-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy path186.py
More file actions
1125 lines (1009 loc) · 43.7 KB
/
Copy path186.py
File metadata and controls
1125 lines (1009 loc) · 43.7 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 ast
import logging
from enum import Enum, EnumMeta
from json import JSONDecodeError
from types import MappingProxyType
from typing import TYPE_CHECKING, Any, List, Literal, NamedTuple
from pydantic import SecretStr
from backend.integrations.providers import ProviderName
if TYPE_CHECKING:
from enum import _EnumMemberT
import anthropic
import ollama
import openai
from groq import Groq
from backend.data.block import Block, BlockCategory, BlockOutput, BlockSchema
from backend.data.model import (
APIKeyCredentials,
CredentialsField,
CredentialsMetaInput,
SchemaField,
)
from backend.util import json
from backend.util.settings import BehaveAs, Settings
logger = logging.getLogger(__name__)
LLMProviderName = Literal[
ProviderName.ANTHROPIC,
ProviderName.GROQ,
ProviderName.OLLAMA,
ProviderName.OPENAI,
ProviderName.OPEN_ROUTER,
]
AICredentials = CredentialsMetaInput[LLMProviderName, Literal["api_key"]]
TEST_CREDENTIALS = APIKeyCredentials(
id="ed55ac19-356e-4243-a6cb-bc599e9b716f",
provider="openai",
api_key=SecretStr("mock-openai-api-key"),
title="Mock OpenAI API key",
expires_at=None,
)
TEST_CREDENTIALS_INPUT = {
"provider": TEST_CREDENTIALS.provider,
"id": TEST_CREDENTIALS.id,
"type": TEST_CREDENTIALS.type,
"title": TEST_CREDENTIALS.title,
}
def AICredentialsField() -> AICredentials:
return CredentialsField(
description="API key for the LLM provider.",
discriminator="model",
discriminator_mapping={
model.value: model.metadata.provider for model in LlmModel
},
)
class ModelMetadata(NamedTuple):
provider: str
context_window: int
class LlmModelMeta(EnumMeta):
@property
def __members__(
self: type["_EnumMemberT"],
) -> MappingProxyType[str, "_EnumMemberT"]:
if Settings().config.behave_as == BehaveAs.LOCAL:
members = super().__members__
return members
else:
removed_providers = ["ollama"]
existing_members = super().__members__
members = {
name: member
for name, member in existing_members.items()
if LlmModel[name].provider not in removed_providers
}
return MappingProxyType(members)
class LlmModel(str, Enum, metaclass=LlmModelMeta):
# OpenAI models
O1_PREVIEW = "o1-preview"
O1_MINI = "o1-mini"
GPT4O_MINI = "gpt-4o-mini"
GPT4O = "gpt-4o"
GPT4_TURBO = "gpt-4-turbo"
GPT3_5_TURBO = "gpt-3.5-turbo"
# Anthropic models
CLAUDE_3_5_SONNET = "claude-3-5-sonnet-latest"
CLAUDE_3_HAIKU = "claude-3-haiku-20240307"
# Groq models
LLAMA3_8B = "llama3-8b-8192"
LLAMA3_70B = "llama3-70b-8192"
MIXTRAL_8X7B = "mixtral-8x7b-32768"
GEMMA_7B = "gemma-7b-it"
GEMMA2_9B = "gemma2-9b-it"
# New Groq models (Preview)
LLAMA3_1_405B = "llama-3.1-405b-reasoning"
LLAMA3_1_70B = "llama-3.1-70b-versatile"
LLAMA3_1_8B = "llama-3.1-8b-instant"
# Ollama models
OLLAMA_LLAMA3_8B = "llama3"
OLLAMA_LLAMA3_405B = "llama3.1:405b"
OLLAMA_DOLPHIN = "dolphin-mistral:latest"
# OpenRouter models
GEMINI_FLASH_1_5_8B = "google/gemini-flash-1.5"
GROK_BETA = "x-ai/grok-beta"
MISTRAL_NEMO = "mistralai/mistral-nemo"
COHERE_COMMAND_R_08_2024 = "cohere/command-r-08-2024"
COHERE_COMMAND_R_PLUS_08_2024 = "cohere/command-r-plus-08-2024"
EVA_QWEN_2_5_32B = "eva-unit-01/eva-qwen-2.5-32b"
DEEPSEEK_CHAT = "deepseek/deepseek-chat"
PERPLEXITY_LLAMA_3_1_SONAR_LARGE_128K_ONLINE = (
"perplexity/llama-3.1-sonar-large-128k-online"
)
QWEN_QWQ_32B_PREVIEW = "qwen/qwq-32b-preview"
NOUSRESEARCH_HERMES_3_LLAMA_3_1_405B = "nousresearch/hermes-3-llama-3.1-405b"
NOUSRESEARCH_HERMES_3_LLAMA_3_1_70B = "nousresearch/hermes-3-llama-3.1-70b"
AMAZON_NOVA_LITE_V1 = "amazon/nova-lite-v1"
AMAZON_NOVA_MICRO_V1 = "amazon/nova-micro-v1"
AMAZON_NOVA_PRO_V1 = "amazon/nova-pro-v1"
MICROSOFT_WIZARDLM_2_8X22B = "microsoft/wizardlm-2-8x22b"
GRYPHE_MYTHOMAX_L2_13B = "gryphe/mythomax-l2-13b"
@property
def metadata(self) -> ModelMetadata:
return MODEL_METADATA[self]
@property
def provider(self) -> str:
return self.metadata.provider
@property
def context_window(self) -> int:
return self.metadata.context_window
MODEL_METADATA = {
LlmModel.O1_PREVIEW: ModelMetadata("openai", 32000),
LlmModel.O1_MINI: ModelMetadata("openai", 62000),
LlmModel.GPT4O_MINI: ModelMetadata("openai", 128000),
LlmModel.GPT4O: ModelMetadata("openai", 128000),
LlmModel.GPT4_TURBO: ModelMetadata("openai", 128000),
LlmModel.GPT3_5_TURBO: ModelMetadata("openai", 16385),
LlmModel.CLAUDE_3_5_SONNET: ModelMetadata("anthropic", 200000),
LlmModel.CLAUDE_3_HAIKU: ModelMetadata("anthropic", 200000),
LlmModel.LLAMA3_8B: ModelMetadata("groq", 8192),
LlmModel.LLAMA3_70B: ModelMetadata("groq", 8192),
LlmModel.MIXTRAL_8X7B: ModelMetadata("groq", 32768),
LlmModel.GEMMA_7B: ModelMetadata("groq", 8192),
LlmModel.GEMMA2_9B: ModelMetadata("groq", 8192),
LlmModel.LLAMA3_1_405B: ModelMetadata("groq", 8192),
# Limited to 16k during preview
LlmModel.LLAMA3_1_70B: ModelMetadata("groq", 131072),
LlmModel.LLAMA3_1_8B: ModelMetadata("groq", 131072),
LlmModel.OLLAMA_LLAMA3_8B: ModelMetadata("ollama", 8192),
LlmModel.OLLAMA_LLAMA3_405B: ModelMetadata("ollama", 8192),
LlmModel.OLLAMA_DOLPHIN: ModelMetadata("ollama", 32768),
LlmModel.GEMINI_FLASH_1_5_8B: ModelMetadata("open_router", 8192),
LlmModel.GROK_BETA: ModelMetadata("open_router", 8192),
LlmModel.MISTRAL_NEMO: ModelMetadata("open_router", 4000),
LlmModel.COHERE_COMMAND_R_08_2024: ModelMetadata("open_router", 4000),
LlmModel.COHERE_COMMAND_R_PLUS_08_2024: ModelMetadata("open_router", 4000),
LlmModel.EVA_QWEN_2_5_32B: ModelMetadata("open_router", 4000),
LlmModel.DEEPSEEK_CHAT: ModelMetadata("open_router", 8192),
LlmModel.PERPLEXITY_LLAMA_3_1_SONAR_LARGE_128K_ONLINE: ModelMetadata(
"open_router", 8192
),
LlmModel.QWEN_QWQ_32B_PREVIEW: ModelMetadata("open_router", 4000),
LlmModel.NOUSRESEARCH_HERMES_3_LLAMA_3_1_405B: ModelMetadata("open_router", 4000),
LlmModel.NOUSRESEARCH_HERMES_3_LLAMA_3_1_70B: ModelMetadata("open_router", 4000),
LlmModel.AMAZON_NOVA_LITE_V1: ModelMetadata("open_router", 4000),
LlmModel.AMAZON_NOVA_MICRO_V1: ModelMetadata("open_router", 4000),
LlmModel.AMAZON_NOVA_PRO_V1: ModelMetadata("open_router", 4000),
LlmModel.MICROSOFT_WIZARDLM_2_8X22B: ModelMetadata("open_router", 4000),
LlmModel.GRYPHE_MYTHOMAX_L2_13B: ModelMetadata("open_router", 4000),
}
for model in LlmModel:
if model not in MODEL_METADATA:
raise ValueError(f"Missing MODEL_METADATA metadata for model: {model}")
class MessageRole(str, Enum):
SYSTEM = "system"
USER = "user"
ASSISTANT = "assistant"
class Message(BlockSchema):
role: MessageRole
content: str
class AIStructuredResponseGeneratorBlock(Block):
class Input(BlockSchema):
prompt: str = SchemaField(
description="The prompt to send to the language model.",
placeholder="Enter your prompt here...",
)
expected_format: dict[str, str] = SchemaField(
description="Expected format of the response. If provided, the response will be validated against this format. "
"The keys should be the expected fields in the response, and the values should be the description of the field.",
)
model: LlmModel = SchemaField(
title="LLM Model",
default=LlmModel.GPT4_TURBO,
description="The language model to use for answering the prompt.",
advanced=False,
)
credentials: AICredentials = AICredentialsField()
sys_prompt: str = SchemaField(
title="System Prompt",
default="",
description="The system prompt to provide additional context to the model.",
)
conversation_history: list[Message] = SchemaField(
default=[],
description="The conversation history to provide context for the prompt.",
)
retry: int = SchemaField(
title="Retry Count",
default=3,
description="Number of times to retry the LLM call if the response does not match the expected format.",
)
prompt_values: dict[str, str] = SchemaField(
advanced=False, default={}, description="Values used to fill in the prompt."
)
max_tokens: int | None = SchemaField(
advanced=True,
default=None,
description="The maximum number of tokens to generate in the chat completion.",
)
ollama_host: str = SchemaField(
advanced=True,
default="localhost:11434",
description="Ollama host for local models",
)
class Output(BlockSchema):
response: dict[str, Any] = SchemaField(
description="The response object generated by the language model."
)
error: str = SchemaField(description="Error message if the API call failed.")
def __init__(self):
super().__init__(
id="ed55ac19-356e-4243-a6cb-bc599e9b716f",
description="Call a Large Language Model (LLM) to generate formatted object based on the given prompt.",
categories={BlockCategory.AI},
input_schema=AIStructuredResponseGeneratorBlock.Input,
output_schema=AIStructuredResponseGeneratorBlock.Output,
test_input={
"model": LlmModel.GPT4_TURBO,
"credentials": TEST_CREDENTIALS_INPUT,
"expected_format": {
"key1": "value1",
"key2": "value2",
},
"prompt": "User prompt",
},
test_credentials=TEST_CREDENTIALS,
test_output=("response", {"key1": "key1Value", "key2": "key2Value"}),
test_mock={
"llm_call": lambda *args, **kwargs: (
json.dumps(
{
"key1": "key1Value",
"key2": "key2Value",
}
),
0,
0,
)
},
)
@staticmethod
def llm_call(
credentials: APIKeyCredentials,
llm_model: LlmModel,
prompt: list[dict],
json_format: bool,
max_tokens: int | None = None,
ollama_host: str = "localhost:11434",
) -> tuple[str, int, int]:
"""
Args:
api_key: API key for the LLM provider.
llm_model: The LLM model to use.
prompt: The prompt to send to the LLM.
json_format: Whether the response should be in JSON format.
max_tokens: The maximum number of tokens to generate in the chat completion.
ollama_host: The host for ollama to use
Returns:
The response from the LLM.
The number of tokens used in the prompt.
The number of tokens used in the completion.
"""
provider = llm_model.metadata.provider
if provider == "openai":
oai_client = openai.OpenAI(api_key=credentials.api_key.get_secret_value())
response_format = None
if llm_model in [LlmModel.O1_MINI, LlmModel.O1_PREVIEW]:
sys_messages = [p["content"] for p in prompt if p["role"] == "system"]
usr_messages = [p["content"] for p in prompt if p["role"] != "system"]
prompt = [
{"role": "user", "content": "\n".join(sys_messages)},
{"role": "user", "content": "\n".join(usr_messages)},
]
elif json_format:
response_format = {"type": "json_object"}
response = oai_client.chat.completions.create(
model=llm_model.value,
messages=prompt, # type: ignore
response_format=response_format, # type: ignore
max_completion_tokens=max_tokens,
)
return (
response.choices[0].message.content or "",
response.usage.prompt_tokens if response.usage else 0,
response.usage.completion_tokens if response.usage else 0,
)
elif provider == "anthropic":
system_messages = [p["content"] for p in prompt if p["role"] == "system"]
sysprompt = " ".join(system_messages)
messages = []
last_role = None
for p in prompt:
if p["role"] in ["user", "assistant"]:
if p["role"] != last_role:
messages.append({"role": p["role"], "content": p["content"]})
last_role = p["role"]
else:
# If the role is the same as the last one, combine the content
messages[-1]["content"] += "\n" + p["content"]
client = anthropic.Anthropic(api_key=credentials.api_key.get_secret_value())
try:
resp = client.messages.create(
model=llm_model.value,
system=sysprompt,
messages=messages,
max_tokens=max_tokens or 8192,
)
if not resp.content:
raise ValueError("No content returned from Anthropic.")
return (
(
resp.content[0].name
if isinstance(resp.content[0], anthropic.types.ToolUseBlock)
else resp.content[0].text
),
resp.usage.input_tokens,
resp.usage.output_tokens,
)
except anthropic.APIError as e:
error_message = f"Anthropic API error: {str(e)}"
logger.error(error_message)
raise ValueError(error_message)
elif provider == "groq":
client = Groq(api_key=credentials.api_key.get_secret_value())
response_format = {"type": "json_object"} if json_format else None
response = client.chat.completions.create(
model=llm_model.value,
messages=prompt, # type: ignore
response_format=response_format, # type: ignore
max_tokens=max_tokens,
)
return (
response.choices[0].message.content or "",
response.usage.prompt_tokens if response.usage else 0,
response.usage.completion_tokens if response.usage else 0,
)
elif provider == "ollama":
client = ollama.Client(host=ollama_host)
sys_messages = [p["content"] for p in prompt if p["role"] == "system"]
usr_messages = [p["content"] for p in prompt if p["role"] != "system"]
response = client.generate(
model=llm_model.value,
prompt=f"{sys_messages}\n\n{usr_messages}",
stream=False,
)
return (
response.get("response") or "",
response.get("prompt_eval_count") or 0,
response.get("eval_count") or 0,
)
elif provider == "open_router":
client = openai.OpenAI(
base_url="https://openrouter.ai/api/v1",
api_key=credentials.api_key.get_secret_value(),
)
response = client.chat.completions.create(
extra_headers={
"HTTP-Referer": "https://agpt.co",
"X-Title": "AutoGPT",
},
model=llm_model.value,
messages=prompt, # type: ignore
max_tokens=max_tokens,
)
# If there's no response, raise an error
if not response.choices:
if response:
raise ValueError(f"OpenRouter error: {response}")
else:
raise ValueError("No response from OpenRouter.")
return (
response.choices[0].message.content or "",
response.usage.prompt_tokens if response.usage else 0,
response.usage.completion_tokens if response.usage else 0,
)
else:
raise ValueError(f"Unsupported LLM provider: {provider}")
def run(
self, input_data: Input, *, credentials: APIKeyCredentials, **kwargs
) -> BlockOutput:
logger.debug(f"Calling LLM with input data: {input_data}")
prompt = [p.model_dump() for p in input_data.conversation_history]
def trim_prompt(s: str) -> str:
lines = s.strip().split("\n")
return "\n".join([line.strip().lstrip("|") for line in lines])
values = input_data.prompt_values
if values:
input_data.prompt = input_data.prompt.format(**values)
input_data.sys_prompt = input_data.sys_prompt.format(**values)
if input_data.sys_prompt:
prompt.append({"role": "system", "content": input_data.sys_prompt})
if input_data.expected_format:
expected_format = [
f'"{k}": "{v}"' for k, v in input_data.expected_format.items()
]
format_prompt = ",\n ".join(expected_format)
sys_prompt = trim_prompt(
f"""
|Reply strictly only in the following JSON format:
|{{
| {format_prompt}
|}}
"""
)
prompt.append({"role": "system", "content": sys_prompt})
if input_data.prompt:
prompt.append({"role": "user", "content": input_data.prompt})
def parse_response(resp: str) -> tuple[dict[str, Any], str | None]:
try:
parsed = json.loads(resp)
if not isinstance(parsed, dict):
return {}, f"Expected a dictionary, but got {type(parsed)}"
miss_keys = set(input_data.expected_format.keys()) - set(parsed.keys())
if miss_keys:
return parsed, f"Missing keys: {miss_keys}"
return parsed, None
except JSONDecodeError as e:
return {}, f"JSON decode error: {e}"
logger.info(f"LLM request: {prompt}")
retry_prompt = ""
llm_model = input_data.model
for retry_count in range(input_data.retry):
try:
response_text, input_token, output_token = self.llm_call(
credentials=credentials,
llm_model=llm_model,
prompt=prompt,
json_format=bool(input_data.expected_format),
ollama_host=input_data.ollama_host,
max_tokens=input_data.max_tokens,
)
self.merge_stats(
{
"input_token_count": input_token,
"output_token_count": output_token,
}
)
logger.info(f"LLM attempt-{retry_count} response: {response_text}")
if input_data.expected_format:
parsed_dict, parsed_error = parse_response(response_text)
if not parsed_error:
yield "response", {
k: (
json.loads(v)
if isinstance(v, str)
and v.startswith("[")
and v.endswith("]")
else (", ".join(v) if isinstance(v, list) else v)
)
for k, v in parsed_dict.items()
}
return
else:
yield "response", {"response": response_text}
return
retry_prompt = trim_prompt(
f"""
|This is your previous error response:
|--
|{response_text}
|--
|
|And this is the error:
|--
|{parsed_error}
|--
"""
)
prompt.append({"role": "user", "content": retry_prompt})
except Exception as e:
logger.exception(f"Error calling LLM: {e}")
retry_prompt = f"Error calling LLM: {e}"
finally:
self.merge_stats(
{
"llm_call_count": retry_count + 1,
"llm_retry_count": retry_count,
}
)
raise RuntimeError(retry_prompt)
class AITextGeneratorBlock(Block):
class Input(BlockSchema):
prompt: str = SchemaField(
description="The prompt to send to the language model. You can use any of the {keys} from Prompt Values to fill in the prompt with values from the prompt values dictionary by putting them in curly braces.",
placeholder="Enter your prompt here...",
)
model: LlmModel = SchemaField(
title="LLM Model",
default=LlmModel.GPT4_TURBO,
description="The language model to use for answering the prompt.",
advanced=False,
)
credentials: AICredentials = AICredentialsField()
sys_prompt: str = SchemaField(
title="System Prompt",
default="",
description="The system prompt to provide additional context to the model.",
)
retry: int = SchemaField(
title="Retry Count",
default=3,
description="Number of times to retry the LLM call if the response does not match the expected format.",
)
prompt_values: dict[str, str] = SchemaField(
advanced=False, default={}, description="Values used to fill in the prompt."
)
ollama_host: str = SchemaField(
advanced=True,
default="localhost:11434",
description="Ollama host for local models",
)
max_tokens: int | None = SchemaField(
advanced=True,
default=None,
description="The maximum number of tokens to generate in the chat completion.",
)
class Output(BlockSchema):
response: str = SchemaField(
description="The response generated by the language model."
)
error: str = SchemaField(description="Error message if the API call failed.")
def __init__(self):
super().__init__(
id="1f292d4a-41a4-4977-9684-7c8d560b9f91",
description="Call a Large Language Model (LLM) to generate a string based on the given prompt.",
categories={BlockCategory.AI},
input_schema=AITextGeneratorBlock.Input,
output_schema=AITextGeneratorBlock.Output,
test_input={
"prompt": "User prompt",
"credentials": TEST_CREDENTIALS_INPUT,
},
test_credentials=TEST_CREDENTIALS,
test_output=("response", "Response text"),
test_mock={"llm_call": lambda *args, **kwargs: "Response text"},
)
def llm_call(
self,
input_data: AIStructuredResponseGeneratorBlock.Input,
credentials: APIKeyCredentials,
) -> str:
block = AIStructuredResponseGeneratorBlock()
response = block.run_once(input_data, "response", credentials=credentials)
self.merge_stats(block.execution_stats)
return response["response"]
def run(
self, input_data: Input, *, credentials: APIKeyCredentials, **kwargs
) -> BlockOutput:
object_input_data = AIStructuredResponseGeneratorBlock.Input(
**{attr: getattr(input_data, attr) for attr in input_data.model_fields},
expected_format={},
)
yield "response", self.llm_call(object_input_data, credentials)
class SummaryStyle(Enum):
CONCISE = "concise"
DETAILED = "detailed"
BULLET_POINTS = "bullet points"
NUMBERED_LIST = "numbered list"
class AITextSummarizerBlock(Block):
class Input(BlockSchema):
text: str = SchemaField(
description="The text to summarize.",
placeholder="Enter the text to summarize here...",
)
model: LlmModel = SchemaField(
title="LLM Model",
default=LlmModel.GPT4_TURBO,
description="The language model to use for summarizing the text.",
)
focus: str = SchemaField(
title="Focus",
default="general information",
description="The topic to focus on in the summary",
)
style: SummaryStyle = SchemaField(
title="Summary Style",
default=SummaryStyle.CONCISE,
description="The style of the summary to generate.",
)
credentials: AICredentials = AICredentialsField()
# TODO: Make this dynamic
max_tokens: int = SchemaField(
title="Max Tokens",
default=4096,
description="The maximum number of tokens to generate in the chat completion.",
ge=1,
)
chunk_overlap: int = SchemaField(
title="Chunk Overlap",
default=100,
description="The number of overlapping tokens between chunks to maintain context.",
ge=0,
)
ollama_host: str = SchemaField(
advanced=True,
default="localhost:11434",
description="Ollama host for local models",
)
class Output(BlockSchema):
summary: str = SchemaField(description="The final summary of the text.")
error: str = SchemaField(description="Error message if the API call failed.")
def __init__(self):
super().__init__(
id="a0a69be1-4528-491c-a85a-a4ab6873e3f0",
description="Utilize a Large Language Model (LLM) to summarize a long text.",
categories={BlockCategory.AI, BlockCategory.TEXT},
input_schema=AITextSummarizerBlock.Input,
output_schema=AITextSummarizerBlock.Output,
test_input={
"text": "Lorem ipsum..." * 100,
"credentials": TEST_CREDENTIALS_INPUT,
},
test_credentials=TEST_CREDENTIALS,
test_output=("summary", "Final summary of a long text"),
test_mock={
"llm_call": lambda input_data, credentials: (
{"final_summary": "Final summary of a long text"}
if "final_summary" in input_data.expected_format
else {"summary": "Summary of a chunk of text"}
)
},
)
def run(
self, input_data: Input, *, credentials: APIKeyCredentials, **kwargs
) -> BlockOutput:
for output in self._run(input_data, credentials):
yield output
def _run(self, input_data: Input, credentials: APIKeyCredentials) -> BlockOutput:
chunks = self._split_text(
input_data.text, input_data.max_tokens, input_data.chunk_overlap
)
summaries = []
for chunk in chunks:
chunk_summary = self._summarize_chunk(chunk, input_data, credentials)
summaries.append(chunk_summary)
final_summary = self._combine_summaries(summaries, input_data, credentials)
yield "summary", final_summary
@staticmethod
def _split_text(text: str, max_tokens: int, overlap: int) -> list[str]:
words = text.split()
chunks = []
chunk_size = max_tokens - overlap
for i in range(0, len(words), chunk_size):
chunk = " ".join(words[i : i + max_tokens])
chunks.append(chunk)
return chunks
def llm_call(
self,
input_data: AIStructuredResponseGeneratorBlock.Input,
credentials: APIKeyCredentials,
) -> dict:
block = AIStructuredResponseGeneratorBlock()
response = block.run_once(input_data, "response", credentials=credentials)
self.merge_stats(block.execution_stats)
return response
def _summarize_chunk(
self, chunk: str, input_data: Input, credentials: APIKeyCredentials
) -> str:
prompt = f"Summarize the following text in a {input_data.style} form. Focus your summary on the topic of `{input_data.focus}` if present, otherwise just provide a general summary:\n\n```{chunk}```"
llm_response = self.llm_call(
AIStructuredResponseGeneratorBlock.Input(
prompt=prompt,
credentials=input_data.credentials,
model=input_data.model,
expected_format={"summary": "The summary of the given text."},
),
credentials=credentials,
)
return llm_response["summary"]
def _combine_summaries(
self, summaries: list[str], input_data: Input, credentials: APIKeyCredentials
) -> str:
combined_text = "\n\n".join(summaries)
if len(combined_text.split()) <= input_data.max_tokens:
prompt = f"Provide a final summary of the following section summaries in a {input_data.style} form, focus your summary on the topic of `{input_data.focus}` if present:\n\n ```{combined_text}```\n\n Just respond with the final_summary in the format specified."
llm_response = self.llm_call(
AIStructuredResponseGeneratorBlock.Input(
prompt=prompt,
credentials=input_data.credentials,
model=input_data.model,
expected_format={
"final_summary": "The final summary of all provided summaries."
},
),
credentials=credentials,
)
return llm_response["final_summary"]
else:
# If combined summaries are still too long, recursively summarize
return self._run(
AITextSummarizerBlock.Input(
text=combined_text,
credentials=input_data.credentials,
model=input_data.model,
max_tokens=input_data.max_tokens,
chunk_overlap=input_data.chunk_overlap,
),
credentials=credentials,
).send(None)[
1
] # Get the first yielded value
class AIConversationBlock(Block):
class Input(BlockSchema):
messages: List[Message] = SchemaField(
description="List of messages in the conversation.", min_length=1
)
model: LlmModel = SchemaField(
title="LLM Model",
default=LlmModel.GPT4_TURBO,
description="The language model to use for the conversation.",
)
credentials: AICredentials = AICredentialsField()
max_tokens: int | None = SchemaField(
advanced=True,
default=None,
description="The maximum number of tokens to generate in the chat completion.",
)
ollama_host: str = SchemaField(
advanced=True,
default="localhost:11434",
description="Ollama host for local models",
)
class Output(BlockSchema):
response: str = SchemaField(
description="The model's response to the conversation."
)
error: str = SchemaField(description="Error message if the API call failed.")
def __init__(self):
super().__init__(
id="32a87eab-381e-4dd4-bdb8-4c47151be35a",
description="Advanced LLM call that takes a list of messages and sends them to the language model.",
categories={BlockCategory.AI},
input_schema=AIConversationBlock.Input,
output_schema=AIConversationBlock.Output,
test_input={
"messages": [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "Who won the world series in 2020?"},
{
"role": "assistant",
"content": "The Los Angeles Dodgers won the World Series in 2020.",
},
{"role": "user", "content": "Where was it played?"},
],
"model": LlmModel.GPT4_TURBO,
"credentials": TEST_CREDENTIALS_INPUT,
},
test_credentials=TEST_CREDENTIALS,
test_output=(
"response",
"The 2020 World Series was played at Globe Life Field in Arlington, Texas.",
),
test_mock={
"llm_call": lambda *args, **kwargs: "The 2020 World Series was played at Globe Life Field in Arlington, Texas."
},
)
def llm_call(
self,
input_data: AIStructuredResponseGeneratorBlock.Input,
credentials: APIKeyCredentials,
) -> str:
block = AIStructuredResponseGeneratorBlock()
response = block.run_once(input_data, "response", credentials=credentials)
self.merge_stats(block.execution_stats)
return response["response"]
def run(
self, input_data: Input, *, credentials: APIKeyCredentials, **kwargs
) -> BlockOutput:
response = self.llm_call(
AIStructuredResponseGeneratorBlock.Input(
prompt="",
credentials=input_data.credentials,
model=input_data.model,
conversation_history=input_data.messages,
max_tokens=input_data.max_tokens,
expected_format={},
),
credentials=credentials,
)
yield "response", response
class AIListGeneratorBlock(Block):
class Input(BlockSchema):
focus: str | None = SchemaField(
description="The focus of the list to generate.",
placeholder="The top 5 most interesting news stories in the data.",
default=None,
advanced=False,
)
source_data: str | None = SchemaField(
description="The data to generate the list from.",
placeholder="News Today: Humans land on Mars: Today humans landed on mars. -- AI wins Nobel Prize: AI wins Nobel Prize for solving world hunger. -- New AI Model: A new AI model has been released.",
default=None,
advanced=False,
)
model: LlmModel = SchemaField(
title="LLM Model",
default=LlmModel.GPT4_TURBO,
description="The language model to use for generating the list.",
advanced=True,
)
credentials: AICredentials = AICredentialsField()
max_retries: int = SchemaField(
default=3,
description="Maximum number of retries for generating a valid list.",
ge=1,
le=5,
)
max_tokens: int | None = SchemaField(
advanced=True,
default=None,
description="The maximum number of tokens to generate in the chat completion.",
)
ollama_host: str = SchemaField(
advanced=True,
default="localhost:11434",
description="Ollama host for local models",
)
class Output(BlockSchema):
generated_list: List[str] = SchemaField(description="The generated list.")
list_item: str = SchemaField(
description="Each individual item in the list.",
)
error: str = SchemaField(
description="Error message if the list generation failed."
)
def __init__(self):
super().__init__(
id="9c0b0450-d199-458b-a731-072189dd6593",
description="Generate a Python list based on the given prompt using a Large Language Model (LLM).",
categories={BlockCategory.AI, BlockCategory.TEXT},
input_schema=AIListGeneratorBlock.Input,
output_schema=AIListGeneratorBlock.Output,
test_input={
"focus": "planets",
"source_data": (
"Zylora Prime is a glowing jungle world with bioluminescent plants, "
"while Kharon-9 is a harsh desert planet with underground cities. "
"Vortexia's constant storms power floating cities, and Oceara is a water-covered world home to "
"intelligent marine life. On icy Draknos, ancient ruins lie buried beneath its frozen landscape, "
"drawing explorers to uncover its mysteries. Each planet showcases the limitless possibilities of "
"fictional worlds."
),
"model": LlmModel.GPT4_TURBO,
"credentials": TEST_CREDENTIALS_INPUT,
"max_retries": 3,
},
test_credentials=TEST_CREDENTIALS,
test_output=[
(
"generated_list",
["Zylora Prime", "Kharon-9", "Vortexia", "Oceara", "Draknos"],
),
("list_item", "Zylora Prime"),
("list_item", "Kharon-9"),
("list_item", "Vortexia"),
("list_item", "Oceara"),
("list_item", "Draknos"),
],
test_mock={
"llm_call": lambda input_data, credentials: {
"response": "['Zylora Prime', 'Kharon-9', 'Vortexia', 'Oceara', 'Draknos']"
},
},
)
@staticmethod
def llm_call(
input_data: AIStructuredResponseGeneratorBlock.Input,
credentials: APIKeyCredentials,
) -> dict[str, str]:
llm_block = AIStructuredResponseGeneratorBlock()
response = llm_block.run_once(input_data, "response", credentials=credentials)
return response
@staticmethod
def string_to_list(string):
"""
Converts a string representation of a list into an actual Python list object.
"""
logger.debug(f"Converting string to list. Input string: {string}")
try:
# Use ast.literal_eval to safely evaluate the string
python_list = ast.literal_eval(string)
if isinstance(python_list, list):
logger.debug(f"Successfully converted string to list: {python_list}")
return python_list
else:
logger.error(f"The provided string '{string}' is not a valid list")
raise ValueError(f"The provided string '{string}' is not a valid list.")
except (SyntaxError, ValueError) as e: