12712969950

Committed 10 Jan 2025 04:04PM UTC coverage: 91.26% (+0.2%) from 91.1%

Build # 12712969950

Build Type

Pull #8605

github

Committed by

web-flow

Commit Message

Merge 89b7ad1ba into 741ce5df5

Pull Request Pull Request #8605: feat: add `RecursiveSplitter` component for `Document` preprocessing

Run Details

8844 of 9691 relevant lines covered (91.26%)

0.91 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

53.62

haystack/components/embedders/azure_document_embedder.py

# SPDX-FileCopyrightText: 2022-present deepset GmbH <info@deepset.ai>
#
# SPDX-License-Identifier: Apache-2.0

import os
from typing import Any, Dict, List, Optional, Tuple

from openai.lib.azure import AzureOpenAI
from tqdm import tqdm

from haystack import Document, component, default_from_dict, default_to_dict
from haystack.utils import Secret, deserialize_secrets_inplace


@component
class AzureOpenAIDocumentEmbedder:
    """
    Calculates document embeddings using OpenAI models deployed on Azure.

    ### Usage example

    ```python
    from haystack import Document
    from haystack.components.embedders import AzureOpenAIDocumentEmbedder

    doc = Document(content="I love pizza!")

    document_embedder = AzureOpenAIDocumentEmbedder()

    result = document_embedder.run([doc])
    print(result['documents'][0].embedding)

    # [0.017020374536514282, -0.023255806416273117, ...]
    ```
    """

    def __init__(  # noqa: PLR0913 (too-many-arguments) # pylint: disable=too-many-positional-arguments
        self,
        azure_endpoint: Optional[str] = None,
        api_version: Optional[str] = "2023-05-15",
        azure_deployment: str = "text-embedding-ada-002",
        dimensions: Optional[int] = None,
        api_key: Optional[Secret] = Secret.from_env_var("AZURE_OPENAI_API_KEY", strict=False),
        azure_ad_token: Optional[Secret] = Secret.from_env_var("AZURE_OPENAI_AD_TOKEN", strict=False),
        organization: Optional[str] = None,
        prefix: str = "",
        suffix: str = "",
        batch_size: int = 32,
        progress_bar: bool = True,
        meta_fields_to_embed: Optional[List[str]] = None,
        embedding_separator: str = "\n",
        timeout: Optional[float] = None,
        max_retries: Optional[int] = None,
    ):
        """
        Creates an AzureOpenAIDocumentEmbedder component.

        :param azure_endpoint:
            The endpoint of the model deployed on Azure.
        :param api_version:
            The version of the API to use.
        :param azure_deployment:
            The name of the model deployed on Azure. The default model is text-embedding-ada-002.
        :param dimensions:
            The number of dimensions of the resulting embeddings. Only supported in text-embedding-3
            and later models.
        :param api_key:
            The Azure OpenAI API key.
            You can set it with an environment variable `AZURE_OPENAI_API_KEY`, or pass with this
            parameter during initialization.
        :param azure_ad_token:
            Microsoft Entra ID token, see Microsoft's
            [Entra ID](https://www.microsoft.com/en-us/security/business/identity-access/microsoft-entra-id)
            documentation for more information. You can set it with an environment variable
            `AZURE_OPENAI_AD_TOKEN`, or pass with this parameter during initialization.
            Previously called Azure Active Directory.
        :param organization:
            Your organization ID. See OpenAI's
            [Setting Up Your Organization](https://platform.openai.com/docs/guides/production-best-practices/setting-up-your-organization)
            for more information.
        :param prefix:
            A string to add at the beginning of each text.
        :param suffix:
            A string to add at the end of each text.
        :param batch_size:
            Number of documents to embed at once.
        :param progress_bar:
            If `True`, shows a progress bar when running.
        :param meta_fields_to_embed:
            List of metadata fields to embed along with the document text.
        :param embedding_separator:
            Separator used to concatenate the metadata fields to the document text.
        :param timeout: The timeout for `AzureOpenAI` client calls, in seconds.
            If not set, defaults to either the
            `OPENAI_TIMEOUT` environment variable, or 30 seconds.
        :param max_retries: Maximum number of retries to contact AzureOpenAI after an internal error.
            If not set, defaults to either the `OPENAI_MAX_RETRIES` environment variable or to 5 retries.
        """
        # if not provided as a parameter, azure_endpoint is read from the env var AZURE_OPENAI_ENDPOINT
        azure_endpoint = azure_endpoint or os.environ.get("AZURE_OPENAI_ENDPOINT")
        if not azure_endpoint:
            raise ValueError("Please provide an Azure endpoint or set the environment variable AZURE_OPENAI_ENDPOINT.")

        if api_key is None and azure_ad_token is None:
            raise ValueError("Please provide an API key or an Azure Active Directory token.")

        self.api_key = api_key
        self.azure_ad_token = azure_ad_token
        self.api_version = api_version
        self.azure_endpoint = azure_endpoint
        self.azure_deployment = azure_deployment
        self.dimensions = dimensions
        self.organization = organization
        self.prefix = prefix
        self.suffix = suffix
        self.batch_size = batch_size
        self.progress_bar = progress_bar
        self.meta_fields_to_embed = meta_fields_to_embed or []
        self.embedding_separator = embedding_separator
        self.timeout = timeout or float(os.environ.get("OPENAI_TIMEOUT", 30.0))
        self.max_retries = max_retries or int(os.environ.get("OPENAI_MAX_RETRIES", 5))

        self._client = AzureOpenAI(
            api_version=api_version,
            azure_endpoint=azure_endpoint,
            azure_deployment=azure_deployment,
            api_key=api_key.resolve_value() if api_key is not None else None,
            azure_ad_token=azure_ad_token.resolve_value() if azure_ad_token is not None else None,
            organization=organization,
            timeout=self.timeout,
            max_retries=self.max_retries,
        )

    def _get_telemetry_data(self) -> Dict[str, Any]:
        """
        Data that is sent to Posthog for usage analytics.
        """
        return {"model": self.azure_deployment}

    def to_dict(self) -> Dict[str, Any]:
        """
        Serializes the component to a dictionary.

        :returns:
            Dictionary with serialized data.
        """
        return default_to_dict(
            self,
            azure_endpoint=self.azure_endpoint,
            azure_deployment=self.azure_deployment,
            dimensions=self.dimensions,
            organization=self.organization,
            api_version=self.api_version,
            prefix=self.prefix,
            suffix=self.suffix,
            batch_size=self.batch_size,
            progress_bar=self.progress_bar,
            meta_fields_to_embed=self.meta_fields_to_embed,
            embedding_separator=self.embedding_separator,
            api_key=self.api_key.to_dict() if self.api_key is not None else None,
            azure_ad_token=self.azure_ad_token.to_dict() if self.azure_ad_token is not None else None,
            timeout=self.timeout,
            max_retries=self.max_retries,
        )

    @classmethod
    def from_dict(cls, data: Dict[str, Any]) -> "AzureOpenAIDocumentEmbedder":
        """
        Deserializes the component from a dictionary.

        :param data:
            Dictionary to deserialize from.
        :returns:
            Deserialized component.
        """
        deserialize_secrets_inplace(data["init_parameters"], keys=["api_key", "azure_ad_token"])
        return default_from_dict(cls, data)

    def _prepare_texts_to_embed(self, documents: List[Document]) -> List[str]:
        """
        Prepare the texts to embed by concatenating the Document text with the metadata fields to embed.
        """
        texts_to_embed = []
        for doc in documents:
            meta_values_to_embed = [
                str(doc.meta[key]) for key in self.meta_fields_to_embed if key in doc.meta and doc.meta[key] is not None
            ]

            text_to_embed = (
                self.prefix + self.embedding_separator.join(meta_values_to_embed + [doc.content or ""]) + self.suffix
            ).replace("\n", " ")

            texts_to_embed.append(text_to_embed)
        return texts_to_embed

    def _embed_batch(self, texts_to_embed: List[str], batch_size: int) -> Tuple[List[List[float]], Dict[str, Any]]:
        """
        Embed a list of texts in batches.
        """

        all_embeddings: List[List[float]] = []
        meta: Dict[str, Any] = {"model": "", "usage": {"prompt_tokens": 0, "total_tokens": 0}}
        for i in tqdm(range(0, len(texts_to_embed), batch_size), desc="Embedding Texts"):
            batch = texts_to_embed[i : i + batch_size]
            if self.dimensions is not None:
                response = self._client.embeddings.create(
                    model=self.azure_deployment, dimensions=self.dimensions, input=batch
                )
            else:
                response = self._client.embeddings.create(model=self.azure_deployment, input=batch)

            # Append embeddings to the list
            all_embeddings.extend(el.embedding for el in response.data)

            # Update the meta information only once if it's empty
            if not meta["model"]:
                meta["model"] = response.model
                meta["usage"] = dict(response.usage)
            else:
                # Update the usage tokens
                meta["usage"]["prompt_tokens"] += response.usage.prompt_tokens
                meta["usage"]["total_tokens"] += response.usage.total_tokens

        return all_embeddings, meta

    @component.output_types(documents=List[Document], meta=Dict[str, Any])
    def run(self, documents: List[Document]) -> Dict[str, Any]:
        """
        Embeds a list of documents.

        :param documents:
            Documents to embed.

        :returns:
            A dictionary with the following keys:
            - `documents`: A list of documents with embeddings.
            - `meta`: Information about the usage of the model.
        """
        if not (isinstance(documents, list) and all(isinstance(doc, Document) for doc in documents)):
            raise TypeError("Input must be a list of Document instances. For strings, use AzureOpenAITextEmbedder.")

        texts_to_embed = self._prepare_texts_to_embed(documents=documents)
        embeddings, meta = self._embed_batch(texts_to_embed=texts_to_embed, batch_size=self.batch_size)

        # Assign the corresponding embeddings to each document
        for doc, emb in zip(documents, embeddings):
            doc.embedding = emb

        return {"documents": documents, "meta": meta}

1	# SPDX-FileCopyrightText: 2022-present deepset GmbH <info@deepset.ai>
2	#
3	# SPDX-License-Identifier: Apache-2.0
4
5	import os	1✔
6	from typing import Any, Dict, List, Optional, Tuple	1✔
7
8	from openai.lib.azure import AzureOpenAI	1✔
9	from tqdm import tqdm	1✔
10
11	from haystack import Document, component, default_from_dict, default_to_dict	1✔
12	from haystack.utils import Secret, deserialize_secrets_inplace	1✔
13
14
15	@component	1✔
16	class AzureOpenAIDocumentEmbedder:	1✔
17	"""
18	Calculates document embeddings using OpenAI models deployed on Azure.
19
20	### Usage example
21
22	```python
23	from haystack import Document
24	from haystack.components.embedders import AzureOpenAIDocumentEmbedder
25
26	doc = Document(content="I love pizza!")
27
28	document_embedder = AzureOpenAIDocumentEmbedder()
29
30	result = document_embedder.run([doc])
31	print(result['documents'][0].embedding)
32
33	# [0.017020374536514282, -0.023255806416273117, ...]
34	```
35	"""
36
37	def __init__( # noqa: PLR0913 (too-many-arguments) # pylint: disable=too-many-positional-arguments	1✔
38	self,
39	azure_endpoint: Optional[str] = None,
40	api_version: Optional[str] = "2023-05-15",
41	azure_deployment: str = "text-embedding-ada-002",
42	dimensions: Optional[int] = None,
43	api_key: Optional[Secret] = Secret.from_env_var("AZURE_OPENAI_API_KEY", strict=False),
44	azure_ad_token: Optional[Secret] = Secret.from_env_var("AZURE_OPENAI_AD_TOKEN", strict=False),
45	organization: Optional[str] = None,
46	prefix: str = "",
47	suffix: str = "",
48	batch_size: int = 32,
49	progress_bar: bool = True,
50	meta_fields_to_embed: Optional[List[str]] = None,
51	embedding_separator: str = "\n",
52	timeout: Optional[float] = None,
53	max_retries: Optional[int] = None,
54	):
55	"""
56	Creates an AzureOpenAIDocumentEmbedder component.
57
58	:param azure_endpoint:
59	The endpoint of the model deployed on Azure.
60	:param api_version:
61	The version of the API to use.
62	:param azure_deployment:
63	The name of the model deployed on Azure. The default model is text-embedding-ada-002.
64	:param dimensions:
65	The number of dimensions of the resulting embeddings. Only supported in text-embedding-3
66	and later models.
67	:param api_key:
68	The Azure OpenAI API key.
69	You can set it with an environment variable `AZURE_OPENAI_API_KEY`, or pass with this
70	parameter during initialization.
71	:param azure_ad_token:
72	Microsoft Entra ID token, see Microsoft's
73	[Entra ID](https://www.microsoft.com/en-us/security/business/identity-access/microsoft-entra-id)
74	documentation for more information. You can set it with an environment variable
75	`AZURE_OPENAI_AD_TOKEN`, or pass with this parameter during initialization.
76	Previously called Azure Active Directory.
77	:param organization:
78	Your organization ID. See OpenAI's
79	[Setting Up Your Organization](https://platform.openai.com/docs/guides/production-best-practices/setting-up-your-organization)
80	for more information.
81	:param prefix:
82	A string to add at the beginning of each text.
83	:param suffix:
84	A string to add at the end of each text.
85	:param batch_size:
86	Number of documents to embed at once.
87	:param progress_bar:
88	If `True`, shows a progress bar when running.
89	:param meta_fields_to_embed:
90	List of metadata fields to embed along with the document text.
91	:param embedding_separator:
92	Separator used to concatenate the metadata fields to the document text.
93	:param timeout: The timeout for `AzureOpenAI` client calls, in seconds.
94	If not set, defaults to either the
95	`OPENAI_TIMEOUT` environment variable, or 30 seconds.
96	:param max_retries: Maximum number of retries to contact AzureOpenAI after an internal error.
97	If not set, defaults to either the `OPENAI_MAX_RETRIES` environment variable or to 5 retries.
98	"""
99	# if not provided as a parameter, azure_endpoint is read from the env var AZURE_OPENAI_ENDPOINT
100	azure_endpoint = azure_endpoint or os.environ.get("AZURE_OPENAI_ENDPOINT")	1✔
101	if not azure_endpoint:	1✔
102	raise ValueError("Please provide an Azure endpoint or set the environment variable AZURE_OPENAI_ENDPOINT.")	×
103
104	if api_key is None and azure_ad_token is None:	1✔
105	raise ValueError("Please provide an API key or an Azure Active Directory token.")	×
106
107	self.api_key = api_key	1✔
108	self.azure_ad_token = azure_ad_token	1✔
109	self.api_version = api_version	1✔
110	self.azure_endpoint = azure_endpoint	1✔
111	self.azure_deployment = azure_deployment	1✔
112	self.dimensions = dimensions	1✔
113	self.organization = organization	1✔
114	self.prefix = prefix	1✔
115	self.suffix = suffix	1✔
116	self.batch_size = batch_size	1✔
117	self.progress_bar = progress_bar	1✔
118	self.meta_fields_to_embed = meta_fields_to_embed or []	1✔
119	self.embedding_separator = embedding_separator	1✔
120	self.timeout = timeout or float(os.environ.get("OPENAI_TIMEOUT", 30.0))	1✔
121	self.max_retries = max_retries or int(os.environ.get("OPENAI_MAX_RETRIES", 5))	1✔
122
123	self._client = AzureOpenAI(	1✔
124	api_version=api_version,
125	azure_endpoint=azure_endpoint,
126	azure_deployment=azure_deployment,
127	api_key=api_key.resolve_value() if api_key is not None else None,
128	azure_ad_token=azure_ad_token.resolve_value() if azure_ad_token is not None else None,
129	organization=organization,
130	timeout=self.timeout,
131	max_retries=self.max_retries,
132	)
133
134	def _get_telemetry_data(self) -> Dict[str, Any]:	1✔
135	"""
136	Data that is sent to Posthog for usage analytics.
137	"""
138	return {"model": self.azure_deployment}	×
139
140	def to_dict(self) -> Dict[str, Any]:	1✔
141	"""
142	Serializes the component to a dictionary.
143
144	:returns:
145	Dictionary with serialized data.
146	"""
147	return default_to_dict(	1✔
148	self,
149	azure_endpoint=self.azure_endpoint,
150	azure_deployment=self.azure_deployment,
151	dimensions=self.dimensions,
152	organization=self.organization,
153	api_version=self.api_version,
154	prefix=self.prefix,
155	suffix=self.suffix,
156	batch_size=self.batch_size,
157	progress_bar=self.progress_bar,
158	meta_fields_to_embed=self.meta_fields_to_embed,
159	embedding_separator=self.embedding_separator,
160	api_key=self.api_key.to_dict() if self.api_key is not None else None,
161	azure_ad_token=self.azure_ad_token.to_dict() if self.azure_ad_token is not None else None,
162	timeout=self.timeout,
163	max_retries=self.max_retries,
164	)
165
166	@classmethod	1✔
167	def from_dict(cls, data: Dict[str, Any]) -> "AzureOpenAIDocumentEmbedder":	1✔
168	"""
169	Deserializes the component from a dictionary.
170
171	:param data:
172	Dictionary to deserialize from.
173	:returns:
174	Deserialized component.
175	"""
176	deserialize_secrets_inplace(data["init_parameters"], keys=["api_key", "azure_ad_token"])	×
177	return default_from_dict(cls, data)	×
178
179	def _prepare_texts_to_embed(self, documents: List[Document]) -> List[str]:	1✔
180	"""
181	Prepare the texts to embed by concatenating the Document text with the metadata fields to embed.
182	"""
183	texts_to_embed = []	×
184	for doc in documents:	×
185	meta_values_to_embed = [	×
186	str(doc.meta[key]) for key in self.meta_fields_to_embed if key in doc.meta and doc.meta[key] is not None
187	]
188
189	text_to_embed = (	×
190	self.prefix + self.embedding_separator.join(meta_values_to_embed + [doc.content or ""]) + self.suffix
191	).replace("\n", " ")
192
193	texts_to_embed.append(text_to_embed)	×
194	return texts_to_embed	×
195
196	def _embed_batch(self, texts_to_embed: List[str], batch_size: int) -> Tuple[List[List[float]], Dict[str, Any]]:	1✔
197	"""
198	Embed a list of texts in batches.
199	"""
200
201	all_embeddings: List[List[float]] = []	×
202	meta: Dict[str, Any] = {"model": "", "usage": {"prompt_tokens": 0, "total_tokens": 0}}	×
203	for i in tqdm(range(0, len(texts_to_embed), batch_size), desc="Embedding Texts"):	×
204	batch = texts_to_embed[i : i + batch_size]	×
205	if self.dimensions is not None:	×
206	response = self._client.embeddings.create(	×
207	model=self.azure_deployment, dimensions=self.dimensions, input=batch
208	)
209	else:
210	response = self._client.embeddings.create(model=self.azure_deployment, input=batch)	×
211
212	# Append embeddings to the list
213	all_embeddings.extend(el.embedding for el in response.data)	×
214
215	# Update the meta information only once if it's empty
216	if not meta["model"]:	×
217	meta["model"] = response.model	×
218	meta["usage"] = dict(response.usage)	×
219	else:
220	# Update the usage tokens
221	meta["usage"]["prompt_tokens"] += response.usage.prompt_tokens	×
222	meta["usage"]["total_tokens"] += response.usage.total_tokens	×
223
224	return all_embeddings, meta	×
225
226	@component.output_types(documents=List[Document], meta=Dict[str, Any])	1✔
227	def run(self, documents: List[Document]) -> Dict[str, Any]:	1✔
228	"""
229	Embeds a list of documents.
230
231	:param documents:
232	Documents to embed.
233
234	:returns:
235	A dictionary with the following keys:
236	- `documents`: A list of documents with embeddings.
237	- `meta`: Information about the usage of the model.
238	"""
239	if not (isinstance(documents, list) and all(isinstance(doc, Document) for doc in documents)):	×
240	raise TypeError("Input must be a list of Document instances. For strings, use AzureOpenAITextEmbedder.")	×
241
242	texts_to_embed = self._prepare_texts_to_embed(documents=documents)	×
243	embeddings, meta = self._embed_batch(texts_to_embed=texts_to_embed, batch_size=self.batch_size)	×
244
245	# Assign the corresponding embeddings to each document
246	for doc, emb in zip(documents, embeddings):	×
247	doc.embedding = emb	×
248
249	return {"documents": documents, "meta": meta}	×

deepset-ai / haystack / 12712969950

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous