Skip to content

Cohere

CohereEmbedding #

Bases: BaseEmbedding

CohereEmbedding uses the Cohere API to generate embeddings for text.

Source code in llama-index-integrations/embeddings/llama-index-embeddings-cohere/llama_index/embeddings/cohere/base.py
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
class CohereEmbedding(BaseEmbedding):
    """CohereEmbedding uses the Cohere API to generate embeddings for text."""

    # Instance variables initialized via Pydantic's mechanism
    cohere_client: cohere.Client = Field(description="CohereAI client")
    cohere_async_client: cohere.AsyncClient = Field(description="CohereAI Async client")
    truncate: str = Field(description="Truncation type - START/ END/ NONE")
    input_type: Optional[str] = Field(
        description="Model Input type. If not provided, search_document and search_query are used when needed."
    )
    embedding_type: str = Field(
        description="Embedding type. If not provided float embedding_type is used when needed."
    )

    def __init__(
        self,
        cohere_api_key: Optional[str] = None,
        model_name: str = "embed-english-v3.0",
        truncate: str = "END",
        input_type: Optional[str] = None,
        embedding_type: str = "float",
        embed_batch_size: int = DEFAULT_EMBED_BATCH_SIZE,
        callback_manager: Optional[CallbackManager] = None,
        base_url: Optional[str] = None,
        timeout: Optional[float] = None,
        httpx_client: Optional[httpx.AsyncClient] = None,
    ):
        """
        A class representation for generating embeddings using the Cohere API.

        Args:
            cohere_client (Any): An instance of the Cohere client, which is used to communicate with the Cohere API.
            truncate (str): A string indicating the truncation strategy to be applied to input text. Possible values
                        are 'START', 'END', or 'NONE'.
            input_type (Optional[str]): An optional string that specifies the type of input provided to the model.
                                    This is model-dependent and could be one of the following: 'search_query',
                                    'search_document', 'classification', or 'clustering'.
            model_name (str): The name of the model to be used for generating embeddings. The class ensures that
                          this model is supported and that the input type provided is compatible with the model.
        """
        # Validate model_name and input_type
        if model_name not in VALID_MODEL_INPUT_TYPES:
            raise ValueError(f"{model_name} is not a valid model name")

        if input_type not in VALID_MODEL_INPUT_TYPES[model_name]:
            raise ValueError(
                f"{input_type} is not a valid input type for the provided model."
            )
        if embedding_type not in VALID_MODEL_EMBEDDING_TYPES[model_name]:
            raise ValueError(
                f"{embedding_type} is not a embedding type for the provided model."
            )

        if truncate not in VALID_TRUNCATE_OPTIONS:
            raise ValueError(f"truncate must be one of {VALID_TRUNCATE_OPTIONS}")

        super().__init__(
            cohere_client=cohere.Client(
                cohere_api_key,
                client_name="llama_index",
                base_url=base_url,
                timeout=timeout,
                httpx_client=httpx_client,
            ),
            cohere_async_client=cohere.AsyncClient(
                cohere_api_key,
                client_name="llama_index",
                base_url=base_url,
                timeout=timeout,
                httpx_client=httpx_client,
            ),
            cohere_api_key=cohere_api_key,
            model_name=model_name,
            input_type=input_type,
            embedding_type=embedding_type,
            truncate=truncate,
            embed_batch_size=embed_batch_size,
            callback_manager=callback_manager,
        )

    @classmethod
    def class_name(cls) -> str:
        return "CohereEmbedding"

    def _embed(self, texts: List[str], input_type: str) -> List[List[float]]:
        """Embed sentences using Cohere."""
        if self.model_name in V3_MODELS:
            result = self.cohere_client.embed(
                texts=texts,
                input_type=self.input_type or input_type,
                embedding_types=[self.embedding_type],
                model=self.model_name,
                truncate=self.truncate,
            ).embeddings
        else:
            result = self.cohere_client.embed(
                texts=texts,
                model=self.model_name,
                embedding_types=[self.embedding_type],
                truncate=self.truncate,
            ).embeddings
        return getattr(result, self.embedding_type, None)

    async def _aembed(self, texts: List[str], input_type: str) -> List[List[float]]:
        """Embed sentences using Cohere."""
        if self.model_name in V3_MODELS:
            result = (
                await self.cohere_async_client.embed(
                    texts=texts,
                    input_type=self.input_type or input_type,
                    embedding_types=[self.embedding_type],
                    model=self.model_name,
                    truncate=self.truncate,
                )
            ).embeddings
        else:
            result = (
                await self.cohere_async_client.embed(
                    texts=texts,
                    model=self.model_name,
                    embedding_types=[self.embedding_type],
                    truncate=self.truncate,
                )
            ).embeddings
        return getattr(result, self.embedding_type, None)

    def _get_query_embedding(self, query: str) -> List[float]:
        """Get query embedding. For query embeddings, input_type='search_query'."""
        return self._embed([query], input_type="search_query")[0]

    async def _aget_query_embedding(self, query: str) -> List[float]:
        """Get query embedding async. For query embeddings, input_type='search_query'."""
        return (await self._aembed([query], input_type="search_query"))[0]

    def _get_text_embedding(self, text: str) -> List[float]:
        """Get text embedding."""
        return self._embed([text], input_type="search_document")[0]

    async def _aget_text_embedding(self, text: str) -> List[float]:
        """Get text embedding async."""
        return (await self._aembed([text], input_type="search_document"))[0]

    def _get_text_embeddings(self, texts: List[str]) -> List[List[float]]:
        """Get text embeddings."""
        return self._embed(texts, input_type="search_document")

    async def _aget_text_embeddings(self, texts: List[str]) -> List[List[float]]:
        """Get text embeddings."""
        return await self._aembed(texts, input_type="search_document")