|
| 1 | +""" |
| 2 | +This module is used to transform the request and response for the Voyage contextualized embeddings API. |
| 3 | +This would be used for all the contextualized embeddings models in Voyage. |
| 4 | +""" |
| 5 | +from typing import List, Optional, Union |
| 6 | + |
| 7 | +import httpx |
| 8 | + |
| 9 | +from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj |
| 10 | +from litellm.llms.base_llm.chat.transformation import BaseLLMException |
| 11 | +from litellm.llms.base_llm.embedding.transformation import BaseEmbeddingConfig |
| 12 | +from litellm.secret_managers.main import get_secret_str |
| 13 | +from litellm.types.llms.openai import AllEmbeddingInputValues, AllMessageValues |
| 14 | +from litellm.types.utils import EmbeddingResponse, Usage |
| 15 | + |
| 16 | + |
| 17 | +class VoyageError(BaseLLMException): |
| 18 | + def __init__( |
| 19 | + self, |
| 20 | + status_code: int, |
| 21 | + message: str, |
| 22 | + headers: Union[dict, httpx.Headers] = {}, |
| 23 | + ): |
| 24 | + self.status_code = status_code |
| 25 | + self.message = message |
| 26 | + self.request = httpx.Request( |
| 27 | + method="POST", url="https://api.voyageai.com/v1/contextualizedembeddings" |
| 28 | + ) |
| 29 | + self.response = httpx.Response(status_code=status_code, request=self.request) |
| 30 | + super().__init__( |
| 31 | + status_code=status_code, |
| 32 | + message=message, |
| 33 | + headers=headers, |
| 34 | + ) |
| 35 | + |
| 36 | + |
| 37 | +class VoyageContextualEmbeddingConfig(BaseEmbeddingConfig): |
| 38 | + """ |
| 39 | + Reference: https://docs.voyageai.com/reference/embeddings-api |
| 40 | + """ |
| 41 | + |
| 42 | + def __init__(self) -> None: |
| 43 | + pass |
| 44 | + |
| 45 | + def get_complete_url( |
| 46 | + self, |
| 47 | + api_base: Optional[str], |
| 48 | + api_key: Optional[str], |
| 49 | + model: str, |
| 50 | + optional_params: dict, |
| 51 | + litellm_params: dict, |
| 52 | + stream: Optional[bool] = None, |
| 53 | + ) -> str: |
| 54 | + if api_base: |
| 55 | + if not api_base.endswith("/contextualizedembeddings"): |
| 56 | + api_base = f"{api_base}/contextualizedembeddings" |
| 57 | + return api_base |
| 58 | + return "https://api.voyageai.com/v1/contextualizedembeddings" |
| 59 | + |
| 60 | + def get_supported_openai_params(self, model: str) -> list: |
| 61 | + return ["encoding_format", "dimensions"] |
| 62 | + |
| 63 | + def map_openai_params( |
| 64 | + self, |
| 65 | + non_default_params: dict, |
| 66 | + optional_params: dict, |
| 67 | + model: str, |
| 68 | + drop_params: bool, |
| 69 | + ) -> dict: |
| 70 | + """ |
| 71 | + Map OpenAI params to Voyage params |
| 72 | +
|
| 73 | + Reference: https://docs.voyageai.com/reference/contextualized-embeddings-api |
| 74 | + """ |
| 75 | + if "encoding_format" in non_default_params: |
| 76 | + optional_params["encoding_format"] = non_default_params["encoding_format"] |
| 77 | + if "dimensions" in non_default_params: |
| 78 | + optional_params["output_dimension"] = non_default_params["dimensions"] |
| 79 | + return optional_params |
| 80 | + |
| 81 | + def validate_environment( |
| 82 | + self, |
| 83 | + headers: dict, |
| 84 | + model: str, |
| 85 | + messages: List[AllMessageValues], |
| 86 | + optional_params: dict, |
| 87 | + litellm_params: dict, |
| 88 | + api_key: Optional[str] = None, |
| 89 | + api_base: Optional[str] = None, |
| 90 | + ) -> dict: |
| 91 | + if api_key is None: |
| 92 | + api_key = ( |
| 93 | + get_secret_str("VOYAGE_API_KEY") |
| 94 | + or get_secret_str("VOYAGE_AI_API_KEY") |
| 95 | + or get_secret_str("VOYAGE_AI_TOKEN") |
| 96 | + ) |
| 97 | + return { |
| 98 | + "Authorization": f"Bearer {api_key}", |
| 99 | + } |
| 100 | + |
| 101 | + def transform_embedding_request( |
| 102 | + self, |
| 103 | + model: str, |
| 104 | + input: Union[AllEmbeddingInputValues, List[List[str]]], |
| 105 | + optional_params: dict, |
| 106 | + headers: dict, |
| 107 | + ) -> dict: |
| 108 | + return { |
| 109 | + "inputs": input, |
| 110 | + "model": model, |
| 111 | + **optional_params, |
| 112 | + } |
| 113 | + |
| 114 | + def transform_embedding_response( |
| 115 | + self, |
| 116 | + model: str, |
| 117 | + raw_response: httpx.Response, |
| 118 | + model_response: EmbeddingResponse, |
| 119 | + logging_obj: LiteLLMLoggingObj, |
| 120 | + api_key: Optional[str] = None, |
| 121 | + request_data: dict = {}, |
| 122 | + optional_params: dict = {}, |
| 123 | + litellm_params: dict = {}, |
| 124 | + ) -> EmbeddingResponse: |
| 125 | + try: |
| 126 | + raw_response_json = raw_response.json() |
| 127 | + except Exception: |
| 128 | + raise VoyageError( |
| 129 | + message=raw_response.text, status_code=raw_response.status_code |
| 130 | + ) |
| 131 | + |
| 132 | + # model_response.usage |
| 133 | + model_response.model = raw_response_json.get("model") |
| 134 | + model_response.data = raw_response_json.get("data") |
| 135 | + model_response.object = raw_response_json.get("object") |
| 136 | + |
| 137 | + usage = Usage( |
| 138 | + prompt_tokens=raw_response_json.get("usage", {}).get("total_tokens", 0), |
| 139 | + total_tokens=raw_response_json.get("usage", {}).get("total_tokens", 0), |
| 140 | + ) |
| 141 | + model_response.usage = usage |
| 142 | + return model_response |
| 143 | + |
| 144 | + def get_error_class( |
| 145 | + self, error_message: str, status_code: int, headers: Union[dict, httpx.Headers] |
| 146 | + ) -> BaseLLMException: |
| 147 | + return VoyageError( |
| 148 | + message=error_message, status_code=status_code, headers=headers |
| 149 | + ) |
| 150 | + |
| 151 | + @staticmethod |
| 152 | + def is_contextualized_embeddings(model: str) -> bool: |
| 153 | + return "context" in model.lower() |
0 commit comments