)
Returns the maximum size in bytes of the state (rng, logits, embedding and kv_cache) - will often be smaller after compacting tokens
Source code in llama_cpp/llama_cpp.py
llama_state_get_data(ctx, dst)
Copies the state to the specified destination address. Destination needs to have allocated enough memory. Returns the number of bytes copied
Source code in llama_cpp/llama_cpp.py
llama_copy_state_data(ctx, dst)
Copies the state to the specified destination address. Destination needs to have allocated enough memory. Returns the number of bytes copied
Source code in llama_cpp/llama_cpp.py
llama_state_set_data(ctx, src)
Set the state reading from the specified address Returns the number of bytes read
Source code in llama_cpp/llama_cpp.py
llama_set_state_data(ctx, src)
Set the state reading from the specified address
Source code in llama_cpp/llama_cpp.py
llama_state_load_file(ctx, path_session, tokens_out, n_token_capacity, n_token_count_out)
Source code in llama_cpp/llama_cpp.py
llama_load_session_file(ctx, path_session, tokens_out, n_token_capacity, n_token_count_out)
Source code in llama_cpp/llama_cpp.py
llama_state_save_file(ctx, path_session, tokens, n_token_count)
Source code in llama_cpp/llama_cpp.py
llama_save_session_file(ctx, path_session, tokens, n_token_count)
Source code in llama_cpp/llama_cpp.py
llama_state_seq_get_size(ctx, seq_id)
Get the exact size needed to copy the KV cache of a single sequence
Source code in llama_cpp/llama_cpp.py
llama_state_seq_get_data(ctx, dst, seq_id)
Copy the KV cache of a single sequence into the specified buffer
Source code in llama_cpp/llama_cpp.py
llama_state_seq_set_data(ctx, src, dest_seq_id)
Copy the sequence data (originally copied with llama_state_seq_get_data
) into the specified sequence
Source code in llama_cpp/llama_cpp.py
llama_state_seq_save_file(ctx, filepath, seq_id, tokens, n_token_count)
Source code in llama_cpp/llama_cpp.py
llama_state_seq_load_file(ctx, filepath, dest_seq_id, tokens_out, n_token_capacity, n_token_count_out)
Source code in llama_cpp/llama_cpp.py
llama_batch_get_one(tokens, n_tokens, pos_0, seq_id)
Return batch for single sequence of tokens starting at pos_0
NOTE: this is a helper function to facilitate transition to the new batch API - avoid using it
Source code in llama_cpp/llama_cpp.py
llama_batch_init(n_tokens, embd, n_seq_max)
Allocates a batch of tokens on the heap that can hold a maximum of n_tokens Each token can be assigned up to n_seq_max sequence ids The batch has to be freed with llama_batch_free() If embd != 0, llama_batch.embd will be allocated with size of n_tokens * embd * sizeof(float) Otherwise, llama_batch.token will be allocated to store n_tokens llama_token The rest of the llama_batch members are allocated with size n_tokens All members are left uninitialized
Source code in llama_cpp/llama_cpp.py
llama_batch_free(batch)
llama_decode(ctx, batch)
Positive return values does not mean a fatal error, but rather a warning. 0 - success 1 - could not find a KV slot for the batch (try reducing the size of the batch or increase the context) < 0 - error
Source code in llama_cpp/llama_cpp.py
llama_set_n_threads(ctx, n_threads, n_threads_batch)
Set the number of threads used for decoding n_threads is the number of threads used for generation (single token) n_threads_batch is the number of threads used for prompt and batch processing (multiple tokens)
Source code in llama_cpp/llama_cpp.py
llama_set_causal_attn(ctx, causal_attn)
Set whether to use causal attention or not If set to true, the model will only attend to the past tokens
Source code in llama_cpp/llama_cpp.py
llama_set_abort_callback(ctx, abort_callback, abort_callback_data)
Set abort callback
Source code in llama_cpp/llama_cpp.py
llama_synchronize(ctx)
Wait until all computations are finished This is automatically done when using one of the functions below to obtain the computation results and is not necessary to call it explicitly in most cases
Source code in llama_cpp/llama_cpp.py
llama_get_logits(ctx)
Token logits obtained from the last call to llama_eval() The logits for the last token are stored in the last row Logits for which llama_batch.logits[i] == 0 are undefined Rows: n_tokens provided with llama_batch Cols: n_vocab
Returns:
-
CtypesArray[c_float]
–Pointer to the logits buffer of shape (n_tokens, n_vocab)
Source code in llama_cpp/llama_cpp.py
llama_get_logits_ith(ctx, i)
Logits for the ith token. Equivalent to: llama_get_logits(ctx) + i*n_vocab
Source code in llama_cpp/llama_cpp.py
llama_get_embeddings(ctx)
Get the embeddings for the input shape: [n_embd] (1-dimensional)
Source code in llama_cpp/llama_cpp.py
llama_get_embeddings_ith(ctx, i)
Get the embeddings for the ith sequence llama_get_embeddings(ctx) + i*n_embd
Source code in llama_cpp/llama_cpp.py
llama_get_embeddings_seq(ctx, seq_id)
Get the embeddings for a sequence id Returns NULL if pooling_type is LLAMA_POOLING_TYPE_NONE shape: [n_embd] (1-dimensional)
Source code in llama_cpp/llama_cpp.py
llama_token_get_text(model, token)
llama_token_get_score(model, token)
llama_token_get_type(model, token)
llama_token_is_eog(model, token)
Check if the token is supposed to end generation (end-of-generation, eg. EOS, EOT, etc.)
Source code in llama_cpp/llama_cpp.py
llama_token_bos(model)
llama_token_eos(model)
llama_token_cls(model)
llama_token_sep(model)
llama_token_nl(model)
llama_add_bos_token(model)
Returns -1 if unknown, 1 for true or 0 for false.
llama_add_eos_token(model)
Returns -1 if unknown, 1 for true or 0 for false.
llama_token_prefix(model)
llama_token_middle(model)
llama_token_suffix(model)
llama_token_eot(model)
llama_tokenize(model, text, text_len, tokens, n_tokens_max, add_special, parse_special)
Convert the provided text into tokens.
Parameters:
-
model
(llama_model_p
) –The model to use for tokenization.
-
text
(bytes
) –The text to tokenize.
-
text_len
(Union[c_int, int]
) –The length of the text.
-
tokens
(CtypesArray[llama_token]
) –The tokens pointer must be large enough to hold the resulting tokens.
-
n_max_tokens
–The maximum number of tokens to return.
-
add_special
(Union[c_bool, bool]
) –Allow tokenizing special and/or control tokens which otherwise are not exposed and treated as plaintext. Does not insert a leading space.
-
parse_special
(Union[c_bool, bool]
) –Allow parsing special tokens.
Returns:
-
int
–Returns the number of tokens on success, no more than n_tokens_max
-
int
–Returns a negative number on failure - the number of tokens that would have been returned
Source code in llama_cpp/llama_cpp.py
llama_token_to_piece(model, token, buf, length, special)
Token Id -> Piece. Uses the vocabulary in the provided context. Does not write null terminator to the buffer. User code is responsible to remove the leading whitespace of the first non-BOS token when decoding multiple tokens.
Parameters:
-
model
(llama_model_p
) –The model to use for tokenization.
-
token
(Union[llama_token, int]
) –The token to convert.
-
buf
(Union[c_char_p, bytes, CtypesArray[c_char]]
) –The buffer to write the token to.
-
length
(Union[c_int, int]
) –The length of the buffer.
-
special
(Union[c_bool, bool]
) –If true, special tokens are rendered in the output.
Source code in llama_cpp/llama_cpp.py
llama_chat_apply_template(model, tmpl, chat, n_msg)
Source code in llama_cpp/llama_cpp.py
llama_grammar_init(rules, n_rules, start_rule_index)
Initialize a grammar from a set of rules.
Source code in llama_cpp/llama_cpp.py
llama_grammar_free(grammar)
llama_grammar_copy(grammar)
llama_set_rng_seed(ctx, seed)
Sets the current rng seed.
llama_sample_repetition_penalties(ctx, candidates, last_tokens_data, penalty_last_n, penalty_repeat, penalty_freq, penalty_present)
Repetition penalty described in CTRL academic paper https://arxiv.org/abs/1909.05858, with negative logit fix. Frequency and presence penalties described in OpenAI API https://platform.openai.com/docs/api-reference/parameter-details.
Source code in llama_cpp/llama_cpp.py
llama_sample_apply_guidance(ctx, logits, logits_guidance, scale)
Apply classifier-free guidance to the logits as described in academic paper "Stay on topic with Classifier-Free Guidance" https://arxiv.org/abs/2306.17806
Source code in llama_cpp/llama_cpp.py
llama_sample_softmax(ctx, candidates)
Sorts candidate tokens by their logits in descending order and calculate probabilities based on logits.
Source code in llama_cpp/llama_cpp.py
llama_sample_top_k(ctx, candidates, k, min_keep)
Top-K sampling described in academic paper "The Curious Case of Neural Text Degeneration" https://arxiv.org/abs/1904.09751
Source code in llama_cpp/llama_cpp.py
llama_sample_top_p(ctx, candidates, p, min_keep)
Nucleus sampling described in academic paper "The Curious Case of Neural Text Degeneration" https://arxiv.org/abs/1904.09751
Source code in llama_cpp/llama_cpp.py
llama_sample_min_p(ctx, candidates, p, min_keep)
Minimum P sampling as described in https://github.com/ggerganov/llama.cpp/pull/3841
Source code in llama_cpp/llama_cpp.py
llama_sample_tail_free(ctx, candidates, z, min_keep)
Tail Free Sampling described in https://www.trentonbricken.com/Tail-Free-Sampling/.
Source code in llama_cpp/llama_cpp.py
llama_sample_typical(ctx, candidates, p, min_keep)
Locally Typical Sampling implementation described in the paper https://arxiv.org/abs/2202.00666.
Source code in llama_cpp/llama_cpp.py
llama_sample_entropy(ctx, candidates, min_temp, max_temp, exponent_val)
Dynamic temperature implementation described in the paper https://arxiv.org/abs/2309.02772.
Source code in llama_cpp/llama_cpp.py
llama_sample_temp(ctx, candidates, temp)
Temperature sampling described in academic paper "Generating Long Sequences with Sparse Transformers" https://arxiv.org/abs/1904.10509
Parameters:
-
candidates
(Union[CtypesArray[llama_token_data_array], CtypesPointerOrRef[llama_token_data_array]]
) –A vector of
llama_token_data
containing the candidate tokens, their probabilities (p), and log-odds (logit) for the current position in the generated text. -
temp
(Union[c_float, float]
) –The temperature value to use for the sampling. A higher value corresponds to more surprising or less predictable text, while a lower value corresponds to less surprising or more predictable text.
Source code in llama_cpp/llama_cpp.py
llama_sample_grammar(ctx, candidates, grammar)
Apply constraints from grammar
Parameters:
-
candidates
(Union[CtypesArray[llama_token_data_array], CtypesPointerOrRef[llama_token_data_array]]
) –A vector of
llama_token_data
containing the candidate tokens, their probabilities (p), and log-odds (logit) for the current position in the generated text. -
grammar
–A grammar object containing the rules and constraints to apply to the generated text.
Source code in llama_cpp/llama_cpp.py
llama_sample_token_mirostat(ctx, candidates, tau, eta, m, mu)
Mirostat 1.0 algorithm described in the paper https://arxiv.org/abs/2007.14966. Uses tokens instead of words.
Parameters:
-
candidates
(Union[CtypesArray[llama_token_data_array], CtypesPointerOrRef[llama_token_data_array]]
) –A vector of
llama_token_data
containing the candidate tokens, their probabilities (p), and log-odds (logit) for the current position in the generated text. -
tau
(Union[c_float, float]
) –The target cross-entropy (or surprise) value you want to achieve for the generated text. A higher value corresponds to more surprising or less predictable text, while a lower value corresponds to less surprising or more predictable text.
-
eta
(Union[c_float, float]
) –The learning rate used to update
mu
based on the error between the target and observed surprisal of the sampled word. A larger learning rate will causemu
to be updated more quickly, while a smaller learning rate will result in slower updates. -
m
(Union[c_int, int]
) –The number of tokens considered in the estimation of
s_hat
. This is an arbitrary value that is used to calculates_hat
, which in turn helps to calculate the value ofk
. In the paper, they usem = 100
, but you can experiment with different values to see how it affects the performance of the algorithm. -
mu
(CtypesPointerOrRef[c_float]
) –Maximum cross-entropy. This value is initialized to be twice the target cross-entropy (
2 * tau
) and is updated in the algorithm based on the error between the target and observed surprisal.
Source code in llama_cpp/llama_cpp.py
llama_sample_token_mirostat_v2(ctx, candidates, tau, eta, mu)
Mirostat 2.0 algorithm described in the paper https://arxiv.org/abs/2007.14966. Uses tokens instead of words.
Parameters:
-
candidates
(Union[CtypesArray[llama_token_data_array], CtypesPointerOrRef[llama_token_data_array]]
) –A vector of
llama_token_data
containing the candidate tokens, their probabilities (p), and log-odds (logit) for the current position in the generated text. -
tau
(Union[c_float, float]
) –The target cross-entropy (or surprise) value you want to achieve for the generated text. A higher value corresponds to more surprising or less predictable text, while a lower value corresponds to less surprising or more predictable text.
-
eta
(Union[c_float, float]
) –The learning rate used to update
mu
based on the error between the target and observed surprisal of the sampled word. A larger learning rate will causemu
to be updated more quickly, while a smaller learning rate will result in slower updates. -
mu
(CtypesPointerOrRef[c_float]
) –Maximum cross-entropy. This value is initialized to be twice the target cross-entropy (
2 * tau
) and is updated in the algorithm based on the error between the target and observed surprisal.
Source code in llama_cpp/llama_cpp.py
llama_sample_token_greedy(ctx, candidates)
Selects the token with the highest probability.
Source code in llama_cpp/llama_cpp.py
llama_sample_token(ctx, candidates)
Randomly selects a token from the candidates based on their probabilities.
Source code in llama_cpp/llama_cpp.py
llama_grammar_accept_token(ctx, grammar, token)
Accepts the sampled token into the grammar
Source code in llama_cpp/llama_cpp.py
llama_beam_view
llama_beams_state
Bases: Structure
Source code in llama_cpp/llama_cpp.py
llama_beam_search_callback_fn_t = ctypes.CFUNCTYPE(None, ctypes.c_void_p, llama_beams_state)
module-attribute
llama_beam_search(ctx, callback, callback_data, n_beams, n_past, n_predict)
Source code in llama_cpp/llama_cpp.py
llama_split_path(split_path, maxlen, path_prefix, split_no, split_count)
Build a split GGUF final path for this chunk.
Source code in llama_cpp/llama_cpp.py
llama_split_prefix(split_prefix, maxlen, split_path, split_no, split_count)
Extract the path prefix from the split_path if and only if the split_no and split_count match.
Source code in llama_cpp/llama_cpp.py
llama_get_timings(ctx)
llama_print_timings(ctx)
llama_reset_timings(ctx)
llama_print_system_info()
llama_log_set(log_callback, user_data)
Set callback for all future logging events.
If this is not called, or NULL is supplied, everything is output on stderr.
Source code in llama_cpp/llama_cpp.py
LLAMA_MAX_DEVICES = _lib.llama_max_devices()
module-attribute
LLAMA_DEFAULT_SEED = 4294967295
module-attribute
LLAMA_MAX_RNG_STATE = 64 * 1024
module-attribute
LLAMA_FILE_MAGIC_GGLA = 1734831201
module-attribute
LLAMA_FILE_MAGIC_GGSN = 1734833006
module-attribute
LLAMA_FILE_MAGIC_GGSQ = 1734833009
module-attribute
LLAMA_SESSION_MAGIC = LLAMA_FILE_MAGIC_GGSN
module-attribute
LLAMA_SESSION_VERSION = 5
module-attribute
LLAMA_STATE_SEQ_MAGIC = LLAMA_FILE_MAGIC_GGSQ
module-attribute
LLAMA_STATE_SEQ_VERSION = 1
module-attribute
LLAMA_VOCAB_TYPE_NONE = 0
module-attribute
For models without vocab
LLAMA_VOCAB_TYPE_SPM = 1
module-attribute
LLaMA tokenizer based on byte-level BPE with byte fallback
LLAMA_VOCAB_TYPE_BPE = 2
module-attribute
GPT-2 tokenizer based on byte-level BPE
LLAMA_VOCAB_TYPE_WPM = 3
module-attribute
BERT tokenizer based on WordPiece
LLAMA_ROPE_TYPE_NONE = -1
module-attribute
LLAMA_ROPE_TYPE_NORM = 0
module-attribute
LLAMA_ROPE_TYPE_NEOX = 2
module-attribute
LLAMA_ROPE_TYPE_GLM = 4
module-attribute
LLAMA_TOKEN_TYPE_UNDEFINED = 0
module-attribute
LLAMA_TOKEN_TYPE_NORMAL = 1
module-attribute
LLAMA_TOKEN_TYPE_UNKNOWN = 2
module-attribute
LLAMA_TOKEN_TYPE_CONTROL = 3
module-attribute
LLAMA_TOKEN_TYPE_USER_DEFINED = 4
module-attribute
LLAMA_TOKEN_TYPE_UNUSED = 5
module-attribute
LLAMA_TOKEN_TYPE_BYTE = 6
module-attribute
LLAMA_FTYPE_ALL_F32 = 0
module-attribute
LLAMA_FTYPE_MOSTLY_F16 = 1
module-attribute
LLAMA_FTYPE_MOSTLY_Q4_0 = 2
module-attribute
LLAMA_FTYPE_MOSTLY_Q4_1 = 3
module-attribute
LLAMA_FTYPE_MOSTLY_Q4_1_SOME_F16 = 4
module-attribute
LLAMA_FTYPE_MOSTLY_Q8_0 = 7
module-attribute
LLAMA_FTYPE_MOSTLY_Q5_0 = 8
module-attribute
LLAMA_FTYPE_MOSTLY_Q5_1 = 9
module-attribute
LLAMA_FTYPE_MOSTLY_Q2_K = 10
module-attribute
LLAMA_FTYPE_MOSTLY_Q3_K_S = 11
module-attribute
LLAMA_FTYPE_MOSTLY_Q3_K_M = 12
module-attribute
LLAMA_FTYPE_MOSTLY_Q3_K_L = 13
module-attribute
LLAMA_FTYPE_MOSTLY_Q4_K_S = 14
module-attribute
LLAMA_FTYPE_MOSTLY_Q4_K_M = 15
module-attribute
LLAMA_FTYPE_MOSTLY_Q5_K_S = 16
module-attribute
LLAMA_FTYPE_MOSTLY_Q5_K_M = 17
module-attribute
LLAMA_FTYPE_MOSTLY_Q6_K = 18
module-attribute
LLAMA_FTYPE_MOSTLY_IQ2_XXS = 19
module-attribute
LLAMA_FTYPE_MOSTLY_IQ2_XS = 20
module-attribute
LLAMA_FTYPE_MOSTLY_Q2_K_S = 21
module-attribute
LLAMA_FTYPE_MOSTLY_IQ3_XS = 22
module-attribute
LLAMA_FTYPE_MOSTLY_IQ3_XXS = 23
module-attribute
LLAMA_FTYPE_MOSTLY_IQ1_S = 24
module-attribute
LLAMA_FTYPE_MOSTLY_IQ4_NL = 25
module-attribute
LLAMA_FTYPE_MOSTLY_IQ3_S = 26
module-attribute
LLAMA_FTYPE_MOSTLY_IQ3_M = 27
module-attribute
LLAMA_FTYPE_MOSTLY_IQ2_S = 28
module-attribute
LLAMA_FTYPE_MOSTLY_IQ2_M = 29
module-attribute
LLAMA_FTYPE_MOSTLY_IQ4_XS = 30
module-attribute
LLAMA_FTYPE_GUESSED = 1024
module-attribute
LLAMA_ROPE_SCALING_TYPE_UNSPECIFIED = -1
module-attribute
LLAMA_ROPE_SCALING_TYPE_NONE = 0
module-attribute
LLAMA_ROPE_SCALING_TYPE_LINEAR = 1
module-attribute
LLAMA_ROPE_SCALING_TYPE_YARN = 2
module-attribute
LLAMA_ROPE_SCALING_TYPE_MAX_VALUE = LLAMA_ROPE_SCALING_TYPE_YARN
module-attribute
LLAMA_POOLING_TYPE_UNSPECIFIED = -1
module-attribute
LLAMA_POOLING_TYPE_NONE = 0
module-attribute
LLAMA_POOLING_TYPE_MEAN = 1
module-attribute
LLAMA_POOLING_TYPE_CLS = 2
module-attribute
LLAMA_SPLIT_MODE_NONE = 0
module-attribute
LLAMA_SPLIT_MODE_LAYER = 1
module-attribute
LLAMA_SPLIT_MODE_ROW = 2
module-attribute
LLAMA_KV_OVERRIDE_TYPE_INT = 0
module-attribute
LLAMA_KV_OVERRIDE_TYPE_FLOAT = 1
module-attribute
LLAMA_KV_OVERRIDE_TYPE_BOOL = 2
module-attribute
LLAMA_KV_OVERRIDE_TYPE_STR = 3
module-attribute
LLAMA_GRETYPE_END = 0
module-attribute
LLAMA_GRETYPE_ALT = 1
module-attribute
LLAMA_GRETYPE_RULE_REF = 2
module-attribute
LLAMA_GRETYPE_CHAR = 3
module-attribute
LLAMA_GRETYPE_CHAR_NOT = 4
module-attribute
LLAMA_GRETYPE_CHAR_RNG_UPPER = 5
module-attribute
LLAMA_GRETYPE_CHAR_ALT = 6
module-attribute
Misc
llama_cpp.llama_types
Types and request signatures for OpenAI compatibility
NOTE: These types may change to match the OpenAI OpenAPI specification.
Based on the OpenAI OpenAPI specification: https://github.com/openai/openai-openapi/blob/master/openapi.yaml
JsonType = Union[None, int, str, bool, List[Any], Dict[str, Any]]
module-attribute
EmbeddingUsage
Embedding
CreateEmbeddingResponse
Bases: TypedDict
Source code in llama_cpp/llama_types.py
object: Literal['list']
instance-attribute
model: str
instance-attribute
data: List[Embedding]
instance-attribute
usage: EmbeddingUsage
instance-attribute
CompletionLogprobs
Bases: TypedDict
Source code in llama_cpp/llama_types.py
text_offset: List[int]
instance-attribute
token_logprobs: List[Optional[float]]
instance-attribute
tokens: List[str]
instance-attribute
top_logprobs: List[Optional[Dict[str, float]]]
instance-attribute
CompletionChoice
Bases: TypedDict
Source code in llama_cpp/llama_types.py
text: str
instance-attribute
index: int
instance-attribute
logprobs: Optional[CompletionLogprobs]
instance-attribute
finish_reason: Optional[Literal['stop', 'length']]
instance-attribute
CompletionUsage
CreateCompletionResponse
Bases: TypedDict
Source code in llama_cpp/llama_types.py
id: str
instance-attribute
object: Literal['text_completion']
instance-attribute
created: int
instance-attribute
model: str
instance-attribute
choices: List[CompletionChoice]
instance-attribute
usage: NotRequired[CompletionUsage]
instance-attribute
ChatCompletionResponseFunctionCall
ChatCompletionResponseMessage
Bases: TypedDict
Source code in llama_cpp/llama_types.py
content: Optional[str]
instance-attribute
tool_calls: NotRequired[ChatCompletionMessageToolCalls]
instance-attribute
role: Literal['assistant', 'function']
instance-attribute
function_call: NotRequired[ChatCompletionResponseFunctionCall]
instance-attribute
ChatCompletionFunction
Bases: TypedDict
Source code in llama_cpp/llama_types.py
name: str
instance-attribute
description: NotRequired[str]
instance-attribute
parameters: Dict[str, JsonType]
instance-attribute
ChatCompletionResponseChoice
Bases: TypedDict
Source code in llama_cpp/llama_types.py
index: int
instance-attribute
message: ChatCompletionResponseMessage
instance-attribute
logprobs: Optional[CompletionLogprobs]
instance-attribute
finish_reason: Optional[str]
instance-attribute
CreateChatCompletionResponse
Bases: TypedDict
Source code in llama_cpp/llama_types.py
id: str
instance-attribute
object: Literal['chat.completion']
instance-attribute
created: int
instance-attribute
model: str
instance-attribute
choices: List[ChatCompletionResponseChoice]
instance-attribute
usage: CompletionUsage
instance-attribute
ChatCompletionMessageToolCallChunkFunction
ChatCompletionMessageToolCallChunk
Bases: TypedDict
Source code in llama_cpp/llama_types.py
index: int
instance-attribute
id: NotRequired[str]
instance-attribute
type: Literal['function']
instance-attribute
function: ChatCompletionMessageToolCallChunkFunction
instance-attribute
ChatCompletionStreamResponseDeltaEmpty
ChatCompletionStreamResponseDeltaFunctionCall
ChatCompletionStreamResponseDelta
Bases: TypedDict
Source code in llama_cpp/llama_types.py
content: NotRequired[Optional[str]]
instance-attribute
function_call: NotRequired[Optional[ChatCompletionStreamResponseDeltaFunctionCall]]
instance-attribute
tool_calls: NotRequired[Optional[List[ChatCompletionMessageToolCallChunk]]]
instance-attribute
role: NotRequired[Optional[Literal['system', 'user', 'assistant', 'tool']]]
instance-attribute
ChatCompletionStreamResponseChoice
Bases: TypedDict
Source code in llama_cpp/llama_types.py
index: int
instance-attribute
delta: Union[ChatCompletionStreamResponseDelta, ChatCompletionStreamResponseDeltaEmpty]
instance-attribute
finish_reason: Optional[Literal['stop', 'length', 'tool_calls', 'function_call']]
instance-attribute
logprobs: NotRequired[Optional[CompletionLogprobs]]
instance-attribute
CreateChatCompletionStreamResponse
Bases: TypedDict
Source code in llama_cpp/llama_types.py
id: str
instance-attribute
model: str
instance-attribute
object: Literal['chat.completion.chunk']
instance-attribute
created: int
instance-attribute
choices: List[ChatCompletionStreamResponseChoice]
instance-attribute
ChatCompletionFunctions
Bases: TypedDict
Source code in llama_cpp/llama_types.py
name: str
instance-attribute
description: NotRequired[str]
instance-attribute
parameters: Dict[str, JsonType]
instance-attribute
ChatCompletionFunctionCallOption
ChatCompletionRequestResponseFormat
Bases: TypedDict
Source code in llama_cpp/llama_types.py
type: Literal['text', 'json_object']
instance-attribute
schema: NotRequired[JsonType]
instance-attribute
ChatCompletionRequestMessageContentPartText
ChatCompletionRequestMessageContentPartImageImageUrl
ChatCompletionRequestMessageContentPartImage
Bases: TypedDict
Source code in llama_cpp/llama_types.py
type: Literal['image_url']
instance-attribute
image_url: Union[str, ChatCompletionRequestMessageContentPartImageImageUrl]
instance-attribute
ChatCompletionRequestMessageContentPart = Union[ChatCompletionRequestMessageContentPartText, ChatCompletionRequestMessageContentPartImage]
module-attribute
ChatCompletionRequestSystemMessage
ChatCompletionRequestUserMessage
Bases: TypedDict
Source code in llama_cpp/llama_types.py
role: Literal['user']
instance-attribute
content: Optional[Union[str, List[ChatCompletionRequestMessageContentPart]]]
instance-attribute
ChatCompletionMessageToolCallFunction
ChatCompletionMessageToolCall
Bases: TypedDict
Source code in llama_cpp/llama_types.py
id: str
instance-attribute
type: Literal['function']
instance-attribute
function: ChatCompletionMessageToolCallFunction
instance-attribute
ChatCompletionMessageToolCalls = List[ChatCompletionMessageToolCall]
module-attribute
ChatCompletionRequestAssistantMessageFunctionCall
ChatCompletionRequestAssistantMessage
Bases: TypedDict
Source code in llama_cpp/llama_types.py
role: Literal['assistant']
instance-attribute
content: Optional[str]
instance-attribute
tool_calls: NotRequired[ChatCompletionMessageToolCalls]
instance-attribute
function_call: NotRequired[ChatCompletionRequestAssistantMessageFunctionCall]
instance-attribute
ChatCompletionRequestToolMessage
Bases: TypedDict
Source code in llama_cpp/llama_types.py
role: Literal['tool']
instance-attribute
content: Optional[str]
instance-attribute
tool_call_id: str
instance-attribute
ChatCompletionRequestFunctionMessage
Bases: TypedDict
Source code in llama_cpp/llama_types.py
role: Literal['function']
instance-attribute
content: Optional[str]
instance-attribute
name: str
instance-attribute
ChatCompletionRequestMessage = Union[ChatCompletionRequestSystemMessage, ChatCompletionRequestUserMessage, ChatCompletionRequestAssistantMessage, ChatCompletionRequestUserMessage, ChatCompletionRequestToolMessage, ChatCompletionRequestFunctionMessage]
module-attribute
ChatCompletionRequestFunctionCallOption
ChatCompletionRequestFunctionCall = Union[Literal['none', 'auto'], ChatCompletionRequestFunctionCallOption]
module-attribute
ChatCompletionFunctionParameters = Dict[str, JsonType]
module-attribute
ChatCompletionToolFunction
Bases: TypedDict
Source code in llama_cpp/llama_types.py
name: str
instance-attribute
description: NotRequired[str]
instance-attribute
parameters: ChatCompletionFunctionParameters
instance-attribute
ChatCompletionTool
ChatCompletionNamedToolChoiceFunction
ChatCompletionNamedToolChoice
Bases: TypedDict