Skip to content
Navigation Menu
{{ message }}
forked from abetlen/llama-cpp-python
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathllama_cpp.py
More file actions
2534 lines (1989 loc) · 89.6 KB
/
Copy pathllama_cpp.py
File metadata and controls
2534 lines (1989 loc) · 89.6 KB
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
import sys
import os
import ctypes
from ctypes import (
c_bool,
c_char_p,
c_int,
c_int8,
c_int32,
c_uint8,
c_uint32,
c_int64,
c_size_t,
c_float,
c_double,
c_void_p,
POINTER,
_Pointer, # type: ignore
Structure,
Union as CtypesUnion,
Array,
)
import pathlib
from typing import List, Union
# Load the library
def _load_shared_library(lib_base_name: str):
# Construct the paths to the possible shared library names
_base_path = pathlib.Path(os.path.abspath(os.path.dirname(__file__)))
# Searching for the library in the current directory under the name "libllama" (default name
# for llamacpp) and "llama" (default name for this repo)
_lib_paths: List[pathlib.Path] = []
# Determine the file extension based on the platform
if sys.platform.startswith("linux"):
_lib_paths += [
_base_path / f"lib{lib_base_name}.so",
]
elif sys.platform == "darwin":
_lib_paths += [
_base_path / f"lib{lib_base_name}.so",
_base_path / f"lib{lib_base_name}.dylib",
]
elif sys.platform == "win32":
_lib_paths += [
_base_path / f"{lib_base_name}.dll",
_base_path / f"lib{lib_base_name}.dll",
]
else:
raise RuntimeError("Unsupported platform")
if "LLAMA_CPP_LIB" in os.environ:
lib_base_name = os.environ["LLAMA_CPP_LIB"]
_lib = pathlib.Path(lib_base_name)
_base_path = _lib.parent.resolve()
_lib_paths = [_lib.resolve()]
cdll_args = dict() # type: ignore
# Add the library directory to the DLL search path on Windows (if needed)
if sys.platform == "win32" and sys.version_info >= (3, 8):
os.add_dll_directory(str(_base_path))
if "CUDA_PATH" in os.environ:
os.add_dll_directory(os.path.join(os.environ["CUDA_PATH"], "bin"))
os.add_dll_directory(os.path.join(os.environ["CUDA_PATH"], "lib"))
if "HIP_PATH" in os.environ:
os.add_dll_directory(os.path.join(os.environ["HIP_PATH"], "bin"))
os.add_dll_directory(os.path.join(os.environ["HIP_PATH"], "lib"))
cdll_args["winmode"] = ctypes.RTLD_GLOBAL
# Try to load the shared library, handling potential errors
for _lib_path in _lib_paths:
if _lib_path.exists():
try:
return ctypes.CDLL(str(_lib_path), **cdll_args)
except Exception as e:
raise RuntimeError(f"Failed to load shared library '{_lib_path}': {e}")
raise FileNotFoundError(
f"Shared library with base name '{lib_base_name}' not found"
)
# Specify the base name of the shared library to load
_lib_base_name = "llama"
# Load the library
_lib = _load_shared_library(_lib_base_name)
# Misc
c_float_p = POINTER(c_float)
c_uint8_p = POINTER(c_uint8)
c_size_t_p = POINTER(c_size_t)
# llama.h bindings
_lib.llama_max_devices.argtypes = []
_lib.llama_max_devices.restype = ctypes.c_int32
LLAMA_MAX_DEVICES = _lib.llama_max_devices()
# define LLAMA_DEFAULT_SEED 0xFFFFFFFF
LLAMA_DEFAULT_SEED = 0xFFFFFFFF
# define LLAMA_MAX_RNG_STATE (64*1024)
LLAMA_MAX_RNG_STATE = 64 * 1024
# define LLAMA_FILE_MAGIC_GGLA 0x67676c61u // 'ggla'
LLAMA_FILE_MAGIC_GGLA = 0x67676C61
# define LLAMA_FILE_MAGIC_GGSN 0x6767736eu // 'ggsn'
LLAMA_FILE_MAGIC_GGSN = 0x6767736E
# define LLAMA_SESSION_MAGIC LLAMA_FILE_MAGIC_GGSN
LLAMA_SESSION_MAGIC = LLAMA_FILE_MAGIC_GGSN
# define LLAMA_SESSION_VERSION 4
LLAMA_SESSION_VERSION = 4
# struct llama_model;
llama_model_p = c_void_p
# struct llama_context;
llama_context_p = c_void_p
# typedef int32_t llama_pos;
llama_pos = c_int32
# typedef int32_t llama_token;
llama_token = c_int32
llama_token_p = POINTER(llama_token)
# typedef int32_t llama_seq_id;
llama_seq_id = c_int32
# enum llama_vocab_type {
# LLAMA_VOCAB_TYPE_SPM = 0, // SentencePiece
# LLAMA_VOCAB_TYPE_BPE = 1, // Byte Pair Encoding
# };
LLAMA_VOCAB_TYPE_SPM = 0
LLAMA_VOCAB_TYPE_BPE = 1
# enum llama_token_type {
# LLAMA_TOKEN_TYPE_UNDEFINED = 0,
# LLAMA_TOKEN_TYPE_NORMAL = 1,
# LLAMA_TOKEN_TYPE_UNKNOWN = 2,
# LLAMA_TOKEN_TYPE_CONTROL = 3,
# LLAMA_TOKEN_TYPE_USER_DEFINED = 4,
# LLAMA_TOKEN_TYPE_UNUSED = 5,
# LLAMA_TOKEN_TYPE_BYTE = 6,
# };
LLAMA_TOKEN_TYPE_UNDEFINED = 0
LLAMA_TOKEN_TYPE_NORMAL = 1
LLAMA_TOKEN_TYPE_UNKNOWN = 2
LLAMA_TOKEN_TYPE_CONTROL = 3
LLAMA_TOKEN_TYPE_USER_DEFINED = 4
LLAMA_TOKEN_TYPE_UNUSED = 5
LLAMA_TOKEN_TYPE_BYTE = 6
# // model file types
# enum llama_ftype {
# LLAMA_FTYPE_ALL_F32 = 0,
# LLAMA_FTYPE_MOSTLY_F16 = 1, // except 1d tensors
# LLAMA_FTYPE_MOSTLY_Q4_0 = 2, // except 1d tensors
# LLAMA_FTYPE_MOSTLY_Q4_1 = 3, // except 1d tensors
# LLAMA_FTYPE_MOSTLY_Q4_1_SOME_F16 = 4, // tok_embeddings.weight and output.weight are F16
# // LLAMA_FTYPE_MOSTLY_Q4_2 = 5, // support has been removed
# // LLAMA_FTYPE_MOSTLY_Q4_3 = 6, // support has been removed
# LLAMA_FTYPE_MOSTLY_Q8_0 = 7, // except 1d tensors
# LLAMA_FTYPE_MOSTLY_Q5_0 = 8, // except 1d tensors
# LLAMA_FTYPE_MOSTLY_Q5_1 = 9, // except 1d tensors
# LLAMA_FTYPE_MOSTLY_Q2_K = 10, // except 1d tensors
# LLAMA_FTYPE_MOSTLY_Q3_K_S = 11, // except 1d tensors
# LLAMA_FTYPE_MOSTLY_Q3_K_M = 12, // except 1d tensors
# LLAMA_FTYPE_MOSTLY_Q3_K_L = 13, // except 1d tensors
# LLAMA_FTYPE_MOSTLY_Q4_K_S = 14, // except 1d tensors
# LLAMA_FTYPE_MOSTLY_Q4_K_M = 15, // except 1d tensors
# LLAMA_FTYPE_MOSTLY_Q5_K_S = 16, // except 1d tensors
# LLAMA_FTYPE_MOSTLY_Q5_K_M = 17, // except 1d tensors
# LLAMA_FTYPE_MOSTLY_Q6_K = 18, // except 1d tensors
# LLAMA_FTYPE_MOSTLY_IQ2_XXS = 19, // except 1d tensors
# LLAMA_FTYPE_MOSTLY_IQ2_XS = 20, // except 1d tensors
# LLAMA_FTYPE_MOSTLY_Q2_K_S = 21, // except 1d tensors
# LLAMA_FTYPE_GUESSED = 1024, // not specified in the model file
# };
LLAMA_FTYPE_ALL_F32 = 0
LLAMA_FTYPE_MOSTLY_F16 = 1
LLAMA_FTYPE_MOSTLY_Q4_0 = 2
LLAMA_FTYPE_MOSTLY_Q4_1 = 3
LLAMA_FTYPE_MOSTLY_Q4_1_SOME_F16 = 4
LLAMA_FTYPE_MOSTLY_Q8_0 = 7
LLAMA_FTYPE_MOSTLY_Q5_0 = 8
LLAMA_FTYPE_MOSTLY_Q5_1 = 9
LLAMA_FTYPE_MOSTLY_Q2_K = 10
LLAMA_FTYPE_MOSTLY_Q3_K_S = 11
LLAMA_FTYPE_MOSTLY_Q3_K_M = 12
LLAMA_FTYPE_MOSTLY_Q3_K_L = 13
LLAMA_FTYPE_MOSTLY_Q4_K_S = 14
LLAMA_FTYPE_MOSTLY_Q4_K_M = 15
LLAMA_FTYPE_MOSTLY_Q5_K_S = 16
LLAMA_FTYPE_MOSTLY_Q5_K_M = 17
LLAMA_FTYPE_MOSTLY_Q6_K = 18
LLAMA_FTYPE_MOSTLY_IQ2_XXS = 19
LLAMA_FTYPE_MOSTLY_IQ2_XS = 20
LLAMA_FTYPE_MOSTLY_Q2_K_S = 21
LLAMA_FTYPE_GUESSED = 1024
# enum llama_rope_scaling_type {
# LLAMA_ROPE_SCALING_UNSPECIFIED = -1,
# LLAMA_ROPE_SCALING_NONE = 0,
# LLAMA_ROPE_SCALING_LINEAR = 1,
# LLAMA_ROPE_SCALING_YARN = 2,
# LLAMA_ROPE_SCALING_MAX_VALUE = LLAMA_ROPE_SCALING_YARN,
# };
LLAMA_ROPE_SCALING_UNSPECIFIED = -1
LLAMA_ROPE_SCALING_NONE = 0
LLAMA_ROPE_SCALING_LINEAR = 1
LLAMA_ROPE_SCALING_YARN = 2
LLAMA_ROPE_SCALING_MAX_VALUE = LLAMA_ROPE_SCALING_YARN
# enum llama_split_mode {
# LLAMA_SPLIT_NONE = 0, // single GPU
# LLAMA_SPLIT_LAYER = 1, // split layers and KV across GPUs
# LLAMA_SPLIT_ROW = 2, // split rows across GPUs
# };
LLAMA_SPLIT_NONE = 0
LLAMA_SPLIT_LAYER = 1
LLAMA_SPLIT_ROW = 2
# typedef struct llama_token_data {
# llama_token id; // token id
# float logit; // log-odds of the token
# float p; // probability of the token
# } llama_token_data;
class llama_token_data(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"""
_fields_ = [
("id", llama_token),
("logit", c_float),
("p", c_float),
]
llama_token_data_p = POINTER(llama_token_data)
# typedef struct llama_token_data_array {
# llama_token_data * data;
# size_t size;
# bool sorted;
# } llama_token_data_array;
class llama_token_data_array(Structure):
"""Used to sample tokens given logits
Attributes:
data (ctypes.Array[llama_token_data]): token data
size (int): size of the array
sorted (bool): whether the array is sorted"""
_fields_ = [
("data", llama_token_data_p),
("size", c_size_t),
("sorted", c_bool),
]
llama_token_data_array_p = POINTER(llama_token_data_array)
# typedef bool (*llama_progress_callback)(float progress, void *ctx);
llama_progress_callback = ctypes.CFUNCTYPE(c_bool, c_float, c_void_p)
# // 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
# //
# // - token : the token ids of the input (used when embd is NULL)
# // - embd : token embeddings (i.e. float vector of size n_embd) (used when token is NULL)
# // - pos : the positions of the respective token in the sequence
# // - seq_id : the sequence to which the respective token belongs
# // - logits : if zero, the logits for the respective token will not be output
# //
# typedef struct llama_batch {
# int32_t n_tokens;
# llama_token * token;
# float * embd;
# llama_pos * pos;
# int32_t * n_seq_id;
# llama_seq_id ** seq_id;
# int8_t * logits;
# // NOTE: helpers for smooth API transition - can be deprecated in the future
# // for future-proof code, use the above fields instead and ignore everything below
# //
# // pos[i] = all_pos_0 + i*all_pos_1
# //
# llama_pos all_pos_0; // used if pos == NULL
# llama_pos all_pos_1; // used if pos == NULL
# llama_seq_id all_seq_id; // used if seq_id == NULL
# } llama_batch;
class llama_batch(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:
token (ctypes.Array[llama_token]): the token ids of the input (used when embd is NULL)
embd (ctypes.Array[ctypes.c_float]): token embeddings (i.e. float vector of size n_embd) (used when token is NULL)
pos (ctypes.Array[ctypes.Array[llama_pos]]): the positions of the respective token in the sequence
seq_id (ctypes.Array[ctypes.Array[llama_seq_id]]): the sequence to which the respective token belongs
"""
_fields_ = [
("n_tokens", c_int32),
("token", POINTER(llama_token)),
("embd", c_float_p),
("pos", POINTER(llama_pos)),
("n_seq_id", POINTER(c_int32)),
("seq_id", POINTER(POINTER(llama_seq_id))),
("logits", POINTER(c_int8)),
("all_pos_0", llama_pos),
("all_pos_1", llama_pos),
("all_seq_id", llama_seq_id),
]
# enum llama_model_kv_override_type {
# LLAMA_KV_OVERRIDE_INT,
# LLAMA_KV_OVERRIDE_FLOAT,
# LLAMA_KV_OVERRIDE_BOOL,
# };
LLAMA_KV_OVERRIDE_INT = 0
LLAMA_KV_OVERRIDE_FLOAT = 1
LLAMA_KV_OVERRIDE_BOOL = 2
# struct llama_model_kv_override {
# char key[128];
# enum llama_model_kv_override_type tag;
# union {
# int64_t int_value;
# double float_value;
# bool bool_value;
# };
# };
class llama_model_kv_override_value(CtypesUnion):
_fields_ = [
("int_value", c_int64),
("float_value", c_double),
("bool_value", c_bool),
]
class llama_model_kv_override(Structure):
_fields_ = [
("key", ctypes.c_char * 128),
("tag", c_int),
("value", llama_model_kv_override_value),
]
# struct llama_model_params {
# int32_t n_gpu_layers; // number of layers to store in VRAM
# enum llama_split_mode split_mode; // how to split the model across multiple GPUs
# // 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
# int32_t main_gpu;
# // proportion of the model (layers or rows) to offload to each GPU, size: LLAMA_MAX_DEVICES
# const float * tensor_split;
# // 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.
# llama_progress_callback progress_callback;
# // context pointer passed to the progress callback
# void * progress_callback_user_data;
# // override key-value pairs of the model meta data
# const struct llama_model_kv_override * kv_overrides;
# // Keep the booleans together to avoid misalignment during copy-by-value.
# bool vocab_only; // only load the vocabulary, no weights
# bool use_mmap; // use mmap if possible
# bool use_mlock; // force system to keep model in RAM
# };
class llama_model_params(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 (ctypes.Array[ctypes.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 (ctypes.c_void_p): context pointer passed to the progress callback
kv_overrides (ctypes.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"""
_fields_ = [
("n_gpu_layers", c_int32),
("split_mode", c_int),
("main_gpu", c_int32),
("tensor_split", c_float_p),
("progress_callback", llama_progress_callback),
("progress_callback_user_data", c_void_p),
("kv_overrides", POINTER(llama_model_kv_override)),
("vocab_only", c_bool),
("use_mmap", c_bool),
("use_mlock", c_bool),
]
# struct llama_context_params {
# uint32_t seed; // RNG seed, -1 for random
# uint32_t n_ctx; // text context, 0 = from model
# uint32_t n_batch; // prompt processing maximum batch size
# uint32_t n_threads; // number of threads to use for generation
# uint32_t n_threads_batch; // number of threads to use for batch processing
# int8_t rope_scaling_type; // RoPE scaling type, from `enum llama_rope_scaling_type`
# // ref: https://github.com/ggerganov/llama.cpp/pull/2054
# float rope_freq_base; // RoPE base frequency, 0 = from model
# float rope_freq_scale; // RoPE frequency scaling factor, 0 = from model
# float yarn_ext_factor; // YaRN extrapolation mix factor, negative = from model
# float yarn_attn_factor; // YaRN magnitude scaling factor
# float yarn_beta_fast; // YaRN low correction dim
# float yarn_beta_slow; // YaRN high correction dim
# uint32_t yarn_orig_ctx; // YaRN original context size
# enum ggml_type type_k; // data type for K cache
# enum ggml_type type_v; // data type for V cache
# // Keep the booleans together to avoid misalignment during copy-by-value.
# bool mul_mat_q; // if true, use experimental mul_mat_q kernels (DEPRECATED - always true)
# bool logits_all; // the llama_eval() call computes all logits, not just the last one (DEPRECATED - set llama_batch.logits instead)
# bool embedding; // embedding mode only
# bool offload_kqv; // whether to offload the KQV ops (including the KV cache) to GPU
# };
class llama_context_params(Structure):
"""Parameters for llama_context
Attributes:
seed (int): RNG seed, -1 for random
n_ctx (int): text context, 0 = from model
n_batch (int): prompt processing maximum batch size
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`
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
type_k (int): data type for K cache
type_v (int): data type for V cache
mul_mat_q (bool): if true, use experimental mul_mat_q kernels (DEPRECATED - always true)
logits_all (bool): the llama_eval() call computes all logits, not just the last one (DEPRECATED - set llama_batch.logits instead)
embedding (bool): embedding mode only
offload_kqv (bool): whether to offload the KQV ops (including the KV cache) to GPU
"""
_fields_ = [
("seed", c_uint32),
("n_ctx", c_uint32),
("n_batch", c_uint32),
("n_threads", c_uint32),
("n_threads_batch", c_uint32),
("rope_scaling_type", c_int8),
("rope_freq_base", c_float),
("rope_freq_scale", c_float),
("yarn_ext_factor", c_float),
("yarn_attn_factor", c_float),
("yarn_beta_fast", c_float),
("yarn_beta_slow", c_float),
("yarn_orig_ctx", c_uint32),
("type_k", c_int),
("type_v", c_int),
("mul_mat_q", c_bool),
("logits_all", c_bool),
("embedding", c_bool),
("offload_kqv", c_bool),
]
# // 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 '\n' and strip it
# // if it exists.
# // It might not exist for progress report where '.' is output repeatedly.
# typedef void (*llama_log_callback)(enum llama_log_level level, const char * text, void * user_data);
llama_log_callback = ctypes.CFUNCTYPE(None, c_int, c_char_p, c_void_p)
"""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 '\n' and strip it
if it exists.
It might not exist for progress report where '.' is output repeatedly."""
# // model quantization parameters
# typedef struct llama_model_quantize_params {
# int32_t nthread; // number of threads to use for quantizing, if <=0 will use std::thread::hardware_concurrency()
# enum llama_ftype ftype; // quantize to this llama_ftype
# bool allow_requantize; // allow quantizing non-f32/f16 tensors
# bool quantize_output_tensor; // quantize output.weight
# bool only_copy; // only copy tensors - ftype, allow_requantize and quantize_output_tensor are ignored
# bool pure; // disable k-quant mixtures and quantize all tensors to the same type
# void * imatrix; // pointer to importance matrix data
# } llama_model_quantize_params;
class llama_model_quantize_params(Structure):
"""Parameters for llama_model_quantize
Attributes:
nthread (int): number of threads to use for quantizing, if <=0 will use std::thread::hardware_concurrency()
ftype (int): quantize to this llama_ftype
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): disable k-quant mixtures and quantize all tensors to the same type
imatrix (ctypes.c_void_p): pointer to importance matrix data
"""
_fields_ = [
("nthread", c_int32),
("ftype", c_int),
("allow_requantize", c_bool),
("quantize_output_tensor", c_bool),
("only_copy", c_bool),
("pure", c_bool),
("imatrix", c_void_p),
]
# // grammar types
# struct llama_grammar;
llama_grammar_p = c_void_p
# // grammar element type
# enum llama_gretype {
# // end of rule definition
# LLAMA_GRETYPE_END = 0,
# // start of alternate definition for rule
# LLAMA_GRETYPE_ALT = 1,
# // non-terminal element: reference to rule
# LLAMA_GRETYPE_RULE_REF = 2,
# // terminal element: character (code point)
# LLAMA_GRETYPE_CHAR = 3,
# // inverse char(s) ([^a], [^a-b] [^abc])
# LLAMA_GRETYPE_CHAR_NOT = 4,
# // modifies a preceding LLAMA_GRETYPE_CHAR or LLAMA_GRETYPE_CHAR_ALT to
# // be an inclusive range ([a-z])
# LLAMA_GRETYPE_CHAR_RNG_UPPER = 5,
# // modifies a preceding LLAMA_GRETYPE_CHAR or
# // LLAMA_GRETYPE_CHAR_RNG_UPPER to add an alternate char to match ([ab], [a-zA])
# LLAMA_GRETYPE_CHAR_ALT = 6,
# };
LLAMA_GRETYPE_END = 0
LLAMA_GRETYPE_ALT = 1
LLAMA_GRETYPE_RULE_REF = 2
LLAMA_GRETYPE_CHAR = 3
LLAMA_GRETYPE_CHAR_NOT = 4
LLAMA_GRETYPE_CHAR_RNG_UPPER = 5
LLAMA_GRETYPE_CHAR_ALT = 6
# typedef struct llama_grammar_element {
# enum llama_gretype type;
# uint32_t value; // Unicode code point or rule ID
# } llama_grammar_element;
class llama_grammar_element(Structure):
_fields_ = [
("type", c_int),
("value", c_uint32),
]
llama_grammar_element_p = POINTER(llama_grammar_element)
# // performance timing information
# struct llama_timings {
# double t_start_ms;
# double t_end_ms;
# double t_load_ms;
# double t_sample_ms;
# double t_p_eval_ms;
# double t_eval_ms;
# int32_t n_sample;
# int32_t n_p_eval;
# int32_t n_eval;
# };
class llama_timings(Structure):
_fields_ = [
("t_start_ms", c_double),
("t_end_ms", c_double),
("t_load_ms", c_double),
("t_sample_ms", c_double),
("t_p_eval_ms", c_double),
("t_eval_ms", c_double),
("n_sample", c_int32),
("n_p_eval", c_int32),
("n_eval", c_int32),
]
# // Helpers for getting default parameters
# LLAMA_API struct llama_model_params llama_model_default_params(void);
def llama_model_default_params() -> llama_model_params:
"""Get default parameters for llama_model"""
return _lib.llama_model_default_params()
_lib.llama_model_default_params.argtypes = []
_lib.llama_model_default_params.restype = llama_model_params
# LLAMA_API struct llama_context_params llama_context_default_params(void);
def llama_context_default_params() -> llama_context_params:
"""Get default parameters for llama_context"""
return _lib.llama_context_default_params()
_lib.llama_context_default_params.argtypes = []
_lib.llama_context_default_params.restype = llama_context_params
# LLAMA_API struct llama_model_quantize_params llama_model_quantize_default_params(void);
def llama_model_quantize_default_params() -> llama_model_quantize_params:
"""Get default parameters for llama_model_quantize"""
return _lib.llama_model_quantize_default_params()
_lib.llama_model_quantize_default_params.argtypes = []
_lib.llama_model_quantize_default_params.restype = llama_model_quantize_params
# // Initialize the llama + ggml backend
# // If numa is true, use NUMA optimizations
# // Call once at the start of the program
# LLAMA_API void llama_backend_init(bool numa);
def llama_backend_init(numa: Union[c_bool, bool]):
"""Initialize the llama + ggml backend
If numa is true, use NUMA optimizations
Call once at the start of the program"""
return _lib.llama_backend_init(numa)
_lib.llama_backend_init.argtypes = [c_bool]
_lib.llama_backend_init.restype = None
# // Call once at the end of the program - currently only used for MPI
# LLAMA_API void llama_backend_free(void);
def llama_backend_free():
"""Call once at the end of the program - currently only used for MPI"""
return _lib.llama_backend_free()
_lib.llama_backend_free.argtypes = []
_lib.llama_backend_free.restype = None
# LLAMA_API struct llama_model * llama_load_model_from_file(
# const char * path_model,
# struct llama_model_params params);
def llama_load_model_from_file(
path_model: bytes, params: llama_model_params
) -> llama_model_p:
return _lib.llama_load_model_from_file(path_model, params)
_lib.llama_load_model_from_file.argtypes = [c_char_p, llama_model_params]
_lib.llama_load_model_from_file.restype = llama_model_p
# LLAMA_API void llama_free_model(struct llama_model * model);
def llama_free_model(model: llama_model_p):
return _lib.llama_free_model(model)
_lib.llama_free_model.argtypes = [llama_model_p]
_lib.llama_free_model.restype = None
# LLAMA_API struct llama_context * llama_new_context_with_model(
# struct llama_model * model,
# struct llama_context_params params);
def llama_new_context_with_model(
model: llama_model_p, params: llama_context_params
) -> llama_context_p:
return _lib.llama_new_context_with_model(model, params)
_lib.llama_new_context_with_model.argtypes = [llama_model_p, llama_context_params]
_lib.llama_new_context_with_model.restype = llama_context_p
# // Frees all allocated memory
# LLAMA_API void llama_free(struct llama_context * ctx);
def llama_free(ctx: llama_context_p):
"""Frees all allocated memory"""
return _lib.llama_free(ctx)
_lib.llama_free.argtypes = [llama_context_p]
_lib.llama_free.restype = None
# LLAMA_API int64_t llama_time_us(void);
def llama_time_us() -> int:
return _lib.llama_time_us()
_lib.llama_time_us.argtypes = []
_lib.llama_time_us.restype = ctypes.c_int64
# LLAMA_API int32_t llama_max_devices(void);
def llama_max_devices() -> int:
return _lib.llama_max_devices()
_lib.llama_max_devices.argtypes = []
_lib.llama_max_devices.restype = ctypes.c_int32
# LLAMA_API bool llama_mmap_supported (void);
def llama_mmap_supported() -> bool:
return _lib.llama_mmap_supported()
_lib.llama_mmap_supported.argtypes = []
_lib.llama_mmap_supported.restype = c_bool
# LLAMA_API bool llama_mlock_supported(void);
def llama_mlock_supported() -> bool:
return _lib.llama_mlock_supported()
_lib.llama_mlock_supported.argtypes = []
_lib.llama_mlock_supported.restype = c_bool
# LLAMA_API const struct llama_model * llama_get_model(const struct llama_context * ctx);
def llama_get_model(ctx: llama_context_p) -> llama_model_p:
return _lib.llama_get_model(ctx)
_lib.llama_get_model.argtypes = [llama_context_p]
_lib.llama_get_model.restype = llama_model_p
# LLAMA_API uint32_t llama_n_ctx (const struct llama_context * ctx);
def llama_n_ctx(ctx: llama_context_p) -> int:
return _lib.llama_n_ctx(ctx)
_lib.llama_n_ctx.argtypes = [llama_context_p]
_lib.llama_n_ctx.restype = c_uint32
# LLAMA_API uint32_t llama_n_batch (const struct llama_context * ctx);
def llama_n_batch(ctx: llama_context_p) -> int:
return _lib.llama_n_batch(ctx)
_lib.llama_n_batch.argtypes = [llama_context_p]
_lib.llama_n_batch.restype = c_uint32
# LLAMA_API enum llama_vocab_type llama_vocab_type(const struct llama_model * model);
def llama_vocab_type(model: llama_model_p) -> int:
return _lib.llama_vocab_type(model)
_lib.llama_vocab_type.argtypes = [llama_model_p]
_lib.llama_vocab_type.restype = c_int
# LLAMA_API int32_t llama_n_vocab (const struct llama_model * model);
def llama_n_vocab(model: llama_model_p) -> int:
return _lib.llama_n_vocab(model)
_lib.llama_n_vocab.argtypes = [llama_model_p]
_lib.llama_n_vocab.restype = c_int32
# LLAMA_API int32_t llama_n_ctx_train(const struct llama_model * model);
def llama_n_ctx_train(model: llama_model_p) -> int:
return _lib.llama_n_ctx_train(model)
_lib.llama_n_ctx_train.argtypes = [llama_model_p]
_lib.llama_n_ctx_train.restype = c_int32
# LLAMA_API int32_t llama_n_embd (const struct llama_model * model);
def llama_n_embd(model: llama_model_p) -> int:
return _lib.llama_n_embd(model)
_lib.llama_n_embd.argtypes = [llama_model_p]
_lib.llama_n_embd.restype = c_int32
# // Get the model's RoPE frequency scaling factor
# LLAMA_API float llama_rope_freq_scale_train(const struct llama_model * model);
def llama_rope_freq_scale_train(model: llama_model_p) -> float:
"""Get the model's RoPE frequency scaling factor"""
return _lib.llama_rope_freq_scale_train(model)
_lib.llama_rope_freq_scale_train.argtypes = [llama_model_p]
_lib.llama_rope_freq_scale_train.restype = c_float
# // Functions to access the model's GGUF metadata scalar values
# // - The functions return the length of the string on success, or -1 on failure
# // - The output string is always null-terminated and cleared on failure
# // - GGUF array values are not supported by these functions
# // Get metadata value as a string by key name
# LLAMA_API int32_t llama_model_meta_val_str(const struct llama_model * model, const char * key, char * buf, size_t buf_size);
def llama_model_meta_val_str(
model: llama_model_p, key: Union[c_char_p, bytes], buf: bytes, buf_size: int
) -> int:
"""Get metadata value as a string by key name"""
return _lib.llama_model_meta_val_str(model, key, buf, buf_size)
_lib.llama_model_meta_val_str.argtypes = [llama_model_p, c_char_p, c_char_p, c_size_t]
_lib.llama_model_meta_val_str.restype = c_int32
# // Get the number of metadata key/value pairs
# LLAMA_API int32_t llama_model_meta_count(const struct llama_model * model);
def llama_model_meta_count(model: llama_model_p) -> int:
"""Get the number of metadata key/value pairs"""
return _lib.llama_model_meta_count(model)
_lib.llama_model_meta_count.argtypes = [llama_model_p]
_lib.llama_model_meta_count.restype = c_int32
# // Get metadata key name by index
# LLAMA_API int32_t llama_model_meta_key_by_index(const struct llama_model * model, int32_t i, char * buf, size_t buf_size);
def llama_model_meta_key_by_index(
model: llama_model_p, i: Union[c_int, int], buf: bytes, buf_size: int
) -> int:
"""Get metadata key name by index"""
return _lib.llama_model_meta_key_by_index(model, i, buf, buf_size)
_lib.llama_model_meta_key_by_index.argtypes = [
llama_model_p,
c_int32,
c_char_p,
c_size_t,
]
_lib.llama_model_meta_key_by_index.restype = c_int32
# // Get metadata value as a string by index
# LLAMA_API int32_t llama_model_meta_val_str_by_index(const struct llama_model * model, int32_t i, char * buf, size_t buf_size);
def llama_model_meta_val_str_by_index(
model: llama_model_p, i: Union[c_int, int], buf: bytes, buf_size: int
) -> int:
"""Get metadata value as a string by index"""
return _lib.llama_model_meta_val_str_by_index(model, i, buf, buf_size)
_lib.llama_model_meta_val_str_by_index.argtypes = [
llama_model_p,
c_int32,
c_char_p,
c_size_t,
]
_lib.llama_model_meta_val_str_by_index.restype = c_int32
# // Get a string describing the model type
# LLAMA_API int32_t llama_model_desc(const struct llama_model * model, char * buf, size_t buf_size);
def llama_model_desc(
model: llama_model_p, buf: bytes, buf_size: Union[c_size_t, int]
) -> int:
"""Get a string describing the model type"""
return _lib.llama_model_desc(model, buf, buf_size)
_lib.llama_model_desc.argtypes = [llama_model_p, c_char_p, c_size_t]
_lib.llama_model_desc.restype = c_int32
# // Returns the total size of all the tensors in the model in bytes
# LLAMA_API uint64_t llama_model_size(const struct llama_model * model);
def llama_model_size(model: llama_model_p) -> int:
"""Returns the total size of all the tensors in the model in bytes"""
return _lib.llama_model_size(model)
_lib.llama_model_size.argtypes = [llama_model_p]
_lib.llama_model_size.restype = ctypes.c_uint64
# // Returns the total number of parameters in the model
# LLAMA_API uint64_t llama_model_n_params(const struct llama_model * model);
def llama_model_n_params(model: llama_model_p) -> int:
"""Returns the total number of parameters in the model"""
return _lib.llama_model_n_params(model)
_lib.llama_model_n_params.argtypes = [llama_model_p]
_lib.llama_model_n_params.restype = ctypes.c_uint64
# // Get a llama model tensor
# LLAMA_API struct ggml_tensor * llama_get_model_tensor(struct llama_model * model, const char * name);
def llama_get_model_tensor(
model: llama_model_p, name: Union[c_char_p, bytes]
) -> c_void_p:
"""Get a llama model tensor"""
return _lib.llama_get_model_tensor(model, name)
_lib.llama_get_model_tensor.argtypes = [llama_model_p, c_char_p]
_lib.llama_get_model_tensor.restype = c_void_p
# // Returns 0 on success
# LLAMA_API uint32_t llama_model_quantize(
# const char * fname_inp,
# const char * fname_out,
# const llama_model_quantize_params * params);
def llama_model_quantize(
fname_inp: bytes,
fname_out: bytes,
params, # type: POINTER(llama_model_quantize_params) # type: ignore
) -> int:
"""Returns 0 on success"""
return _lib.llama_model_quantize(fname_inp, fname_out, params)
_lib.llama_model_quantize.argtypes = [
c_char_p,
c_char_p,
POINTER(llama_model_quantize_params),
]
_lib.llama_model_quantize.restype = c_uint32
# // 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
# LLAMA_API DEPRECATED(int32_t llama_apply_lora_from_file(
# struct llama_context * ctx,
# const char * path_lora,
# float scale,
# const char * path_base_model,
# int32_t n_threads),
# "use llama_model_apply_lora_from_file instead");
def llama_apply_lora_from_file(
ctx: llama_context_p,
path_lora: Union[c_char_p, bytes],
scale: Union[c_float, float],
You can’t perform that action at this time.