openai/openai-python
Publicmirrored from https://github.com/openai/openai-pythonAvailable
openai/api_resources/embedding.py
91lines · modecode
| 1 | import base64 |
| 2 | import time |
| 3 | |
| 4 | from openai import util |
| 5 | from openai.api_resources.abstract.engine_api_resource import EngineAPIResource |
| 6 | from openai.datalib.numpy_helper import assert_has_numpy |
| 7 | from openai.datalib.numpy_helper import numpy as np |
| 8 | from openai.error import TryAgain |
| 9 | |
| 10 | |
| 11 | class Embedding(EngineAPIResource): |
| 12 | OBJECT_NAME = "embeddings" |
| 13 | |
| 14 | @classmethod |
| 15 | def create(cls, *args, **kwargs): |
| 16 | """ |
| 17 | Creates a new embedding for the provided input and parameters. |
| 18 | |
| 19 | See https://platform.openai.com/docs/api-reference/embeddings for a list |
| 20 | of valid parameters. |
| 21 | """ |
| 22 | start = time.time() |
| 23 | timeout = kwargs.pop("timeout", None) |
| 24 | |
| 25 | user_provided_encoding_format = kwargs.get("encoding_format", None) |
| 26 | |
| 27 | # If encoding format was not explicitly specified, we opaquely use base64 for performance |
| 28 | if not user_provided_encoding_format: |
| 29 | kwargs["encoding_format"] = "base64" |
| 30 | |
| 31 | while True: |
| 32 | try: |
| 33 | response = super().create(*args, **kwargs) |
| 34 | |
| 35 | # If a user specifies base64, we'll just return the encoded string. |
| 36 | # This is only for the default case. |
| 37 | if not user_provided_encoding_format: |
| 38 | for data in response.data: |
| 39 | |
| 40 | # If an engine isn't using this optimization, don't do anything |
| 41 | if type(data["embedding"]) == str: |
| 42 | assert_has_numpy() |
| 43 | data["embedding"] = np.frombuffer( |
| 44 | base64.b64decode(data["embedding"]), dtype="float32" |
| 45 | ).tolist() |
| 46 | |
| 47 | return response |
| 48 | except TryAgain as e: |
| 49 | if timeout is not None and time.time() > start + timeout: |
| 50 | raise |
| 51 | |
| 52 | util.log_info("Waiting for model to warm up", error=e) |
| 53 | |
| 54 | @classmethod |
| 55 | async def acreate(cls, *args, **kwargs): |
| 56 | """ |
| 57 | Creates a new embedding for the provided input and parameters. |
| 58 | |
| 59 | See https://platform.openai.com/docs/api-reference/embeddings for a list |
| 60 | of valid parameters. |
| 61 | """ |
| 62 | start = time.time() |
| 63 | timeout = kwargs.pop("timeout", None) |
| 64 | |
| 65 | user_provided_encoding_format = kwargs.get("encoding_format", None) |
| 66 | |
| 67 | # If encoding format was not explicitly specified, we opaquely use base64 for performance |
| 68 | if not user_provided_encoding_format: |
| 69 | kwargs["encoding_format"] = "base64" |
| 70 | |
| 71 | while True: |
| 72 | try: |
| 73 | response = await super().acreate(*args, **kwargs) |
| 74 | |
| 75 | # If a user specifies base64, we'll just return the encoded string. |
| 76 | # This is only for the default case. |
| 77 | if not user_provided_encoding_format: |
| 78 | for data in response.data: |
| 79 | |
| 80 | # If an engine isn't using this optimization, don't do anything |
| 81 | if type(data["embedding"]) == str: |
| 82 | data["embedding"] = np.frombuffer( |
| 83 | base64.b64decode(data["embedding"]), dtype="float32" |
| 84 | ).tolist() |
| 85 | |
| 86 | return response |
| 87 | except TryAgain as e: |
| 88 | if timeout is not None and time.time() > start + timeout: |
| 89 | raise |
| 90 | |
| 91 | util.log_info("Waiting for model to warm up", error=e) |
| 92 | |