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_chat_format.py
More file actions
1306 lines (1168 loc) · 50 KB
/
Copy pathllama_chat_format.py
File metadata and controls
1306 lines (1168 loc) · 50 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
from __future__ import annotations
import os
import json
import ctypes
import dataclasses
from typing import Any, Dict, Iterator, List, Optional, Tuple, Union, Protocol
import llama_cpp.llama as llama
import llama_cpp.llama_types as llama_types
import llama_cpp.llama_grammar as llama_grammar
from ._utils import suppress_stdout_stderr
class LlamaChatCompletionHandler(Protocol):
def __call__(
self,
*,
llama: llama.Llama,
messages: List[llama_types.ChatCompletionRequestMessage],
functions: Optional[List[llama_types.ChatCompletionFunction]] = None,
function_call: Optional[llama_types.ChatCompletionRequestFunctionCall] = None,
tools: Optional[List[llama_types.ChatCompletionTool]] = None,
tool_choice: Optional[llama_types.ChatCompletionToolChoiceOption] = None,
temperature: float = 0.2,
top_p: float = 0.95,
top_k: int = 40,
min_p: float = 0.05,
typical_p: float = 1.0,
stream: bool = False,
stop: Optional[Union[str, List[str]]] = [],
seed: Optional[int] = None,
response_format: Optional[
llama_types.ChatCompletionRequestResponseFormat
] = None,
max_tokens: Optional[int] = None,
presence_penalty: float = 0.0,
frequency_penalty: float = 0.0,
repeat_penalty: float = 1.1,
tfs_z: float = 1.0,
mirostat_mode: int = 0,
mirostat_tau: float = 5.0,
mirostat_eta: float = 0.1,
model: Optional[str] = None,
logits_processor: Optional[llama.LogitsProcessorList] = None,
grammar: Optional[llama.LlamaGrammar] = None,
logit_bias: Optional[Dict[str, float]] = None,
**kwargs, # type: ignore
) -> Union[
llama_types.CreateChatCompletionResponse,
Iterator[llama_types.CreateChatCompletionStreamResponse],
]:
...
CHAT_HANDLERS: Dict[str, LlamaChatCompletionHandler] = {}
def get_chat_completion_handler(name: str) -> LlamaChatCompletionHandler:
return CHAT_HANDLERS[name]
def register_chat_completion_handler(name: str):
def decorator(f: LlamaChatCompletionHandler):
CHAT_HANDLERS[name] = f
return f
return decorator
def _get_system_message(
messages: List[llama_types.ChatCompletionRequestMessage],
) -> str:
"""Get the first system message."""
for message in messages:
if message["role"] == "system":
return message["content"] or ""
return ""
def _map_roles(
messages: List[llama_types.ChatCompletionRequestMessage], role_map: Dict[str, str]
) -> List[Tuple[str, Optional[str]]]:
"""Map the message roles."""
output: List[Tuple[str, Optional[str]]] = []
for message in messages:
role = message["role"]
if role in role_map:
output.append((role_map[role], message["content"]))
return output
def _format_llama2(
system_message: str, messages: List[Tuple[str, Optional[str]]], sep: str, sep2: str
) -> str:
"""Format the prompt with the llama2 style."""
seps = [sep, sep2]
ret = system_message + sep
for i, (role, message) in enumerate(messages):
if system_message and i == 0:
ret += message + seps[i % 2]
elif message:
ret += role + message + " " + seps[i % 2]
else:
ret += role + " "
return ret
def _format_add_colon_single(
system_message: str, messages: List[Tuple[str, Optional[str]]], sep: str
) -> str:
"""Format the prompt with the add-colon-single style."""
ret = system_message + sep
for role, message in messages:
if message:
ret += role + ": " + message + sep
else:
ret += role + ":"
return ret
def _format_add_colon_two(
system_message: str, messages: List[Tuple[str, Optional[str]]], sep: str, sep2: str
) -> str:
"""Format the prompt with the add-colon-two style."""
seps = [sep, sep2]
ret = system_message + seps[0]
for i, (role, message) in enumerate(messages):
if message:
ret += role + ": " + message + seps[i % 2]
else:
ret += role + ":"
return ret
def _format_no_colon_single(
system_message: str, messages: List[Tuple[str, Optional[str]]], sep: str
) -> str:
"""Format the prompt with the no-colon-single style."""
ret = system_message
for role, message in messages:
if message:
ret += role + message + sep
else:
ret += role
return ret
def _format_add_colon_space_single(
system_message: str, messages: List[Tuple[str, Optional[str]]], sep: str
) -> str:
"""Format the prompt with the add-colon-space-single style."""
ret = system_message + sep
for role, message in messages:
if message:
ret += role + ": " + message + sep
else:
ret += role + ": " # must be end with a space
return ret
def _format_chatml(
system_message: str, messages: List[Tuple[str, Optional[str]]], sep: str
) -> str:
"""Format the prompt with the chatml style."""
ret = "" if system_message == "" else system_message + sep + "\n"
for role, message in messages:
if message:
ret += role + "\n" + message + sep + "\n"
else:
ret += role + "\n"
return ret
def _format_chatglm3(
system_message: str, messages: List[Tuple[str, Optional[str]]], sep: str
) -> str:
"""Format the prompt with the chatglm3 style."""
ret = ""
if system_message:
ret += system_message
for role, message in messages:
if message:
ret += role + "\n" + " " + message
else:
ret += role
return ret
@dataclasses.dataclass
class ChatFormatterResponse:
prompt: str
stop: Optional[Union[str, List[str]]] = None
class ChatFormatter(Protocol):
def __call__(
self,
*,
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
...
class BasicChatHandler:
def __init__(self, chat_format: str):
self.chat_format = chat_format
def _convert_text_completion_to_chat(
completion: llama_types.Completion,
) -> llama_types.ChatCompletion:
return {
"id": "chat" + completion["id"],
"object": "chat.completion",
"created": completion["created"],
"model": completion["model"],
"choices": [
{
"index": 0,
"message": {
"role": "assistant",
"content": completion["choices"][0]["text"],
},
"finish_reason": completion["choices"][0]["finish_reason"],
}
],
"usage": completion["usage"],
}
def _convert_text_completion_chunks_to_chat(
chunks: Iterator[llama_types.CreateCompletionStreamResponse],
) -> Iterator[llama_types.ChatCompletionChunk]:
for i, chunk in enumerate(chunks):
if i == 0:
yield {
"id": "chat" + chunk["id"],
"model": chunk["model"],
"created": chunk["created"],
"object": "chat.completion.chunk",
"choices": [
{
"index": 0,
"delta": {
"role": "assistant",
},
"finish_reason": None,
}
],
}
yield {
"id": "chat" + chunk["id"],
"model": chunk["model"],
"created": chunk["created"],
"object": "chat.completion.chunk",
"choices": [
{
"index": 0,
"delta": {
"content": chunk["choices"][0]["text"],
},
"finish_reason": None,
}
],
}
if chunk["choices"][0]["finish_reason"] is not None:
yield {
"id": "chat" + chunk["id"],
"model": chunk["model"],
"created": chunk["created"],
"object": "chat.completion.chunk",
"choices": [
{
"index": 0,
"delta": {},
"finish_reason": chunk["choices"][0]["finish_reason"],
}
],
}
def _convert_completion_to_chat(
completion_or_chunks: Union[
llama_types.CreateCompletionResponse,
Iterator[llama_types.CreateCompletionStreamResponse],
],
stream: bool = False,
) -> Union[
llama_types.CreateChatCompletionResponse, Iterator[llama_types.ChatCompletionChunk]
]:
if stream:
chunks: Iterator[llama_types.CreateCompletionStreamResponse] = completion_or_chunks # type: ignore
return _convert_text_completion_chunks_to_chat(chunks)
else:
completion: llama_types.Completion = completion_or_chunks # type: ignore
return _convert_text_completion_to_chat(completion)
_CHAT_FORMATS: Dict[str, ChatFormatter] = {}
def register_chat_format(name: str):
def decorator(f: ChatFormatter):
def basic_create_chat_completion(
*,
llama: llama.Llama,
messages: List[llama_types.ChatCompletionRequestMessage],
functions: Optional[List[llama_types.ChatCompletionFunction]] = None,
function_call: Optional[
llama_types.ChatCompletionRequestFunctionCall
] = None,
tools: Optional[List[llama_types.ChatCompletionTool]] = None,
tool_choice: Optional[llama_types.ChatCompletionToolChoiceOption] = None,
temperature: float = 0.2,
top_p: float = 0.95,
top_k: int = 40,
min_p: float = 0.05,
typical_p: float = 1.0,
stream: bool = False,
stop: Optional[Union[str, List[str]]] = [],
seed: Optional[int] = None,
response_format: Optional[
llama_types.ChatCompletionRequestResponseFormat
] = None,
max_tokens: Optional[int] = None,
presence_penalty: float = 0.0,
frequency_penalty: float = 0.0,
repeat_penalty: float = 1.1,
tfs_z: float = 1.0,
mirostat_mode: int = 0,
mirostat_tau: float = 5.0,
mirostat_eta: float = 0.1,
model: Optional[str] = None,
logits_processor: Optional[llama.LogitsProcessorList] = None,
grammar: Optional[llama.LlamaGrammar] = None,
logit_bias: Optional[Dict[str, float]] = None,
**kwargs, # type: ignore
) -> Union[
llama_types.CreateChatCompletionResponse,
Iterator[llama_types.CreateChatCompletionStreamResponse],
]:
result = f(
messages=messages,
functions=functions,
function_call=function_call,
)
prompt = result.prompt
if result.stop is not None:
stop = [] if stop is None else [stop] if isinstance(stop, str) else stop
rstop = result.stop if isinstance(result.stop, list) else [result.stop]
stop = stop + rstop
if response_format is not None and response_format["type"] == "json_object":
grammar = llama_grammar.LlamaGrammar.from_string(
llama_grammar.JSON_GBNF
)
completion_or_chunks = llama.create_completion(
prompt=prompt,
temperature=temperature,
top_p=top_p,
top_k=top_k,
min_p=min_p,
typical_p=typical_p,
stream=stream,
stop=stop,
seed=seed,
max_tokens=max_tokens,
presence_penalty=presence_penalty,
frequency_penalty=frequency_penalty,
repeat_penalty=repeat_penalty,
tfs_z=tfs_z,
mirostat_mode=mirostat_mode,
mirostat_tau=mirostat_tau,
mirostat_eta=mirostat_eta,
model=model,
logits_processor=logits_processor,
grammar=grammar,
logit_bias=logit_bias,
)
return _convert_completion_to_chat(completion_or_chunks, stream=stream)
register_chat_completion_handler(name)(basic_create_chat_completion)
return f
return decorator
def get_chat_format(name: str):
try:
return _CHAT_FORMATS[name]
except KeyError:
raise ValueError(
f"Invalid chat format: {name} (valid formats: {list(_CHAT_FORMATS.keys())})"
)
def hf_autotokenizer_to_chat_formatter(
pretrained_model_name_or_path: Union[str, os.PathLike[str]]
) -> ChatFormatter:
# https://huggingface.co/docs/transformers/main/chat_templating
# https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1#instruction-format
# https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1/blob/main/tokenizer_config.json
from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained(pretrained_model_name_or_path)
def format_autotokenizer(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
tokenizer.use_default_system_prompt = False
_prompt = tokenizer.apply_chat_template(messages, tokenize=False)
# Return formatted prompt and eos token by default
return ChatFormatterResponse(prompt=_prompt, stop=tokenizer.eos_token)
return format_autotokenizer
# see https://github.com/huggingface/transformers/blob/main/src/transformers/models/llama/tokenization_llama.py
# system prompt is "embedded" in the first message
@register_chat_format("llama-2")
def format_llama2(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
_system_template = "<s>[INST] <<SYS>>\n{system_message}\n<</SYS>>"
_roles = dict(user="<s>[INST]", assistant="[/INST]")
_messages = _map_roles(messages, _roles)
system_message = _get_system_message(messages)
if system_message:
system_message = _system_template.format(system_message=system_message)
_prompt = _format_llama2(system_message, _messages, " ", "</s>") + "[/INST]"
return ChatFormatterResponse(prompt=_prompt)
@register_chat_format("alpaca")
def format_alpaca(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
_roles = dict(user="### Instruction", assistant="### Response")
_sep = "\n\n"
_sep2 = "</s>"
system_message = _get_system_message(messages)
_messages = _map_roles(messages, _roles)
_prompt = _format_add_colon_two(system_message, _messages, _sep, _sep2)
return ChatFormatterResponse(prompt=_prompt)
@register_chat_format("qwen")
def format_qwen(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
_roles = dict(user="<|im_start|>user", assistant="<|im_start|>assistant")
system_message="You are a helpful assistant."
system_template="<|im_start|>system\n{system_message}"
system_message=system_template.format(system_message=system_message)
_messages = _map_roles(messages, _roles)
_messages.append((_roles["assistant"], None))
_sep = "<|im_end|>"
_prompt = _format_chatml(system_message, _messages, _sep)
_sep2 = "<|endoftext|>"
return ChatFormatterResponse(prompt=_prompt,stop=_sep2)
@register_chat_format("vicuna")
def format(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
_system_message = "A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions."
_roles = dict(user="USER", assistant="ASSISTANT")
_sep = " "
_sep2 = "</s>"
system_message = _system_message
_messages = _map_roles(messages, _roles)
_messages.append((_roles["assistant"], None))
_prompt = _format_add_colon_two(system_message, _messages, _sep, _sep2)
return ChatFormatterResponse(prompt=_prompt)
@register_chat_format("oasst_llama")
def format_oasst_llama(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
_system_template = "[INST] <<SYS>>\n{system_message}\n<</SYS>>\n\n"
_roles = dict(user="<|prompter|>", assistant="<|assistant|>")
_sep = "</s>"
system_message = _get_system_message(messages)
system_message = _system_template.format(system_message=system_message)
_messages = _map_roles(messages, _roles)
_messages.append((_roles["assistant"], None))
_prompt = _format_no_colon_single(system_message, _messages, _sep)
return ChatFormatterResponse(prompt=_prompt)
@register_chat_format("baichuan-2")
def format_baichuan2(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
_system_template = "{system_message}"
_roles = dict(user="<reserved_106>", assistant="<reserved_107>")
_sep = ""
system_message = _get_system_message(messages)
system_message = _system_template.format(system_message=system_message)
_messages = _map_roles(messages, _roles)
_messages.append((_roles["assistant"], None))
_prompt = _format_no_colon_single(system_message, _messages, _sep)
return ChatFormatterResponse(prompt=_prompt)
@register_chat_format("baichuan")
def format_baichuan(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
_system_template = "{system_message}"
_roles = dict(user="<reserved_102>", assistant="<reserved_103>")
_sep = ""
system_message = _get_system_message(messages)
system_message = _system_template.format(system_message=system_message)
_messages = _map_roles(messages, _roles)
_messages.append((_roles["assistant"], None))
_prompt = _format_no_colon_single(system_message, _messages, _sep)
return ChatFormatterResponse(prompt=_prompt)
@register_chat_format("openbuddy")
def format_openbuddy(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
_system_message = """Consider a conversation between User (a human) and Assistant (named Buddy).
Buddy is an INTP-T, a friendly, intelligent and multilingual AI assistant, by OpenBuddy team. GitHub: https://github.com/OpenBuddy/OpenBuddy
Buddy cannot access the Internet.
Buddy can fluently speak the user's language (e.g. English, Chinese).
Buddy can generate poems, stories, code, essays, songs, parodies, and more.
Buddy possesses vast knowledge about the world, history, and culture.
Buddy's responses are always safe, creative, high-quality, human-like, and interesting.
Buddy strictly refuses to discuss political, NSFW, or other unsafe topics.
User: Hi.
Assistant: Hi, I'm Buddy, your AI assistant. How can I help you today?"""
_roles = dict(user="User", assistant="Assistant")
_sep = "\n"
system_message = _system_message
_messages = _map_roles(messages, _roles)
_messages.append((_roles["assistant"], None))
_prompt = _format_add_colon_single(system_message, _messages, _sep)
return ChatFormatterResponse(prompt=_prompt)
@register_chat_format("redpajama-incite")
def format_redpajama_incite(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
_system_message = _get_system_message(messages)
_roles = dict(user="<human>", assistant="<bot>")
_sep = "\n"
_stop = "<human>"
system_message = _system_message
_messages = _map_roles(messages, _roles)
_messages.append((_roles["assistant"], None))
_prompt = _format_add_colon_single(system_message, _messages, _sep)
return ChatFormatterResponse(prompt=_prompt, stop=_stop)
@register_chat_format("snoozy")
def format_snoozy(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
system_template = "### Instruction:\n{system_message}"
default_system_message = "The prompt below is a question to answer, a task to complete, or a conversation to respond to; decide which and write an appropriate response."
_system_message = _get_system_message(messages)
_system_message = (
_system_message if _system_message != "" else default_system_message
)
system_message = system_template.format(system_message=_system_message)
_roles = dict(user="### Prompt", assistant="### Response")
_sep = "\n"
_stop = "###"
system_message = _system_message
_messages = _map_roles(messages, _roles)
_messages.append((_roles["assistant"], None))
_prompt = _format_add_colon_single(system_message, _messages, _sep)
return ChatFormatterResponse(prompt=_prompt, stop=_stop)
@register_chat_format("phind")
def format_phind(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
_roles = dict(user="### User Message", assistant="### Assistant")
_sep = "\n\n"
_system_message = "### System Prompt\nYou are an intelligent programming assistant."
_messages = _map_roles(messages, _roles)
_messages.append((_roles["assistant"], None))
_prompt = _format_add_colon_single(_system_message, _messages, _sep)
return ChatFormatterResponse(prompt=_prompt)
@register_chat_format("intel")
def format_intel(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
_roles = dict(user="### User:", assistant="### Assistant:")
_sep = "\n"
_system_message = "### System:\n{system_message}"
_messages = _map_roles(messages, _roles)
_messages.append((_roles["assistant"], None))
_prompt = _format_add_colon_single(_system_message, _messages, _sep)
return ChatFormatterResponse(prompt=_prompt)
@register_chat_format("open-orca")
def format_open_orca(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
system_template = "{system_message}"
system_message = (
"You are a helpful assistant. Please answer truthfully and write out your "
"thinking step by step to be sure you get the right answer. If you make a mistake or encounter "
"an error in your thinking, say so out loud and attempt to correct it. If you don't know or "
"aren't sure about something, say so clearly. You will act as a professional logician, mathematician, "
"and physicist. You will also act as the most appropriate type of expert to answer any particular "
"question or solve the relevant problem; state which expert type your are, if so. Also think of "
"any particular named expert that would be ideal to answer the relevant question or solve the "
"relevant problem; name and act as them, if appropriate."
)
roles = ("User", "Assistant")
sep = "<|end_of_turn|>\n"
# stop_token_ids=[32000, 32001], # "<|end_of_turn|>"
stop_str = "User"
system_message = system_template.format(system_message=system_message)
_messages = _map_roles(messages, dict(zip(roles, roles)))
_messages.append((roles[1], None))
_prompt = _format_add_colon_space_single(system_message, _messages, sep)
return ChatFormatterResponse(prompt=_prompt, stop=stop_str)
@register_chat_format("mistrallite")
def format_mistrallite(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
_roles = dict(user="<|prompter|>", assistant="</s>\n<|assistant|>")
_sep = " "
system_template = """<|system|>{system_message}</s>"""
system_message = _get_system_message(messages)
system_message = system_template.format(system_message=system_message)
_messages = _map_roles(messages, _roles)
_messages.append((_roles["assistant"], None))
_prompt = _format_no_colon_single(system_message, _messages, _sep)
return ChatFormatterResponse(prompt=_prompt)
@register_chat_format("zephyr")
def format_zephyr(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
system_template = """<|system|>
{system_message}"""
system_message = _get_system_message(messages)
system_message = system_template.format(system_message=system_message)
_roles = dict(user="<|user|>\n", assistant="<|assistant|>\n")
_sep = "</s>"
_messages = _map_roles(messages, _roles)
_messages.append((_roles["assistant"], None))
_prompt = _format_chatml(system_message, _messages, _sep)
return ChatFormatterResponse(prompt=_prompt, stop=_sep)
@register_chat_format("pygmalion")
def format_pygmalion(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
system_template = """<|system|>{system_message}"""
system_message = _get_system_message(messages)
system_message = system_template.format(system_message=system_message)
_roles = dict(user="<|user|>", assistant="<|model|>")
_sep = "\n"
_messages = _map_roles(messages, _roles)
_messages.append((_roles["assistant"], None))
_prompt = _format_chatml(system_message, _messages, _sep)
return ChatFormatterResponse(prompt=_prompt, stop=_sep)
@register_chat_format("chatml")
def format_chatml(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
system_template = """<|im_start|>system
{system_message}"""
system_message = _get_system_message(messages)
system_message = system_template.format(system_message=system_message)
_roles = dict(user="<|im_start|>user", assistant="<|im_start|>assistant")
_sep = "<|im_end|>"
_messages = _map_roles(messages, _roles)
_messages.append((_roles["assistant"], None))
_prompt = _format_chatml(system_message, _messages, _sep)
return ChatFormatterResponse(prompt=_prompt, stop=_sep)
@register_chat_format("chatglm3")
def format_chatglm3(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
system_template = """<|system|>
{system_message}"""
system_message = _get_system_message(messages)
system_message = system_template.format(system_message=system_message)
_roles = dict(user="<|user|>", assistant="<|assistant|>")
_sep = "</s>"
_messages = _map_roles(messages, _roles)
_messages.append((_roles["assistant"], None))
_prompt = _format_chatglm3(system_message, _messages, _sep)
return ChatFormatterResponse(prompt=_prompt, stop=_sep)
@register_chat_format("openchat")
def format_openchat(
messages: List[llama_types.ChatCompletionRequestMessage],
**kwargs: Any,
) -> ChatFormatterResponse:
system_template = "{system_message}<|end_of_turn|>"
system_message = _get_system_message(messages)
system_message = system_template.format(system_message=system_message)
_roles = dict(
user="GPT4 Correct User: ", assistant="<|end_of_turn|>GPT4 Correct Assistant: "
)
_sep = "<|end_of_turn|>"
_messages = _map_roles(messages, _roles)
_messages.append((_roles["assistant"], None))
_prompt = _format_chatml(system_message, _messages, _sep)
return ChatFormatterResponse(prompt=_prompt, stop=_sep)
# Chat format for Saiga models, see more details and available models:
# https://huggingface.co/collections/IlyaGusev/saiga2-saigamistral-6505d4ccc3d1e53166b636cd
@register_chat_format("saiga")
def format_saiga(
messages: list[llama_types.ChatCompletionRequestMessage],
**kwargs,
) -> ChatFormatterResponse:
_message_template = "<s>{role}\n{content}</s>"
_roles = dict(user="user", bot="bot", system="system")
_messages = _map_roles(messages, _roles)
_prompt = ""
for role, content in _messages:
if content:
_prompt += _message_template.format(role=role, content=content)
else:
_prompt += f"<s>{role}\n"
# Response template
_prompt += "<s>bot"
return ChatFormatterResponse(prompt=_prompt.strip())
@register_chat_completion_handler("functionary")
def functionary_chat_handler(
llama: llama.Llama,
messages: List[llama_types.ChatCompletionRequestMessage],
functions: Optional[List[llama_types.ChatCompletionFunction]] = None,
function_call: Optional[llama_types.ChatCompletionRequestFunctionCall] = None,
tools: Optional[List[llama_types.ChatCompletionTool]] = None,
tool_choice: Optional[llama_types.ChatCompletionToolChoiceOption] = None,
temperature: float = 0.2,
top_p: float = 0.95,
top_k: int = 40,
min_p: float = 0.05,
typical_p: float = 1.0,
stream: bool = False,
stop: Optional[Union[str, List[str]]] = [],
response_format: Optional[llama_types.ChatCompletionRequestResponseFormat] = None,
max_tokens: Optional[int] = None,
presence_penalty: float = 0.0,
frequency_penalty: float = 0.0,
repeat_penalty: float = 1.1,
tfs_z: float = 1.0,
mirostat_mode: int = 0,
mirostat_tau: float = 5.0,
mirostat_eta: float = 0.1,
model: Optional[str] = None,
logits_processor: Optional[llama.LogitsProcessorList] = None,
grammar: Optional[llama.LlamaGrammar] = None,
**kwargs, # type: ignore
) -> Union[llama_types.ChatCompletion, Iterator[llama_types.ChatCompletionChunk]]:
SYSTEM_MESSAGE = """A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. The assistant calls functions with appropriate input when necessary"""
def generate_type_definition(
param: Dict[str, llama_types.JsonType], indent_level: int, shared_defs
) -> str:
indent = " " * indent_level
if "$ref" in param:
# Reference to a shared definition
ref_name = param["$ref"].split("/")[
-1
] # Extract the type name from the reference
return ref_name
elif param.get("type") == "array":
items = param.get("items", {})
item_type = generate_type_definition(items, indent_level + 1, shared_defs)
return f"Array<{item_type}>"
elif param.get("type") == "object":
properties = param.get("properties", {})
nested_schema = "{\n"
for nested_param_name, nested_param in properties.items():
nested_param_type = generate_type_definition(
nested_param, indent_level + 1, shared_defs
)
nested_schema += (
f"{indent} {nested_param_name}: {nested_param_type},\n"
)
nested_schema += indent + "}"
return nested_schema
elif "enum" in param:
# Enum type
return " | ".join([f'"{enum_value}"' for enum_value in param["enum"]])
else:
# Simple type
return param.get("type", "any")
def generate_shared_definitions(shared_defs, indent_level: int) -> str:
indent = " " * indent_level
shared_definitions = ""
for def_name, def_properties in shared_defs.items():
shared_definitions += f"{indent}type {def_name} = "
if def_properties.get("type") == "object":
shared_definitions += generate_type_definition(
def_properties, indent_level, shared_defs
)
elif "enum" in def_properties:
# Enum type
shared_definitions += " | ".join(
[f'"{enum_value}"' for enum_value in def_properties["enum"]]
)
shared_definitions += ";\n"
return shared_definitions
def generate_schema_from_functions(functions, namespace="functions") -> str:
schema = (
"// Supported function definitions that should be called when necessary.\n"
)
schema += f"namespace {namespace} {{\n\n"
# Generate shared definitions
shared_definitions = {}
for function in functions:
parameters = function.get("parameters", {})
shared_definitions.update(parameters.get("$defs", {}))
schema += generate_shared_definitions(shared_definitions, 1)
for function in functions:
function_name = function["name"]
description = function.get("description", "")
parameters = function.get("parameters", {})
required_params = parameters.get("required", [])
schema += f" // {description}\n"
schema += f" type {function_name} = (_: {{\n"
for param_name, param in parameters.get("properties", {}).items():
param_description = param.get("description", "")
param_type = generate_type_definition(param, 2, shared_definitions)
optional_indicator = "" if param_name in required_params else "?"
schema += f" // {param_description}\n"
schema += f" {param_name}{optional_indicator}: {param_type},\n"
schema += " }) => any;\n\n"
schema += "}} // namespace {}\n".format(namespace)
return schema
def prepare_messages_for_inference(
messages: List[llama_types.ChatCompletionRequestMessage],
functions: Optional[List[llama_types.ChatCompletionFunctions]] = None,
tools: Optional[List[llama_types.ChatCompletionTool]] = None,
):
all_messages: List[llama_types.ChatCompletionRequestMessage] = []
if functions is not None:
all_messages.append(
llama_types.ChatCompletionRequestSystemMessage(
role="system", content=generate_schema_from_functions(functions)
)
)
if tools is not None:
all_messages.append(
llama_types.ChatCompletionRequestSystemMessage(
role="system",
content=generate_schema_from_functions(
[
tool["function"]
for tool in tools
if tool["type"] == "function"
]
),
)
)
all_messages.append(
llama_types.ChatCompletionRequestSystemMessage(
role="system", content=SYSTEM_MESSAGE
)
)
for message in messages:
# Function call responses
if message["role"] == "function" and "name" in message:
message["name"] = f"functions.{message['name']}"
# Function call requests by assistant
if "function_call" in message:
message["function_call"][
"name"
] = f"functions.{message['function_call']['name']}"
all_messages.append(message)
all_messages.append(
llama_types.ChatCompletionRequestAssistantMessage(
role="assistant", content=None
)
)
def message_to_str(msg: llama_types.ChatCompletionRequestMessage):
if msg["role"] == "system":
return f"system:\n{msg['content']}\n"
elif msg["role"] == "function" and "name" in msg:
return f"function name={msg['name']}:\n{msg['content']}\n"
elif msg["role"] == "function" and "function_call" in msg:
return f"function name={msg['function_call']['name']}:\n{msg['function_call']['arguments']}\n"
elif msg["role"] == "tool":
if msg["content"] is not None:
return f"function name={msg['tool_call_id']}:\n{msg['content']}\n"
else:
return f"function name={msg['tool_call_id']}\n"
elif msg["role"] == "user":
if msg["content"] is None:
return "user:\n</s></s>\n"
else:
return f"user:\n</s>{msg['content']}</s>\n"
elif msg["role"] == "assistant":
if msg["content"] is not None and "function_call" in msg:
return f"assistant:\n{msg['content']}\nassistant to={msg['function_call']['name']}:\n{msg['function_call']['arguments']}</s>\n"
elif "function_call" in msg:
return f"assistant to={msg['function_call']['name']}:\n{msg['function_call']['arguments']}</s>\n"
elif "tool_calls" in msg and len(msg["tool_calls"]) > 0:
for tool_call in msg[
"tool_calls"
]: # NOTE: probably doesn't work with the functionary model
return f"assistant to={tool_call['id']}:\n{tool_call['function']['arguments']}</s>\n"
elif msg["content"] is None:
return "assistant"
else:
return f"assistant:\n{msg['content']}\n"
else:
raise ValueError(f"Unsupported role: {msg['role']}")
return "".join([message_to_str(msg) for msg in all_messages])
if tools is not None:
functions = [tool["function"] for tool in tools if tool["type"] == "function"]
if tool_choice is not None:
function_call = (
tool_choice if isinstance(tool_choice, str) else tool_choice["function"]
)
prompt = prepare_messages_for_inference(messages, functions, tools)
if function_call is None and (functions is None or len(functions) == 0):
completion_or_completion_chunks = llama.create_completion(
prompt=prompt + ":\n",
temperature=temperature,
top_p=top_p,
top_k=top_k,
min_p=min_p,
typical_p=typical_p,
stream=stream,
stop=["user:", "</s>"],
max_tokens=max_tokens,
presence_penalty=presence_penalty,
frequency_penalty=frequency_penalty,
repeat_penalty=repeat_penalty,
tfs_z=tfs_z,
mirostat_mode=mirostat_mode,
mirostat_tau=mirostat_tau,
mirostat_eta=mirostat_eta,
You can’t perform that action at this time.