13972131258

Committed 20 Mar 2025 02:43PM UTC coverage: 90.021% (-0.03%) from 90.054%

Build # 13972131258

Build Type

Pull #9069

github

Committed by

web-flow

Commit Message

Merge 8371761b0 into 67ab3788e

Pull Request Pull Request #9069: refactor!: `ChatMessage` serialization-deserialization updates

Run Details

9833 of 10923 relevant lines covered (90.02%)

0.9 hits per line

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

96.59

haystack/components/embedders/hugging_face_api_document_embedder.py

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

import warnings
from typing import Any, Dict, List, Optional, Union

from tqdm import tqdm

from haystack import component, default_from_dict, default_to_dict
from haystack.dataclasses import Document
from haystack.lazy_imports import LazyImport
from haystack.utils import Secret, deserialize_secrets_inplace
from haystack.utils.hf import HFEmbeddingAPIType, HFModelType, check_valid_model
from haystack.utils.url_validation import is_valid_http_url

with LazyImport(message="Run 'pip install \"huggingface_hub>=0.27.0\"'") as huggingface_hub_import:
    from huggingface_hub import InferenceClient


@component
class HuggingFaceAPIDocumentEmbedder:
    """
    Embeds documents using Hugging Face APIs.

    Use it with the following Hugging Face APIs:
    - [Free Serverless Inference API](https://huggingface.co/inference-api)
    - [Paid Inference Endpoints](https://huggingface.co/inference-endpoints)
    - [Self-hosted Text Embeddings Inference](https://github.com/huggingface/text-embeddings-inference)


    ### Usage examples

    #### With free serverless inference API

    ```python
    from haystack.components.embedders import HuggingFaceAPIDocumentEmbedder
    from haystack.utils import Secret
    from haystack.dataclasses import Document

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

    doc_embedder = HuggingFaceAPIDocumentEmbedder(api_type="serverless_inference_api",
                                                  api_params={"model": "BAAI/bge-small-en-v1.5"},
                                                  token=Secret.from_token("<your-api-key>"))

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

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

    #### With paid inference endpoints

    ```python
    from haystack.components.embedders import HuggingFaceAPIDocumentEmbedder
    from haystack.utils import Secret
    from haystack.dataclasses import Document

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

    doc_embedder = HuggingFaceAPIDocumentEmbedder(api_type="inference_endpoints",
                                                  api_params={"url": "<your-inference-endpoint-url>"},
                                                  token=Secret.from_token("<your-api-key>"))

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

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

    #### With self-hosted text embeddings inference

    ```python
    from haystack.components.embedders import HuggingFaceAPIDocumentEmbedder
    from haystack.dataclasses import Document

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

    doc_embedder = HuggingFaceAPIDocumentEmbedder(api_type="text_embeddings_inference",
                                                  api_params={"url": "http://localhost:8080"})

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

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

    def __init__(
        self,
        api_type: Union[HFEmbeddingAPIType, str],
        api_params: Dict[str, str],
        token: Optional[Secret] = Secret.from_env_var(["HF_API_TOKEN", "HF_TOKEN"], strict=False),
        prefix: str = "",
        suffix: str = "",
        truncate: Optional[bool] = True,
        normalize: Optional[bool] = False,
        batch_size: int = 32,
        progress_bar: bool = True,
        meta_fields_to_embed: Optional[List[str]] = None,
        embedding_separator: str = "\n",
    ):  # pylint: disable=too-many-positional-arguments
        """
        Creates a HuggingFaceAPIDocumentEmbedder component.

        :param api_type:
            The type of Hugging Face API to use.
        :param api_params:
            A dictionary with the following keys:
            - `model`: Hugging Face model ID. Required when `api_type` is `SERVERLESS_INFERENCE_API`.
            - `url`: URL of the inference endpoint. Required when `api_type` is `INFERENCE_ENDPOINTS` or
            `TEXT_EMBEDDINGS_INFERENCE`.
        :param token: The Hugging Face token to use as HTTP bearer authorization.
            Check your HF token in your [account settings](https://huggingface.co/settings/tokens).
        :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 truncate:
            Truncates the input text to the maximum length supported by the model.
            Applicable when `api_type` is `TEXT_EMBEDDINGS_INFERENCE`, or `INFERENCE_ENDPOINTS`
            if the backend uses Text Embeddings Inference.
            If `api_type` is `SERVERLESS_INFERENCE_API`, this parameter is ignored.
        :param normalize:
            Normalizes the embeddings to unit length.
            Applicable when `api_type` is `TEXT_EMBEDDINGS_INFERENCE`, or `INFERENCE_ENDPOINTS`
            if the backend uses Text Embeddings Inference.
            If `api_type` is `SERVERLESS_INFERENCE_API`, this parameter is ignored.
        :param batch_size:
            Number of documents to process 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.
        """
        huggingface_hub_import.check()

        if isinstance(api_type, str):
            api_type = HFEmbeddingAPIType.from_str(api_type)

        api_params = api_params or {}

        if api_type == HFEmbeddingAPIType.SERVERLESS_INFERENCE_API:
            model = api_params.get("model")
            if model is None:
                raise ValueError(
                    "To use the Serverless Inference API, you need to specify the `model` parameter in `api_params`."
                )
            check_valid_model(model, HFModelType.EMBEDDING, token)
            model_or_url = model
        elif api_type in [HFEmbeddingAPIType.INFERENCE_ENDPOINTS, HFEmbeddingAPIType.TEXT_EMBEDDINGS_INFERENCE]:
            url = api_params.get("url")
            if url is None:
                msg = (
                    "To use Text Embeddings Inference or Inference Endpoints, you need to specify the `url` "
                    "parameter in `api_params`."
                )
                raise ValueError(msg)
            if not is_valid_http_url(url):
                raise ValueError(f"Invalid URL: {url}")
            model_or_url = url
        else:
            msg = f"Unknown api_type {api_type}"
            raise ValueError(msg)

        self.api_type = api_type
        self.api_params = api_params
        self.token = token
        self.prefix = prefix
        self.suffix = suffix
        self.truncate = truncate
        self.normalize = normalize
        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._client = InferenceClient(model_or_url, token=token.resolve_value() if token else None)

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

        :returns:
            Dictionary with serialized data.
        """
        return default_to_dict(
            self,
            api_type=str(self.api_type),
            api_params=self.api_params,
            prefix=self.prefix,
            suffix=self.suffix,
            token=self.token.to_dict() if self.token else None,
            truncate=self.truncate,
            normalize=self.normalize,
            batch_size=self.batch_size,
            progress_bar=self.progress_bar,
            meta_fields_to_embed=self.meta_fields_to_embed,
            embedding_separator=self.embedding_separator,
        )

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

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

            texts_to_embed.append(text_to_embed)
        return texts_to_embed

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

        if self.api_type == HFEmbeddingAPIType.SERVERLESS_INFERENCE_API:
            if truncate is not None:
                msg = "`truncate` parameter is not supported for Serverless Inference API. It will be ignored."
                warnings.warn(msg)
                truncate = None
            if normalize is not None:
                msg = "`normalize` parameter is not supported for Serverless Inference API. It will be ignored."
                warnings.warn(msg)
                normalize = None

        all_embeddings = []
        for i in tqdm(
            range(0, len(texts_to_embed), batch_size), disable=not self.progress_bar, desc="Calculating embeddings"
        ):
            batch = texts_to_embed[i : i + batch_size]

            np_embeddings = self._client.feature_extraction(
                # this method does not officially support list of strings, but works as expected
                text=batch,  # type: ignore[arg-type]
                truncate=truncate,
                normalize=normalize,
            )

            if np_embeddings.ndim != 2 or np_embeddings.shape[0] != len(batch):
                raise ValueError(f"Expected embedding shape ({batch_size}, embedding_dim), got {np_embeddings.shape}")

            all_embeddings.extend(np_embeddings.tolist())

        return all_embeddings

    @component.output_types(documents=List[Document])
    def run(self, documents: List[Document]):
        """
        Embeds a list of documents.

        :param documents:
            Documents to embed.

        :returns:
            A dictionary with the following keys:
            - `documents`: A list of documents with embeddings.
        """
        if not isinstance(documents, list) or documents and not isinstance(documents[0], Document):
            raise TypeError(
                "HuggingFaceAPIDocumentEmbedder expects a list of Documents as input."
                " In case you want to embed a string, please use the HuggingFaceAPITextEmbedder."
            )

        texts_to_embed = self._prepare_texts_to_embed(documents=documents)

        embeddings = self._embed_batch(texts_to_embed=texts_to_embed, batch_size=self.batch_size)

        for doc, emb in zip(documents, embeddings):
            doc.embedding = emb

        return {"documents": documents}

1	# SPDX-FileCopyrightText: 2022-present deepset GmbH <info@deepset.ai>
2	#
3	# SPDX-License-Identifier: Apache-2.0
4
5	import warnings	1✔
6	from typing import Any, Dict, List, Optional, Union	1✔
7
8	from tqdm import tqdm	1✔
9
10	from haystack import component, default_from_dict, default_to_dict	1✔
11	from haystack.dataclasses import Document	1✔
12	from haystack.lazy_imports import LazyImport	1✔
13	from haystack.utils import Secret, deserialize_secrets_inplace	1✔
14	from haystack.utils.hf import HFEmbeddingAPIType, HFModelType, check_valid_model	1✔
15	from haystack.utils.url_validation import is_valid_http_url	1✔
16
17	with LazyImport(message="Run 'pip install \"huggingface_hub>=0.27.0\"'") as huggingface_hub_import:	1✔
18	from huggingface_hub import InferenceClient	1✔
19
20
21	@component	1✔
22	class HuggingFaceAPIDocumentEmbedder:	1✔
23	"""
24	Embeds documents using Hugging Face APIs.
25
26	Use it with the following Hugging Face APIs:
27	- [Free Serverless Inference API](https://huggingface.co/inference-api)
28	- [Paid Inference Endpoints](https://huggingface.co/inference-endpoints)
29	- [Self-hosted Text Embeddings Inference](https://github.com/huggingface/text-embeddings-inference)
30
31
32	### Usage examples
33
34	#### With free serverless inference API
35
36	```python
37	from haystack.components.embedders import HuggingFaceAPIDocumentEmbedder
38	from haystack.utils import Secret
39	from haystack.dataclasses import Document
40
41	doc = Document(content="I love pizza!")
42
43	doc_embedder = HuggingFaceAPIDocumentEmbedder(api_type="serverless_inference_api",
44	api_params={"model": "BAAI/bge-small-en-v1.5"},
45	token=Secret.from_token("<your-api-key>"))
46
47	result = document_embedder.run([doc])
48	print(result["documents"][0].embedding)
49
50	# [0.017020374536514282, -0.023255806416273117, ...]
51	```
52
53	#### With paid inference endpoints
54
55	```python
56	from haystack.components.embedders import HuggingFaceAPIDocumentEmbedder
57	from haystack.utils import Secret
58	from haystack.dataclasses import Document
59
60	doc = Document(content="I love pizza!")
61
62	doc_embedder = HuggingFaceAPIDocumentEmbedder(api_type="inference_endpoints",
63	api_params={"url": "<your-inference-endpoint-url>"},
64	token=Secret.from_token("<your-api-key>"))
65
66	result = document_embedder.run([doc])
67	print(result["documents"][0].embedding)
68
69	# [0.017020374536514282, -0.023255806416273117, ...]
70	```
71
72	#### With self-hosted text embeddings inference
73
74	```python
75	from haystack.components.embedders import HuggingFaceAPIDocumentEmbedder
76	from haystack.dataclasses import Document
77
78	doc = Document(content="I love pizza!")
79
80	doc_embedder = HuggingFaceAPIDocumentEmbedder(api_type="text_embeddings_inference",
81	api_params={"url": "http://localhost:8080"})
82
83	result = document_embedder.run([doc])
84	print(result["documents"][0].embedding)
85
86	# [0.017020374536514282, -0.023255806416273117, ...]
87	```
88	"""
89
90	def __init__(	1✔
91	self,
92	api_type: Union[HFEmbeddingAPIType, str],
93	api_params: Dict[str, str],
94	token: Optional[Secret] = Secret.from_env_var(["HF_API_TOKEN", "HF_TOKEN"], strict=False),
95	prefix: str = "",
96	suffix: str = "",
97	truncate: Optional[bool] = True,
98	normalize: Optional[bool] = False,
99	batch_size: int = 32,
100	progress_bar: bool = True,
101	meta_fields_to_embed: Optional[List[str]] = None,
102	embedding_separator: str = "\n",
103	): # pylint: disable=too-many-positional-arguments
104	"""
105	Creates a HuggingFaceAPIDocumentEmbedder component.
106
107	:param api_type:
108	The type of Hugging Face API to use.
109	:param api_params:
110	A dictionary with the following keys:
111	- `model`: Hugging Face model ID. Required when `api_type` is `SERVERLESS_INFERENCE_API`.
112	- `url`: URL of the inference endpoint. Required when `api_type` is `INFERENCE_ENDPOINTS` or
113	`TEXT_EMBEDDINGS_INFERENCE`.
114	:param token: The Hugging Face token to use as HTTP bearer authorization.
115	Check your HF token in your [account settings](https://huggingface.co/settings/tokens).
116	:param prefix:
117	A string to add at the beginning of each text.
118	:param suffix:
119	A string to add at the end of each text.
120	:param truncate:
121	Truncates the input text to the maximum length supported by the model.
122	Applicable when `api_type` is `TEXT_EMBEDDINGS_INFERENCE`, or `INFERENCE_ENDPOINTS`
123	if the backend uses Text Embeddings Inference.
124	If `api_type` is `SERVERLESS_INFERENCE_API`, this parameter is ignored.
125	:param normalize:
126	Normalizes the embeddings to unit length.
127	Applicable when `api_type` is `TEXT_EMBEDDINGS_INFERENCE`, or `INFERENCE_ENDPOINTS`
128	if the backend uses Text Embeddings Inference.
129	If `api_type` is `SERVERLESS_INFERENCE_API`, this parameter is ignored.
130	:param batch_size:
131	Number of documents to process at once.
132	:param progress_bar:
133	If `True`, shows a progress bar when running.
134	:param meta_fields_to_embed:
135	List of metadata fields to embed along with the document text.
136	:param embedding_separator:
137	Separator used to concatenate the metadata fields to the document text.
138	"""
139	huggingface_hub_import.check()	1✔
140
141	if isinstance(api_type, str):	1✔
142	api_type = HFEmbeddingAPIType.from_str(api_type)	1✔
143
144	api_params = api_params or {}	1✔
145
146	if api_type == HFEmbeddingAPIType.SERVERLESS_INFERENCE_API:	1✔
147	model = api_params.get("model")	1✔
148	if model is None:	1✔
149	raise ValueError(	1✔
150	"To use the Serverless Inference API, you need to specify the `model` parameter in `api_params`."
151	)
152	check_valid_model(model, HFModelType.EMBEDDING, token)	1✔
153	model_or_url = model	1✔
154	elif api_type in [HFEmbeddingAPIType.INFERENCE_ENDPOINTS, HFEmbeddingAPIType.TEXT_EMBEDDINGS_INFERENCE]:	1✔
155	url = api_params.get("url")	1✔
156	if url is None:	1✔
157	msg = (	1✔
158	"To use Text Embeddings Inference or Inference Endpoints, you need to specify the `url` "
159	"parameter in `api_params`."
160	)
161	raise ValueError(msg)	1✔
162	if not is_valid_http_url(url):	1✔
163	raise ValueError(f"Invalid URL: {url}")	1✔
164	model_or_url = url	1✔
165	else:
166	msg = f"Unknown api_type {api_type}"	×
167	raise ValueError(msg)	×
168
169	self.api_type = api_type	1✔
170	self.api_params = api_params	1✔
171	self.token = token	1✔
172	self.prefix = prefix	1✔
173	self.suffix = suffix	1✔
174	self.truncate = truncate	1✔
175	self.normalize = normalize	1✔
176	self.batch_size = batch_size	1✔
177	self.progress_bar = progress_bar	1✔
178	self.meta_fields_to_embed = meta_fields_to_embed or []	1✔
179	self.embedding_separator = embedding_separator	1✔
180	self._client = InferenceClient(model_or_url, token=token.resolve_value() if token else None)	1✔
181
182	def to_dict(self) -> Dict[str, Any]:	1✔
183	"""
184	Serializes the component to a dictionary.
185
186	:returns:
187	Dictionary with serialized data.
188	"""
189	return default_to_dict(	1✔
190	self,
191	api_type=str(self.api_type),
192	api_params=self.api_params,
193	prefix=self.prefix,
194	suffix=self.suffix,
195	token=self.token.to_dict() if self.token else None,
196	truncate=self.truncate,
197	normalize=self.normalize,
198	batch_size=self.batch_size,
199	progress_bar=self.progress_bar,
200	meta_fields_to_embed=self.meta_fields_to_embed,
201	embedding_separator=self.embedding_separator,
202	)
203
204	@classmethod	1✔
205	def from_dict(cls, data: Dict[str, Any]) -> "HuggingFaceAPIDocumentEmbedder":	1✔
206	"""
207	Deserializes the component from a dictionary.
208
209	:param data:
210	Dictionary to deserialize from.
211	:returns:
212	Deserialized component.
213	"""
214	deserialize_secrets_inplace(data["init_parameters"], keys=["token"])	1✔
215	return default_from_dict(cls, data)	1✔
216
217	def _prepare_texts_to_embed(self, documents: List[Document]) -> List[str]:	1✔
218	"""
219	Prepare the texts to embed by concatenating the Document text with the metadata fields to embed.
220	"""
221	texts_to_embed = []	1✔
222	for doc in documents:	1✔
223	meta_values_to_embed = [	1✔
224	str(doc.meta[key]) for key in self.meta_fields_to_embed if key in doc.meta and doc.meta[key] is not None
225	]
226
227	text_to_embed = (	1✔
228	self.prefix + self.embedding_separator.join(meta_values_to_embed + [doc.content or ""]) + self.suffix
229	)
230
231	texts_to_embed.append(text_to_embed)	1✔
232	return texts_to_embed	1✔
233
234	def _embed_batch(self, texts_to_embed: List[str], batch_size: int) -> List[List[float]]:	1✔
235	"""
236	Embed a list of texts in batches.
237	"""
238	truncate = self.truncate	1✔
239	normalize = self.normalize	1✔
240
241	if self.api_type == HFEmbeddingAPIType.SERVERLESS_INFERENCE_API:	1✔
242	if truncate is not None:	1✔
243	msg = "`truncate` parameter is not supported for Serverless Inference API. It will be ignored."	1✔
244	warnings.warn(msg)	1✔
245	truncate = None	1✔
246	if normalize is not None:	1✔
247	msg = "`normalize` parameter is not supported for Serverless Inference API. It will be ignored."	1✔
248	warnings.warn(msg)	1✔
249	normalize = None	1✔
250
251	all_embeddings = []	1✔
252	for i in tqdm(	1✔
253	range(0, len(texts_to_embed), batch_size), disable=not self.progress_bar, desc="Calculating embeddings"
254	):
255	batch = texts_to_embed[i : i + batch_size]	1✔
256
257	np_embeddings = self._client.feature_extraction(	1✔
258	# this method does not officially support list of strings, but works as expected
259	text=batch, # type: ignore[arg-type]
260	truncate=truncate,
261	normalize=normalize,
262	)
263
264	if np_embeddings.ndim != 2 or np_embeddings.shape[0] != len(batch):	1✔
265	raise ValueError(f"Expected embedding shape ({batch_size}, embedding_dim), got {np_embeddings.shape}")	1✔
266
267	all_embeddings.extend(np_embeddings.tolist())	1✔
268
269	return all_embeddings	1✔
270
271	@component.output_types(documents=List[Document])	1✔
272	def run(self, documents: List[Document]):	1✔
273	"""
274	Embeds a list of documents.
275
276	:param documents:
277	Documents to embed.
278
279	:returns:
280	A dictionary with the following keys:
281	- `documents`: A list of documents with embeddings.
282	"""
283	if not isinstance(documents, list) or documents and not isinstance(documents[0], Document):	1✔
284	raise TypeError(	×
285	"HuggingFaceAPIDocumentEmbedder expects a list of Documents as input."
286	" In case you want to embed a string, please use the HuggingFaceAPITextEmbedder."
287	)
288
289	texts_to_embed = self._prepare_texts_to_embed(documents=documents)	1✔
290
291	embeddings = self._embed_batch(texts_to_embed=texts_to_embed, batch_size=self.batch_size)	1✔
292
293	for doc, emb in zip(documents, embeddings):	1✔
294	doc.embedding = emb	1✔
295
296	return {"documents": documents}	1✔

deepset-ai / haystack / 13972131258

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