Repository navigation
Expand file tree
/
Copy path_transformers.py
More file actions
1395 lines (1194 loc) · 42.9 KB
/
Copy path_transformers.py
File metadata and controls
1395 lines (1194 loc) · 42.9 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
# Copyright 2025 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
"""Transformers for Google GenAI SDK."""
import base64
from collections.abc import Iterable, Mapping
from enum import Enum, EnumMeta
import inspect
import io
import logging
import re
import sys
import time
import types as builtin_types
import typing
from typing import Any, GenericAlias, List, Optional, Sequence, Union # type: ignore[attr-defined]
from ._mcp_utils import mcp_to_gemini_tool
from ._common import get_value_by_path as getv
from ._common import is_duck_type_of
if typing.TYPE_CHECKING:
import PIL.Image
import pydantic
from . import _api_client
from . import _common
from . import types
logger = logging.getLogger('google_genai._transformers')
if sys.version_info >= (3, 10):
VersionedUnionType = builtin_types.UnionType
_UNION_TYPES = (typing.Union, builtin_types.UnionType)
from typing import TypeGuard
else:
VersionedUnionType = typing._UnionGenericAlias # type: ignore[attr-defined]
_UNION_TYPES = (typing.Union,)
from typing_extensions import TypeGuard
if typing.TYPE_CHECKING:
from mcp import ClientSession as McpClientSession
from mcp.types import Tool as McpTool
else:
McpClientSession: typing.Type = Any
McpTool: typing.Type = Any
try:
from mcp import ClientSession as McpClientSession
from mcp.types import Tool as McpTool
except ImportError:
McpClientSession = None
McpTool = None
metric_name_sdk_api_map = {
'exact_match': 'exactMatchSpec',
'bleu': 'bleuSpec',
'rouge_spec': 'rougeSpec',
}
metric_name_api_sdk_map = {v: k for k, v in metric_name_sdk_api_map.items()}
def _resource_name(
client: _api_client.BaseApiClient,
resource_name: str,
*,
collection_identifier: str,
collection_hierarchy_depth: int = 2,
) -> str:
# pylint: disable=line-too-long
"""Prepends resource name with project, location, collection_identifier if needed.
The collection_identifier will only be prepended if it's not present
and the prepending won't violate the collection hierarchy depth.
When the prepending condition doesn't meet, returns the input
resource_name.
Args:
client: The API client.
resource_name: The user input resource name to be completed.
collection_identifier: The collection identifier to be prepended. See
collection identifiers in https://google.aip.dev/122.
collection_hierarchy_depth: The collection hierarchy depth. Only set this
field when the resource has nested collections. For example,
`users/vhugo1802/events/birthday-dinner-226`, the collection_identifier is
`users` and collection_hierarchy_depth is 4. See nested collections in
https://google.aip.dev/122.
Example:
resource_name = 'cachedContents/123'
client.vertexai = True
client.project = 'bar'
client.location = 'us-west1'
_resource_name(client, 'cachedContents/123',
collection_identifier='cachedContents')
returns: 'projects/bar/locations/us-west1/cachedContents/123'
Example:
resource_name = 'projects/foo/locations/us-central1/cachedContents/123'
# resource_name = 'locations/us-central1/cachedContents/123'
client.vertexai = True
client.project = 'bar'
client.location = 'us-west1'
_resource_name(client, resource_name,
collection_identifier='cachedContents')
returns: 'projects/foo/locations/us-central1/cachedContents/123'
Example:
resource_name = '123'
# resource_name = 'cachedContents/123'
client.vertexai = False
_resource_name(client, resource_name,
collection_identifier='cachedContents')
returns 'cachedContents/123'
Example:
resource_name = 'some/wrong/cachedContents/resource/name/123'
resource_prefix = 'cachedContents'
client.vertexai = False
# client.vertexai = True
_resource_name(client, resource_name,
collection_identifier='cachedContents')
returns: 'some/wrong/cachedContents/resource/name/123'
Returns:
The completed resource name.
"""
should_prepend_collection_identifier = (
not resource_name.startswith(f'{collection_identifier}/')
# Check if prepending the collection identifier won't violate the
# collection hierarchy depth.
and f'{collection_identifier}/{resource_name}'.count('/') + 1
== collection_hierarchy_depth
)
if client.vertexai:
if resource_name.startswith('projects/'):
return resource_name
elif resource_name.startswith('locations/'):
return f'projects/{client.project}/{resource_name}'
elif resource_name.startswith(f'{collection_identifier}/'):
return f'projects/{client.project}/locations/{client.location}/{resource_name}'
elif should_prepend_collection_identifier:
return f'projects/{client.project}/locations/{client.location}/{collection_identifier}/{resource_name}'
else:
return resource_name
else:
if should_prepend_collection_identifier:
return f'{collection_identifier}/{resource_name}'
else:
return resource_name
def t_model(client: _api_client.BaseApiClient, model: str) -> str:
if not model:
raise ValueError('model is required.')
if '..' in model or '?' in model or '&' in model:
raise ValueError('invalid model parameter.')
if client.vertexai:
if (
model.startswith('projects/')
or model.startswith('models/')
or model.startswith('publishers/')
):
return model
elif '/' in model:
publisher, model_id = model.split('/', 1)
return f'publishers/{publisher}/models/{model_id}'
else:
return f'publishers/google/models/{model}'
else:
if model.startswith('models/'):
return model
elif model.startswith('tunedModels/'):
return model
else:
return f'models/{model}'
def t_models_url(
api_client: _api_client.BaseApiClient, base_models: bool
) -> str:
if api_client.vertexai:
if base_models:
return 'publishers/google/models'
else:
return 'models'
else:
if base_models:
return 'models'
else:
return 'tunedModels'
def t_extract_models(
response: _common.StringDict,
) -> list[_common.StringDict]:
if not response:
return []
models: Optional[list[_common.StringDict]] = response.get('models')
if models is not None:
return models
tuned_models: Optional[list[_common.StringDict]] = response.get('tunedModels')
if tuned_models is not None:
return tuned_models
publisher_models: Optional[list[_common.StringDict]] = response.get(
'publisherModels'
)
if publisher_models is not None:
return publisher_models
if (
response.get('httpHeaders') is not None
and response.get('jsonPayload') is None
):
return []
else:
logger.warning('Cannot determine the models type.')
logger.debug('Cannot determine the models type for response: %s', response)
return []
def t_caches_model(
api_client: _api_client.BaseApiClient, model: str
) -> Optional[str]:
model = t_model(api_client, model)
if not model:
return None
if model.startswith('publishers/') and api_client.vertexai:
# vertex caches only support model name start with projects.
return (
f'projects/{api_client.project}/locations/{api_client.location}/{model}'
)
elif model.startswith('models/') and api_client.vertexai:
return f'projects/{api_client.project}/locations/{api_client.location}/publishers/google/{model}'
else:
return model
def pil_to_blob(image: Any) -> types.Blob:
image_format = 'PNG'
save_params: dict[str, Any] = dict()
if (
image.format == 'JPEG'
and getattr(image, 'filename', '')
and image.mode in ['1', 'L', 'RGB', 'RGBX', 'CMYK']
):
image_format = 'JPEG'
save_params.update(quality='keep')
image_io = io.BytesIO()
image.save(image_io, image_format, **save_params)
image_bytes = image_io.getvalue()
mime_type = f'image/{image_format.lower()}'
return types.Blob(data=image_bytes, mime_type=mime_type)
def t_function_response(
function_response: types.FunctionResponseOrDict,
) -> types.FunctionResponse:
if not function_response:
raise ValueError('function_response is required.')
if isinstance(function_response, dict):
return types.FunctionResponse.model_validate(function_response)
elif is_duck_type_of(function_response, types.FunctionResponse):
return function_response
else:
raise TypeError(
'Could not parse input as FunctionResponse. Unsupported'
f' function_response type: {type(function_response)}'
)
def t_function_responses(
function_responses: Union[
types.FunctionResponseOrDict,
Sequence[types.FunctionResponseOrDict],
],
) -> list[types.FunctionResponse]:
if not function_responses:
raise ValueError('function_responses are required.')
if isinstance(function_responses, Sequence):
return [t_function_response(response) for response in function_responses]
else:
return [t_function_response(function_responses)]
def t_blobs(
blobs: Union[types.BlobImageUnionDict, list[types.BlobImageUnionDict]],
) -> list[types.Blob]:
if isinstance(blobs, list):
return [t_blob(blob) for blob in blobs]
else:
return [t_blob(blobs)]
def t_blob(blob: types.BlobImageUnionDict) -> types.Blob:
if not blob:
raise ValueError('blob is required.')
if is_duck_type_of(blob, types.Blob):
return blob # type: ignore[return-value]
if isinstance(blob, dict):
return types.Blob.model_validate(blob)
if 'image' in blob.__class__.__name__.lower():
try:
import PIL.Image
PIL_Image = PIL.Image.Image
except ImportError:
PIL_Image = None
if PIL_Image is not None and isinstance(blob, PIL_Image):
return pil_to_blob(blob)
raise TypeError(
f'Could not parse input as Blob. Unsupported blob type: {type(blob)}'
)
def t_image_blob(blob: types.BlobImageUnionDict) -> types.Blob:
blob = t_blob(blob)
if blob.mime_type and blob.mime_type.startswith('image/'):
return blob
raise ValueError(f'Unsupported mime type: {blob.mime_type!r}')
def t_audio_blob(blob: types.BlobOrDict) -> types.Blob:
blob = t_blob(blob)
if blob.mime_type and blob.mime_type.startswith('audio/'):
return blob
raise ValueError(f'Unsupported mime type: {blob.mime_type!r}')
def t_part(part: Optional[types.PartUnionDict]) -> types.Part:
if part is None:
raise ValueError('content part is required.')
if isinstance(part, str):
return types.Part(text=part)
if is_duck_type_of(part, types.File):
if not part.uri or not part.mime_type: # type: ignore[union-attr]
raise ValueError('file uri and mime_type are required.')
return types.Part.from_uri(file_uri=part.uri, mime_type=part.mime_type) # type: ignore[union-attr]
if isinstance(part, dict):
try:
return types.Part.model_validate(part)
except pydantic.ValidationError:
return types.Part(file_data=types.FileData.model_validate(part))
if is_duck_type_of(part, types.Part):
return part # type: ignore[return-value]
if 'image' in part.__class__.__name__.lower():
try:
import PIL.Image
PIL_Image = PIL.Image.Image
except ImportError:
PIL_Image = None
if PIL_Image is not None and isinstance(part, PIL_Image):
return types.Part(inline_data=pil_to_blob(part))
raise ValueError(f'Unsupported content part type: {type(part)}')
def t_parts(
parts: Optional[
Union[list[types.PartUnionDict], types.PartUnionDict, list[types.Part]]
],
) -> list[types.Part]:
#
if parts is None or (isinstance(parts, list) and not parts):
raise ValueError('content parts are required.')
if isinstance(parts, list):
return [t_part(part) for part in parts]
else:
return [t_part(parts)]
def t_image_predictions(
predictions: Optional[Iterable[Mapping[str, Any]]],
) -> Optional[list[types.GeneratedImage]]:
if not predictions:
return None
images = []
for prediction in predictions:
if prediction.get('image'):
images.append(
types.GeneratedImage(
image=types.Image(
gcs_uri=prediction['image']['gcsUri'],
image_bytes=prediction['image']['imageBytes'],
)
)
)
return images
ContentType = Union[types.Content, types.ContentDict, types.PartUnionDict]
def t_content(
content: Union[ContentType, types.ContentDict, None],
) -> types.Content:
if content is None:
raise ValueError('content is required.')
if is_duck_type_of(content, types.Content):
return content # type: ignore[return-value]
if isinstance(content, dict):
try:
return types.Content.model_validate(content)
except pydantic.ValidationError:
possible_part = t_part(content) # type: ignore[arg-type]
return (
types.ModelContent(parts=[possible_part])
if possible_part.function_call
else types.UserContent(parts=[possible_part])
)
if is_duck_type_of(content, types.File):
return types.UserContent(parts=[t_part(content)]) # type: ignore[arg-type]
if is_duck_type_of(content, types.Part):
return (
types.ModelContent(parts=[content]) # type: ignore[arg-type]
if content.function_call # type: ignore[union-attr]
else types.UserContent(parts=[content]) # type: ignore[arg-type]
)
return types.UserContent(parts=content) # type: ignore[arg-type]
def t_contents_for_embed(
client: _api_client.BaseApiClient,
contents: Union[list[types.Content], list[types.ContentDict], ContentType],
) -> Union[list[str], list[types.Content]]:
if isinstance(contents, list):
transformed_contents = [t_content(content) for content in contents]
else:
transformed_contents = [t_content(contents)]
if client.vertexai:
text_parts = []
for content in transformed_contents:
if content is not None:
if isinstance(content, dict):
content = types.Content.model_validate(content)
if content.parts is not None:
for part in content.parts:
if part.text:
text_parts.append(part.text)
else:
logger.warning(f'Non-text part found, only returning text parts.')
return text_parts
else:
return transformed_contents
def t_contents(
contents: Optional[
Union[types.ContentListUnion, types.ContentListUnionDict, types.Content]
],
) -> list[types.Content]:
if contents is None or (isinstance(contents, list) and not contents):
raise ValueError('contents are required.')
if not isinstance(contents, list):
return [t_content(contents)]
result: list[types.Content] = []
accumulated_parts: list[types.Part] = []
def _is_part(
part: Union[types.PartUnionDict, Any],
) -> TypeGuard[types.PartUnionDict]:
if (
isinstance(part, str)
or is_duck_type_of(part, types.File)
or is_duck_type_of(part, types.Part)
):
return True
if isinstance(part, dict):
if not part:
# Empty dict should be considered as Content, not Part.
return False
try:
types.Part.model_validate(part)
return True
except pydantic.ValidationError:
try:
types.FileData.model_validate(part)
return True
except pydantic.ValidationError:
return False
if 'image' in part.__class__.__name__.lower():
try:
import PIL.Image
PIL_Image = PIL.Image.Image
except ImportError:
PIL_Image = None
if PIL_Image is not None and isinstance(part, PIL_Image):
return True
return False
def _is_user_part(part: types.Part) -> bool:
return not part.function_call
def _are_user_parts(parts: list[types.Part]) -> bool:
return all(_is_user_part(part) for part in parts)
def _append_accumulated_parts_as_content(
result: list[types.Content],
accumulated_parts: list[types.Part],
) -> None:
if not accumulated_parts:
return
result.append(
types.UserContent(parts=accumulated_parts)
if _are_user_parts(accumulated_parts)
else types.ModelContent(parts=accumulated_parts)
)
accumulated_parts[:] = []
def _handle_current_part(
result: list[types.Content],
accumulated_parts: list[types.Part],
current_part: types.PartUnionDict,
) -> None:
current_part = t_part(current_part)
if _is_user_part(current_part) == _are_user_parts(accumulated_parts):
accumulated_parts.append(current_part)
else:
_append_accumulated_parts_as_content(result, accumulated_parts)
accumulated_parts[:] = [current_part]
# iterator over contents
# if content type or content dict, append to result
# if consecutive part(s),
# group consecutive user part(s) to a UserContent
# group consecutive model part(s) to a ModelContent
# append to result
# if list, we only accept a list of types.PartUnion
for content in contents:
if is_duck_type_of(content, types.Content) or isinstance(content, list):
_append_accumulated_parts_as_content(result, accumulated_parts)
if isinstance(content, list):
result.append(types.UserContent(parts=content)) # type: ignore[arg-type]
else:
result.append(content) # type: ignore[arg-type]
elif _is_part(content):
_handle_current_part(result, accumulated_parts, content)
elif isinstance(content, dict):
# PactDict is already handled in _is_part
result.append(types.Content.model_validate(content))
else:
raise ValueError(f'Unsupported content type: {type(content)}')
_append_accumulated_parts_as_content(result, accumulated_parts)
return result
def handle_null_fields(schema: _common.StringDict) -> None:
"""Process null fields in the schema so it is compatible with OpenAPI.
The OpenAPI spec does not support 'type: 'null' in the schema. This function
handles this case by adding 'nullable: True' to the null field and removing
the {'type': 'null'} entry.
https://swagger.io/docs/specification/v3_0/data-models/data-types/#null
Example of schema properties before and after handling null fields:
Before:
{
"name": {
"title": "Name",
"type": "string"
},
"total_area_sq_mi": {
"anyOf": [
{
"type": "integer"
},
{
"type": "null"
}
],
"default": None,
"title": "Total Area Sq Mi"
}
}
After:
{
"name": {
"title": "Name",
"type": "string"
},
"total_area_sq_mi": {
"type": "integer",
"nullable": true,
"default": None,
"title": "Total Area Sq Mi"
}
}
"""
if schema.get('type', None) == 'null':
schema['nullable'] = True
del schema['type']
elif 'anyOf' in schema:
for item in schema['anyOf']:
if 'type' in item and item['type'] == 'null':
schema['nullable'] = True
schema['anyOf'].remove({'type': 'null'})
if len(schema['anyOf']) == 1:
# If there is only one type left after removing null, remove the anyOf field.
for key, val in schema['anyOf'][0].items():
schema[key] = val
del schema['anyOf']
def _raise_for_unsupported_schema_type(origin: Any) -> None:
"""Raises an error if the schema type is unsupported."""
raise ValueError(f'Unsupported schema type: {origin}')
def _raise_for_unsupported_mldev_properties(
schema: Any, client: Optional[_api_client.BaseApiClient]
) -> None:
if (
client
and not client.vertexai
and (
schema.get('additionalProperties')
or schema.get('additional_properties')
)
):
raise ValueError('additionalProperties is only supported in Gemini Enterprise Agent Platform mode, , not in Gemini Developer API mode.')
def process_schema(
schema: _common.StringDict,
client: Optional[_api_client.BaseApiClient],
defs: Optional[_common.StringDict] = None,
*,
order_properties: bool = True,
visited_dicts_path: Optional[set[int]] = None,
) -> None:
"""Updates the schema and each sub-schema inplace to be API-compatible.
- Inlines the $defs.
Example of a schema before and after (with mldev):
Before:
`schema`
{
'items': {
'$ref': '#/$defs/CountryInfo'
},
'title': 'Placeholder',
'type': 'array'
}
`defs`
{
'CountryInfo': {
'properties': {
'continent': {
'title': 'Continent',
'type': 'string'
},
'gdp': {
'title': 'Gdp',
'type': 'integer'}
},
}
'required':['continent', 'gdp'],
'title': 'CountryInfo',
'type': 'object'
}
}
After:
`schema`
{
'items': {
'properties': {
'continent': {
'title': 'Continent',
'type': 'string'
},
'gdp': {
'title': 'Gdp',
'type': 'integer'
},
}
'required':['continent', 'gdp'],
'title': 'CountryInfo',
'type': 'object'
},
'type': 'array'
}
"""
if visited_dicts_path is None:
visited_dicts_path = set()
if id(schema) in visited_dicts_path:
return
visited_dicts_path.add(id(schema))
if schema.get('title') == 'PlaceholderLiteralEnum':
del schema['title']
_raise_for_unsupported_mldev_properties(schema, client)
# Standardize spelling for relevant schema fields. For example, if a dict is
# provided directly to response_schema, it may use `any_of` instead of `anyOf.
# Otherwise, model_json_schema() uses `anyOf`.
for from_name, to_name in [
('additional_properties', 'additionalProperties'),
('any_of', 'anyOf'),
('prefix_items', 'prefixItems'),
('property_ordering', 'propertyOrdering'),
]:
if (value := schema.pop(from_name, None)) is not None:
schema[to_name] = value
if defs is None:
defs = schema.pop('$defs', {})
for _, sub_schema in defs.items():
# We can skip the '$ref' check, because JSON schema forbids a '$ref' from
# directly referencing another '$ref':
# https://json-schema.org/understanding-json-schema/structuring#recursion
process_schema(
sub_schema,
client,
defs,
order_properties=order_properties,
visited_dicts_path=visited_dicts_path,
)
handle_null_fields(schema)
# After removing null fields, Optional fields with only one possible type
# will have a $ref key that needs to be flattened
# For example: {'default': None, 'description': 'Name of the person', 'nullable': True, '$ref': '#/$defs/TestPerson'}
if (ref := schema.pop('$ref', None)) is not None:
schema.update(defs[ref.split('defs/')[-1]])
def _recurse(sub_schema: _common.StringDict) -> _common.StringDict:
"""Returns the processed `sub_schema`, resolving its '$ref' if any."""
if (ref := sub_schema.pop('$ref', None)) is not None:
sub_schema = defs[ref.split('defs/')[-1]]
if id(sub_schema) in visited_dicts_path:
return {}
process_schema(
sub_schema,
client,
defs,
order_properties=order_properties,
visited_dicts_path=visited_dicts_path,
)
return sub_schema
if (any_of := schema.get('anyOf')) is not None:
schema['anyOf'] = [_recurse(sub_schema) for sub_schema in any_of]
visited_dicts_path.remove(id(schema))
return
schema_type = schema.get('type')
if isinstance(schema_type, Enum):
schema_type = schema_type.value
if isinstance(schema_type, str):
schema_type = schema_type.upper()
# model_json_schema() returns a schema with a 'const' field when a Literal with one value is provided as a pydantic field
# For example `genre: Literal['action']` becomes: {'const': 'action', 'title': 'Genre', 'type': 'string'}
const = schema.get('const')
if const is not None:
if schema_type == 'STRING':
schema['enum'] = [const]
del schema['const']
else:
raise ValueError('Literal values must be strings.')
if schema_type == 'OBJECT':
if (properties := schema.get('properties')) is not None:
for name, sub_schema in list(properties.items()):
properties[name] = _recurse(sub_schema)
if (
len(properties.items()) > 1
and order_properties
and 'propertyOrdering' not in schema
):
schema['property_ordering'] = list(properties.keys())
if (additional := schema.get('additionalProperties')) is not None:
# It is legal to set 'additionalProperties' to a bool:
# https://json-schema.org/understanding-json-schema/reference/object#additionalproperties
if isinstance(additional, dict):
schema['additionalProperties'] = _recurse(additional)
elif schema_type == 'ARRAY':
if (items := schema.get('items')) is not None:
schema['items'] = _recurse(items)
if (prefixes := schema.get('prefixItems')) is not None:
schema['prefixItems'] = [_recurse(prefix) for prefix in prefixes]
visited_dicts_path.remove(id(schema))
def _process_enum(
enum: EnumMeta, client: Optional[_api_client.BaseApiClient]
) -> types.Schema:
is_integer_enum = False
for member in enum: # type: ignore
if isinstance(member.value, int):
is_integer_enum = True
elif not isinstance(member.value, str):
raise TypeError(
f'Enum member {member.name} value must be a string or integer, got'
f' {type(member.value)}'
)
enum_to_process = enum
if is_integer_enum:
str_members = [str(member.value) for member in enum] # type: ignore
str_enum = Enum(enum.__name__, str_members, type=str) # type: ignore
enum_to_process = str_enum
class Placeholder(pydantic.BaseModel):
placeholder: enum_to_process # type: ignore[valid-type]
enum_schema = Placeholder.model_json_schema()
process_schema(enum_schema, client)
enum_schema = enum_schema['properties']['placeholder']
return types.Schema.model_validate(enum_schema)
def _is_type_dict_str_any(
origin: Union[types.SchemaUnionDict, Any],
) -> TypeGuard[_common.StringDict]:
"""Verifies the schema is of type dict[str, Any] for mypy type checking."""
return isinstance(origin, dict) and all(
isinstance(key, str) for key in origin
)
def t_schema(
client: Optional[_api_client.BaseApiClient],
origin: Union[types.SchemaUnionDict, Any],
) -> Optional[types.Schema]:
if not origin:
return None
if isinstance(origin, dict) and _is_type_dict_str_any(origin):
process_schema(origin, client)
return types.Schema.model_validate(origin)
if isinstance(origin, EnumMeta):
return _process_enum(origin, client)
if is_duck_type_of(origin, types.Schema):
if dict(origin) == dict(types.Schema()): # type: ignore [arg-type]
# response_schema value was coerced to an empty Schema instance because
# it did not adhere to the Schema field annotation
_raise_for_unsupported_schema_type(origin)
schema = origin.model_dump(exclude_unset=True) # type: ignore[union-attr]
process_schema(schema, client)
return types.Schema.model_validate(schema)
if (
# in Python 3.9 Generic alias list[int] counts as a type,
# and breaks issubclass because it's not a class.
not isinstance(origin, GenericAlias)
and isinstance(origin, type)
and issubclass(origin, pydantic.BaseModel)
):
schema = origin.model_json_schema()
process_schema(schema, client)
return types.Schema.model_validate(schema)
elif (
isinstance(origin, GenericAlias)
or isinstance(origin, type)
or isinstance(origin, VersionedUnionType)
or typing.get_origin(origin) in _UNION_TYPES
):
class Placeholder(pydantic.BaseModel):
placeholder: origin # type: ignore[valid-type]
schema = Placeholder.model_json_schema()
process_schema(schema, client)
schema = schema['properties']['placeholder']
return types.Schema.model_validate(schema)
raise ValueError(f'Unsupported schema type: {origin}')
def t_speech_config(
origin: Union[types.SpeechConfigUnionDict, Any],
) -> Optional[types.SpeechConfig]:
if not origin:
return None
if is_duck_type_of(origin, types.SpeechConfig):
return origin # type: ignore[return-value]
if isinstance(origin, str):
return types.SpeechConfig(
voice_config=types.VoiceConfig(
prebuilt_voice_config=types.PrebuiltVoiceConfig(voice_name=origin)
)
)
if isinstance(origin, dict):
return types.SpeechConfig.model_validate(origin)
raise ValueError(f'Unsupported speechConfig type: {type(origin)}')
def t_live_speech_config(
origin: types.SpeechConfigOrDict,
) -> Optional[types.SpeechConfig]:
if is_duck_type_of(origin, types.SpeechConfig):
speech_config = origin
if isinstance(origin, dict):
speech_config = types.SpeechConfig.model_validate(origin)
if speech_config.multi_speaker_voice_config is not None: # type: ignore[union-attr]
raise ValueError(
'multi_speaker_voice_config is not supported in the live API.'
)
return speech_config # type: ignore[return-value]
def t_tool(
client: _api_client.BaseApiClient, origin: Any
) -> Optional[Union[types.Tool, Any]]:
if not origin:
return None
if inspect.isfunction(origin) or inspect.ismethod(origin):
return types.Tool(
function_declarations=[
types.FunctionDeclaration.from_callable(
client=client, callable=origin
)
]
)
elif McpTool is not None and is_duck_type_of(origin, McpTool):
return mcp_to_gemini_tool(origin)
elif isinstance(origin, dict):
return types.Tool.model_validate(origin)
else:
return origin
def t_tools(
client: _api_client.BaseApiClient, origin: list[Any]
) -> list[types.Tool]:
if not origin:
return []
function_tool = types.Tool(function_declarations=[])
tools = []
for tool in origin:
transformed_tool = t_tool(client, tool)
# All functions should be merged into one tool.
if transformed_tool is not None:
if (
transformed_tool.function_declarations
and function_tool.function_declarations is not None
):
function_tool.function_declarations += (
transformed_tool.function_declarations
)
tool_copy = transformed_tool.model_copy()
tool_copy.function_declarations = None
if tool_copy.model_dump(exclude_none=True):
tools.append(tool_copy)
else: