File size: 10,596 Bytes
e3278e4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import json
from typing import List, Literal, Optional, Union

import httpx

import litellm
from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
from litellm.llms.custom_httpx.http_handler import (
    AsyncHTTPHandler,
    HTTPHandler,
    get_async_httpx_client,
)
from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import (
    VertexAIError,
    VertexLLM,
)
from litellm.types.llms.vertex_ai import (
    Instance,
    InstanceImage,
    InstanceVideo,
    MultimodalPredictions,
    VertexMultimodalEmbeddingRequest,
)
from litellm.types.utils import Embedding, EmbeddingResponse
from litellm.utils import is_base64_encoded


class VertexMultimodalEmbedding(VertexLLM):
    def __init__(self) -> None:
        super().__init__()
        self.SUPPORTED_MULTIMODAL_EMBEDDING_MODELS = [
            "multimodalembedding",
            "multimodalembedding@001",
        ]

    def multimodal_embedding(
        self,
        model: str,
        input: Union[list, str],
        print_verbose,
        model_response: EmbeddingResponse,
        custom_llm_provider: Literal["gemini", "vertex_ai"],
        optional_params: dict,
        logging_obj: LiteLLMLoggingObj,
        api_key: Optional[str] = None,
        api_base: Optional[str] = None,
        encoding=None,
        vertex_project=None,
        vertex_location=None,
        vertex_credentials=None,
        aembedding=False,
        timeout=300,
        client=None,
    ) -> EmbeddingResponse:

        _auth_header, vertex_project = self._ensure_access_token(
            credentials=vertex_credentials,
            project_id=vertex_project,
            custom_llm_provider=custom_llm_provider,
        )

        auth_header, url = self._get_token_and_url(
            model=model,
            auth_header=_auth_header,
            gemini_api_key=api_key,
            vertex_project=vertex_project,
            vertex_location=vertex_location,
            vertex_credentials=vertex_credentials,
            stream=None,
            custom_llm_provider=custom_llm_provider,
            api_base=api_base,
            should_use_v1beta1_features=False,
            mode="embedding",
        )

        if client is None:
            _params = {}
            if timeout is not None:
                if isinstance(timeout, float) or isinstance(timeout, int):
                    _httpx_timeout = httpx.Timeout(timeout)
                    _params["timeout"] = _httpx_timeout
            else:
                _params["timeout"] = httpx.Timeout(timeout=600.0, connect=5.0)

            sync_handler: HTTPHandler = HTTPHandler(**_params)  # type: ignore
        else:
            sync_handler = client  # type: ignore

        optional_params = optional_params or {}

        request_data = VertexMultimodalEmbeddingRequest()

        if "instances" in optional_params:
            request_data["instances"] = optional_params["instances"]
        elif isinstance(input, list):
            vertex_instances: List[Instance] = self.process_openai_embedding_input(
                _input=input
            )
            request_data["instances"] = vertex_instances

        else:
            # construct instances
            vertex_request_instance = Instance(**optional_params)

            if isinstance(input, str):
                vertex_request_instance = self._process_input_element(input)

            request_data["instances"] = [vertex_request_instance]

        headers = {
            "Content-Type": "application/json; charset=utf-8",
            "Authorization": f"Bearer {auth_header}",
        }

        ## LOGGING
        logging_obj.pre_call(
            input=input,
            api_key="",
            additional_args={
                "complete_input_dict": request_data,
                "api_base": url,
                "headers": headers,
            },
        )

        if aembedding is True:
            return self.async_multimodal_embedding(  # type: ignore
                model=model,
                api_base=url,
                data=request_data,
                timeout=timeout,
                headers=headers,
                client=client,
                model_response=model_response,
            )

        response = sync_handler.post(
            url=url,
            headers=headers,
            data=json.dumps(request_data),
        )

        if response.status_code != 200:
            raise Exception(f"Error: {response.status_code} {response.text}")

        _json_response = response.json()
        if "predictions" not in _json_response:
            raise litellm.InternalServerError(
                message=f"embedding response does not contain 'predictions', got {_json_response}",
                llm_provider="vertex_ai",
                model=model,
            )
        _predictions = _json_response["predictions"]
        vertex_predictions = MultimodalPredictions(predictions=_predictions)
        model_response.data = self.transform_embedding_response_to_openai(
            predictions=vertex_predictions
        )
        model_response.model = model

        return model_response

    async def async_multimodal_embedding(
        self,
        model: str,
        api_base: str,
        data: VertexMultimodalEmbeddingRequest,
        model_response: litellm.EmbeddingResponse,
        timeout: Optional[Union[float, httpx.Timeout]],
        headers={},
        client: Optional[AsyncHTTPHandler] = None,
    ) -> litellm.EmbeddingResponse:
        if client is None:
            _params = {}
            if timeout is not None:
                if isinstance(timeout, float) or isinstance(timeout, int):
                    timeout = httpx.Timeout(timeout)
                _params["timeout"] = timeout
            client = get_async_httpx_client(
                llm_provider=litellm.LlmProviders.VERTEX_AI,
                params={"timeout": timeout},
            )
        else:
            client = client  # type: ignore

        try:
            response = await client.post(api_base, headers=headers, json=data)  # type: ignore
            response.raise_for_status()
        except httpx.HTTPStatusError as err:
            error_code = err.response.status_code
            raise VertexAIError(status_code=error_code, message=err.response.text)
        except httpx.TimeoutException:
            raise VertexAIError(status_code=408, message="Timeout error occurred.")

        _json_response = response.json()
        if "predictions" not in _json_response:
            raise litellm.InternalServerError(
                message=f"embedding response does not contain 'predictions', got {_json_response}",
                llm_provider="vertex_ai",
                model=model,
            )
        _predictions = _json_response["predictions"]

        vertex_predictions = MultimodalPredictions(predictions=_predictions)
        model_response.data = self.transform_embedding_response_to_openai(
            predictions=vertex_predictions
        )
        model_response.model = model

        return model_response

    def _process_input_element(self, input_element: str) -> Instance:
        """
        Process the input element for multimodal embedding requests. checks if the if the input is gcs uri, base64 encoded image or plain text.

        Args:
            input_element (str): The input element to process.

        Returns:
            Dict[str, Any]: A dictionary representing the processed input element.
        """
        if len(input_element) == 0:
            return Instance(text=input_element)
        elif "gs://" in input_element:
            if "mp4" in input_element:
                return Instance(video=InstanceVideo(gcsUri=input_element))
            else:
                return Instance(image=InstanceImage(gcsUri=input_element))
        elif is_base64_encoded(s=input_element):
            return Instance(image=InstanceImage(bytesBase64Encoded=input_element))
        else:
            return Instance(text=input_element)

    def process_openai_embedding_input(
        self, _input: Union[list, str]
    ) -> List[Instance]:
        """
        Process the input for multimodal embedding requests.

        Args:
            _input (Union[list, str]): The input data to process.

        Returns:
            List[Instance]: A list of processed VertexAI Instance objects.
        """

        _input_list = None
        if not isinstance(_input, list):
            _input_list = [_input]
        else:
            _input_list = _input

        processed_instances = []
        for element in _input_list:
            if isinstance(element, str):
                instance = Instance(**self._process_input_element(element))
            elif isinstance(element, dict):
                instance = Instance(**element)
            else:
                raise ValueError(f"Unsupported input type: {type(element)}")
            processed_instances.append(instance)

        return processed_instances

    def transform_embedding_response_to_openai(
        self, predictions: MultimodalPredictions
    ) -> List[Embedding]:

        openai_embeddings: List[Embedding] = []
        if "predictions" in predictions:
            for idx, _prediction in enumerate(predictions["predictions"]):
                if _prediction:
                    if "textEmbedding" in _prediction:
                        openai_embedding_object = Embedding(
                            embedding=_prediction["textEmbedding"],
                            index=idx,
                            object="embedding",
                        )
                        openai_embeddings.append(openai_embedding_object)
                    elif "imageEmbedding" in _prediction:
                        openai_embedding_object = Embedding(
                            embedding=_prediction["imageEmbedding"],
                            index=idx,
                            object="embedding",
                        )
                        openai_embeddings.append(openai_embedding_object)
                    elif "videoEmbeddings" in _prediction:
                        for video_embedding in _prediction["videoEmbeddings"]:
                            openai_embedding_object = Embedding(
                                embedding=video_embedding["embedding"],
                                index=idx,
                                object="embedding",
                            )
                            openai_embeddings.append(openai_embedding_object)
        return openai_embeddings