Kilokem/Reinforced-Learning-Godot
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
main
Reinforced-Learning-Godot/rl/Lib/site-packages/huggingface_hub/inference/_client.py
Kilokem 40e2a747cf I am done
2024-10-30 22:14:35 +01:00

3145 lines
143 KiB
Python
Raw Permalink Blame History

# coding=utf-8
# Copyright 2023-present, the HuggingFace Inc. team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
# Related resources:
# https://huggingface.co/tasks
# https://huggingface.co/docs/huggingface.js/inference/README
# https://github.com/huggingface/huggingface.js/tree/main/packages/inference/src
# https://github.com/huggingface/text-generation-inference/tree/main/clients/python
# https://github.com/huggingface/text-generation-inference/blob/main/clients/python/text_generation/client.py
# https://huggingface.slack.com/archives/C03E4DQ9LAJ/p1680169099087869
# https://github.com/huggingface/unity-api#tasks
#
# Some TODO:
# - add all tasks
#
# NOTE: the philosophy of this client is "let's make it as easy as possible to use it, even if less optimized". Some
# examples of how it translates:
# - Timeout / Server unavailable is handled by the client in a single "timeout" parameter.
# - Files can be provided as bytes, file paths, or URLs and the client will try to "guess" the type.
# - Images are parsed as PIL.Image for easier manipulation.
# - Provides a "recommended model" for each task => suboptimal but user-wise quicker to get a first script running.
# - Only the main parameters are publicly exposed. Power users can always read the docs for more options.
import base64
import logging
import re
import time
import warnings
from typing import TYPE_CHECKING, Any, Dict, Iterable, List, Literal, Optional, Union, overload
from requests import HTTPError
from requests.structures import CaseInsensitiveDict
from huggingface_hub.constants import ALL_INFERENCE_API_FRAMEWORKS, INFERENCE_ENDPOINT, MAIN_INFERENCE_API_FRAMEWORKS
from huggingface_hub.errors import BadRequestError, InferenceTimeoutError
from huggingface_hub.inference._common import (
TASKS_EXPECTING_IMAGES,
ContentT,
ModelStatus,
_b64_encode,
_b64_to_image,
_bytes_to_dict,
_bytes_to_image,
_bytes_to_list,
_fetch_recommended_models,
_get_unsupported_text_generation_kwargs,
_import_numpy,
_open_as_binary,
_prepare_payload,
_set_unsupported_text_generation_kwargs,
_stream_chat_completion_response,
_stream_text_generation_response,
raise_text_generation_error,
)
from huggingface_hub.inference._generated.types import (
AudioClassificationOutputElement,
AudioClassificationOutputTransform,
AudioToAudioOutputElement,
AutomaticSpeechRecognitionOutput,
ChatCompletionInputGrammarType,
ChatCompletionInputStreamOptions,
ChatCompletionInputToolType,
ChatCompletionOutput,
ChatCompletionStreamOutput,
DocumentQuestionAnsweringOutputElement,
FillMaskOutputElement,
ImageClassificationOutputElement,
ImageSegmentationOutputElement,
ImageToTextOutput,
ObjectDetectionOutputElement,
QuestionAnsweringOutputElement,
SummarizationOutput,
TableQuestionAnsweringOutputElement,
TextClassificationOutputElement,
TextClassificationOutputTransform,
TextGenerationInputGrammarType,
TextGenerationOutput,
TextGenerationStreamOutput,
TextToImageTargetSize,
TextToSpeechEarlyStoppingEnum,
TokenClassificationOutputElement,
ToolElement,
TranslationOutput,
VisualQuestionAnsweringOutputElement,
ZeroShotClassificationOutputElement,
ZeroShotImageClassificationOutputElement,
)
from huggingface_hub.utils import build_hf_headers, get_session, hf_raise_for_status
from huggingface_hub.utils._deprecation import _deprecate_arguments
if TYPE_CHECKING:
import numpy as np
from PIL.Image import Image
logger = logging.getLogger(__name__)
MODEL_KWARGS_NOT_USED_REGEX = re.compile(r"The following `model_kwargs` are not used by the model: \[(.*?)\]")
class InferenceClient:
"""
Initialize a new Inference Client.
[`InferenceClient`] aims to provide a unified experience to perform inference. The client can be used
seamlessly with either the (free) Inference API or self-hosted Inference Endpoints.
Args:
model (`str`, `optional`):
The model to run inference with. Can be a model id hosted on the Hugging Face Hub, e.g. `meta-llama/Meta-Llama-3-8B-Instruct`
or a URL to a deployed Inference Endpoint. Defaults to None, in which case a recommended model is
automatically selected for the task.
Note: for better compatibility with OpenAI's client, `model` has been aliased as `base_url`. Those 2
arguments are mutually exclusive. If using `base_url` for chat completion, the `/chat/completions` suffix
path will be appended to the base URL (see the [TGI Messages API](https://huggingface.co/docs/text-generation-inference/en/messages_api)
documentation for details). When passing a URL as `model`, the client will not append any suffix path to it.
token (`str` or `bool`, *optional*):
Hugging Face token. Will default to the locally saved token if not provided.
Pass `token=False` if you don't want to send your token to the server.
Note: for better compatibility with OpenAI's client, `token` has been aliased as `api_key`. Those 2
arguments are mutually exclusive and have the exact same behavior.
timeout (`float`, `optional`):
The maximum number of seconds to wait for a response from the server. Loading a new model in Inference
API can take up to several minutes. Defaults to None, meaning it will loop until the server is available.
headers (`Dict[str, str]`, `optional`):
Additional headers to send to the server. By default only the authorization and user-agent headers are sent.
Values in this dictionary will override the default values.
cookies (`Dict[str, str]`, `optional`):
Additional cookies to send to the server.
proxies (`Any`, `optional`):
Proxies to use for the request.
base_url (`str`, `optional`):
Base URL to run inference. This is a duplicated argument from `model` to make [`InferenceClient`]
follow the same pattern as `openai.OpenAI` client. Cannot be used if `model` is set. Defaults to None.
api_key (`str`, `optional`):
Token to use for authentication. This is a duplicated argument from `token` to make [`InferenceClient`]
follow the same pattern as `openai.OpenAI` client. Cannot be used if `token` is set. Defaults to None.
"""
def __init__(
self,
model: Optional[str] = None,
*,
token: Union[str, bool, None] = None,
timeout: Optional[float] = None,
headers: Optional[Dict[str, str]] = None,
cookies: Optional[Dict[str, str]] = None,
proxies: Optional[Any] = None,
# OpenAI compatibility
base_url: Optional[str] = None,
api_key: Optional[str] = None,
) -> None:
if model is not None and base_url is not None:
raise ValueError(
"Received both `model` and `base_url` arguments. Please provide only one of them."
" `base_url` is an alias for `model` to make the API compatible with OpenAI's client."
" If using `base_url` for chat completion, the `/chat/completions` suffix path will be appended to the base url."
" When passing a URL as `model`, the client will not append any suffix path to it."
)
if token is not None and api_key is not None:
raise ValueError(
"Received both `token` and `api_key` arguments. Please provide only one of them."
" `api_key` is an alias for `token` to make the API compatible with OpenAI's client."
" It has the exact same behavior as `token`."
)
self.model: Optional[str] = model
self.token: Union[str, bool, None] = token if token is not None else api_key
self.headers = CaseInsensitiveDict(build_hf_headers(token=self.token)) # 'authorization' + 'user-agent'
if headers is not None:
self.headers.update(headers)
self.cookies = cookies
self.timeout = timeout
self.proxies = proxies
# OpenAI compatibility
self.base_url = base_url
def __repr__(self):
return f"<InferenceClient(model='{self.model if self.model else ''}', timeout={self.timeout})>"
@overload
def post( # type: ignore[misc]
self,
*,
json: Optional[Union[str, Dict, List]] = None,
data: Optional[ContentT] = None,
model: Optional[str] = None,
task: Optional[str] = None,
stream: Literal[False] = ...,
) -> bytes: ...
@overload
def post( # type: ignore[misc]
self,
*,
json: Optional[Union[str, Dict, List]] = None,
data: Optional[ContentT] = None,
model: Optional[str] = None,
task: Optional[str] = None,
stream: Literal[True] = ...,
) -> Iterable[bytes]: ...
@overload
def post(
self,
*,
json: Optional[Union[str, Dict, List]] = None,
data: Optional[ContentT] = None,
model: Optional[str] = None,
task: Optional[str] = None,
stream: bool = False,
) -> Union[bytes, Iterable[bytes]]: ...
def post(
self,
*,
json: Optional[Union[str, Dict, List]] = None,
data: Optional[ContentT] = None,
model: Optional[str] = None,
task: Optional[str] = None,
stream: bool = False,
) -> Union[bytes, Iterable[bytes]]:
"""
Make a POST request to the inference server.
Args:
json (`Union[str, Dict, List]`, *optional*):
The JSON data to send in the request body, specific to each task. Defaults to None.
data (`Union[str, Path, bytes, BinaryIO]`, *optional*):
The content to send in the request body, specific to each task.
It can be raw bytes, a pointer to an opened file, a local file path,
or a URL to an online resource (image, audio file,...). If both `json` and `data` are passed,
`data` will take precedence. At least `json` or `data` must be provided. Defaults to None.
model (`str`, *optional*):
The model to use for inference. Can be a model ID hosted on the Hugging Face Hub or a URL to a deployed
Inference Endpoint. Will override the model defined at the instance level. Defaults to None.
task (`str`, *optional*):
The task to perform on the inference. All available tasks can be found
[here](https://huggingface.co/tasks). Used only to default to a recommended model if `model` is not
provided. At least `model` or `task` must be provided. Defaults to None.
stream (`bool`, *optional*):
Whether to iterate over streaming APIs.
Returns:
bytes: The raw bytes returned by the server.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
"""
url = self._resolve_url(model, task)
if data is not None and json is not None:
warnings.warn("Ignoring `json` as `data` is passed as binary.")
# Set Accept header if relevant
headers = self.headers.copy()
if task in TASKS_EXPECTING_IMAGES and "Accept" not in headers:
headers["Accept"] = "image/png"
t0 = time.time()
timeout = self.timeout
while True:
with _open_as_binary(data) as data_as_binary:
try:
response = get_session().post(
url,
json=json,
data=data_as_binary,
headers=headers,
cookies=self.cookies,
timeout=self.timeout,
stream=stream,
proxies=self.proxies,
)
except TimeoutError as error:
# Convert any `TimeoutError` to a `InferenceTimeoutError`
raise InferenceTimeoutError(f"Inference call timed out: {url}") from error # type: ignore
try:
hf_raise_for_status(response)
return response.iter_lines() if stream else response.content
except HTTPError as error:
if error.response.status_code == 422 and task is not None:
error.args = (
f"{error.args[0]}\nMake sure '{task}' task is supported by the model.",
) + error.args[1:]
if error.response.status_code == 503:
# If Model is unavailable, either raise a TimeoutError...
if timeout is not None and time.time() - t0 > timeout:
raise InferenceTimeoutError(
f"Model not loaded on the server: {url}. Please retry with a higher timeout (current:"
f" {self.timeout}).",
request=error.request,
response=error.response,
) from error
# ...or wait 1s and retry
logger.info(f"Waiting for model to be loaded on the server: {error}")
time.sleep(1)
if "X-wait-for-model" not in headers and url.startswith(INFERENCE_ENDPOINT):
headers["X-wait-for-model"] = "1"
if timeout is not None:
timeout = max(self.timeout - (time.time() - t0), 1) # type: ignore
continue
raise
def audio_classification(
self,
audio: ContentT,
*,
model: Optional[str] = None,
top_k: Optional[int] = None,
function_to_apply: Optional["AudioClassificationOutputTransform"] = None,
) -> List[AudioClassificationOutputElement]:
"""
Perform audio classification on the provided audio content.
Args:
audio (Union[str, Path, bytes, BinaryIO]):
The audio content to classify. It can be raw audio bytes, a local audio file, or a URL pointing to an
audio file.
model (`str`, *optional*):
The model to use for audio classification. Can be a model ID hosted on the Hugging Face Hub
or a URL to a deployed Inference Endpoint. If not provided, the default recommended model for
audio classification will be used.
top_k (`int`, *optional*):
When specified, limits the output to the top K most probable classes.
function_to_apply (`"AudioClassificationOutputTransform"`, *optional*):
The function to apply to the output.
Returns:
`List[AudioClassificationOutputElement]`: List of [`AudioClassificationOutputElement`] items containing the predicted labels and their confidence.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.audio_classification("audio.flac")
[
AudioClassificationOutputElement(score=0.4976358711719513, label='hap'),
AudioClassificationOutputElement(score=0.3677836060523987, label='neu'),
...
]
```
"""
parameters = {"function_to_apply": function_to_apply, "top_k": top_k}
payload = _prepare_payload(audio, parameters=parameters, expect_binary=True)
response = self.post(**payload, model=model, task="audio-classification")
return AudioClassificationOutputElement.parse_obj_as_list(response)
def audio_to_audio(
self,
audio: ContentT,
*,
model: Optional[str] = None,
) -> List[AudioToAudioOutputElement]:
"""
Performs multiple tasks related to audio-to-audio depending on the model (eg: speech enhancement, source separation).
Args:
audio (Union[str, Path, bytes, BinaryIO]):
The audio content for the model. It can be raw audio bytes, a local audio file, or a URL pointing to an
audio file.
model (`str`, *optional*):
The model can be any model which takes an audio file and returns another audio file. Can be a model ID hosted on the Hugging Face Hub
or a URL to a deployed Inference Endpoint. If not provided, the default recommended model for
audio_to_audio will be used.
Returns:
`List[AudioToAudioOutputElement]`: A list of [`AudioToAudioOutputElement`] items containing audios label, content-type, and audio content in blob.
Raises:
`InferenceTimeoutError`:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> audio_output = client.audio_to_audio("audio.flac")
>>> for i, item in enumerate(audio_output):
>>> with open(f"output_{i}.flac", "wb") as f:
f.write(item.blob)
```
"""
response = self.post(data=audio, model=model, task="audio-to-audio")
audio_output = AudioToAudioOutputElement.parse_obj_as_list(response)
for item in audio_output:
item.blob = base64.b64decode(item.blob)
return audio_output
def automatic_speech_recognition(
self,
audio: ContentT,
*,
model: Optional[str] = None,
) -> AutomaticSpeechRecognitionOutput:
"""
Perform automatic speech recognition (ASR or audio-to-text) on the given audio content.
Args:
audio (Union[str, Path, bytes, BinaryIO]):
The content to transcribe. It can be raw audio bytes, local audio file, or a URL to an audio file.
model (`str`, *optional*):
The model to use for ASR. Can be a model ID hosted on the Hugging Face Hub or a URL to a deployed
Inference Endpoint. If not provided, the default recommended model for ASR will be used.
Returns:
[`AutomaticSpeechRecognitionOutput`]: An item containing the transcribed text and optionally the timestamp chunks.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.automatic_speech_recognition("hello_world.flac").text
"hello world"
```
"""
response = self.post(data=audio, model=model, task="automatic-speech-recognition")
return AutomaticSpeechRecognitionOutput.parse_obj_as_instance(response)
@overload
def chat_completion( # type: ignore
self,
messages: List[Dict],
*,
model: Optional[str] = None,
stream: Literal[False] = False,
frequency_penalty: Optional[float] = None,
logit_bias: Optional[List[float]] = None,
logprobs: Optional[bool] = None,
max_tokens: Optional[int] = None,
n: Optional[int] = None,
presence_penalty: Optional[float] = None,
response_format: Optional[ChatCompletionInputGrammarType] = None,
seed: Optional[int] = None,
stop: Optional[List[str]] = None,
stream_options: Optional[ChatCompletionInputStreamOptions] = None,
temperature: Optional[float] = None,
tool_choice: Optional[Union[ChatCompletionInputToolType, str]] = None,
tool_prompt: Optional[str] = None,
tools: Optional[List[ToolElement]] = None,
top_logprobs: Optional[int] = None,
top_p: Optional[float] = None,
) -> ChatCompletionOutput: ...
@overload
def chat_completion( # type: ignore
self,
messages: List[Dict],
*,
model: Optional[str] = None,
stream: Literal[True] = True,
frequency_penalty: Optional[float] = None,
logit_bias: Optional[List[float]] = None,
logprobs: Optional[bool] = None,
max_tokens: Optional[int] = None,
n: Optional[int] = None,
presence_penalty: Optional[float] = None,
response_format: Optional[ChatCompletionInputGrammarType] = None,
seed: Optional[int] = None,
stop: Optional[List[str]] = None,
stream_options: Optional[ChatCompletionInputStreamOptions] = None,
temperature: Optional[float] = None,
tool_choice: Optional[Union[ChatCompletionInputToolType, str]] = None,
tool_prompt: Optional[str] = None,
tools: Optional[List[ToolElement]] = None,
top_logprobs: Optional[int] = None,
top_p: Optional[float] = None,
) -> Iterable[ChatCompletionStreamOutput]: ...
@overload
def chat_completion(
self,
messages: List[Dict],
*,
model: Optional[str] = None,
stream: bool = False,
frequency_penalty: Optional[float] = None,
logit_bias: Optional[List[float]] = None,
logprobs: Optional[bool] = None,
max_tokens: Optional[int] = None,
n: Optional[int] = None,
presence_penalty: Optional[float] = None,
response_format: Optional[ChatCompletionInputGrammarType] = None,
seed: Optional[int] = None,
stop: Optional[List[str]] = None,
stream_options: Optional[ChatCompletionInputStreamOptions] = None,
temperature: Optional[float] = None,
tool_choice: Optional[Union[ChatCompletionInputToolType, str]] = None,
tool_prompt: Optional[str] = None,
tools: Optional[List[ToolElement]] = None,
top_logprobs: Optional[int] = None,
top_p: Optional[float] = None,
) -> Union[ChatCompletionOutput, Iterable[ChatCompletionStreamOutput]]: ...
def chat_completion(
self,
messages: List[Dict],
*,
model: Optional[str] = None,
stream: bool = False,
# Parameters from ChatCompletionInput (handled manually)
frequency_penalty: Optional[float] = None,
logit_bias: Optional[List[float]] = None,
logprobs: Optional[bool] = None,
max_tokens: Optional[int] = None,
n: Optional[int] = None,
presence_penalty: Optional[float] = None,
response_format: Optional[ChatCompletionInputGrammarType] = None,
seed: Optional[int] = None,
stop: Optional[List[str]] = None,
stream_options: Optional[ChatCompletionInputStreamOptions] = None,
temperature: Optional[float] = None,
tool_choice: Optional[Union[ChatCompletionInputToolType, str]] = None,
tool_prompt: Optional[str] = None,
tools: Optional[List[ToolElement]] = None,
top_logprobs: Optional[int] = None,
top_p: Optional[float] = None,
) -> Union[ChatCompletionOutput, Iterable[ChatCompletionStreamOutput]]:
"""
A method for completing conversations using a specified language model.
<Tip>
The `client.chat_completion` method is aliased as `client.chat.completions.create` for compatibility with OpenAI's client.
Inputs and outputs are strictly the same and using either syntax will yield the same results.
Check out the [Inference guide](https://huggingface.co/docs/huggingface_hub/guides/inference#openai-compatibility)
for more details about OpenAI's compatibility.
</Tip>
Args:
messages (List of [`ChatCompletionInputMessage`]):
Conversation history consisting of roles and content pairs.
model (`str`, *optional*):
The model to use for chat-completion. Can be a model ID hosted on the Hugging Face Hub or a URL to a deployed
Inference Endpoint. If not provided, the default recommended model for chat-based text-generation will be used.
See https://huggingface.co/tasks/text-generation for more details.
If `model` is a model ID, it is passed to the server as the `model` parameter. If you want to define a
custom URL while setting `model` in the request payload, you must set `base_url` when initializing [`InferenceClient`].
frequency_penalty (`float`, *optional*):
Penalizes new tokens based on their existing frequency
in the text so far. Range: [-2.0, 2.0]. Defaults to 0.0.
logit_bias (`List[float]`, *optional*):
Modify the likelihood of specified tokens appearing in the completion. Accepts a JSON object that maps tokens
(specified by their token ID in the tokenizer) to an associated bias value from -100 to 100. Mathematically,
the bias is added to the logits generated by the model prior to sampling. The exact effect will vary per model,
but values between -1 and 1 should decrease or increase likelihood of selection; values like -100 or 100 should
result in a ban or exclusive selection of the relevant token. Defaults to None.
logprobs (`bool`, *optional*):
Whether to return log probabilities of the output tokens or not. If true, returns the log
probabilities of each output token returned in the content of message.
max_tokens (`int`, *optional*):
Maximum number of tokens allowed in the response. Defaults to 20.
n (`int`, *optional*):
UNUSED.
presence_penalty (`float`, *optional*):
Number between -2.0 and 2.0. Positive values penalize new tokens based on whether they appear in the
text so far, increasing the model's likelihood to talk about new topics.
response_format ([`ChatCompletionInputGrammarType`], *optional*):
Grammar constraints. Can be either a JSONSchema or a regex.
seed (Optional[`int`], *optional*):
Seed for reproducible control flow. Defaults to None.
stop (Optional[`str`], *optional*):
Up to four strings which trigger the end of the response.
Defaults to None.
stream (`bool`, *optional*):
Enable realtime streaming of responses. Defaults to False.
stream_options ([`ChatCompletionInputStreamOptions`], *optional*):
Options for streaming completions.
temperature (`float`, *optional*):
Controls randomness of the generations. Lower values ensure
less random completions. Range: [0, 2]. Defaults to 1.0.
top_logprobs (`int`, *optional*):
An integer between 0 and 5 specifying the number of most likely tokens to return at each token
position, each with an associated log probability. logprobs must be set to true if this parameter is
used.
top_p (`float`, *optional*):
Fraction of the most likely next words to sample from.
Must be between 0 and 1. Defaults to 1.0.
tool_choice ([`ChatCompletionInputToolType`] or `str`, *optional*):
The tool to use for the completion. Defaults to "auto".
tool_prompt (`str`, *optional*):
A prompt to be appended before the tools.
tools (List of [`ToolElement`], *optional*):
A list of tools the model may call. Currently, only functions are supported as a tool. Use this to
provide a list of functions the model may generate JSON inputs for.
Returns:
[`ChatCompletionOutput`] or Iterable of [`ChatCompletionStreamOutput`]:
Generated text returned from the server:
- if `stream=False`, the generated text is returned as a [`ChatCompletionOutput`] (default).
- if `stream=True`, the generated text is returned token by token as a sequence of [`ChatCompletionStreamOutput`].
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> messages = [{"role": "user", "content": "What is the capital of France?"}]
>>> client = InferenceClient("meta-llama/Meta-Llama-3-8B-Instruct")
>>> client.chat_completion(messages, max_tokens=100)
ChatCompletionOutput(
choices=[
ChatCompletionOutputComplete(
finish_reason='eos_token',
index=0,
message=ChatCompletionOutputMessage(
role='assistant',
content='The capital of France is Paris.',
name=None,
tool_calls=None
),
logprobs=None
)
],
created=1719907176,
id='',
model='meta-llama/Meta-Llama-3-8B-Instruct',
object='text_completion',
system_fingerprint='2.0.4-sha-f426a33',
usage=ChatCompletionOutputUsage(
completion_tokens=8,
prompt_tokens=17,
total_tokens=25
)
)
```
Example using streaming:
```py
>>> from huggingface_hub import InferenceClient
>>> messages = [{"role": "user", "content": "What is the capital of France?"}]
>>> client = InferenceClient("meta-llama/Meta-Llama-3-8B-Instruct")
>>> for token in client.chat_completion(messages, max_tokens=10, stream=True):
... print(token)
ChatCompletionStreamOutput(choices=[ChatCompletionStreamOutputChoice(delta=ChatCompletionStreamOutputDelta(content='The', role='assistant'), index=0, finish_reason=None)], created=1710498504)
ChatCompletionStreamOutput(choices=[ChatCompletionStreamOutputChoice(delta=ChatCompletionStreamOutputDelta(content=' capital', role='assistant'), index=0, finish_reason=None)], created=1710498504)
(...)
ChatCompletionStreamOutput(choices=[ChatCompletionStreamOutputChoice(delta=ChatCompletionStreamOutputDelta(content=' may', role='assistant'), index=0, finish_reason=None)], created=1710498504)
```
Example using OpenAI's syntax:
```py
# instead of `from openai import OpenAI`
from huggingface_hub import InferenceClient
# instead of `client = OpenAI(...)`
client = InferenceClient(
base_url=...,
api_key=...,
)
output = client.chat.completions.create(
model="meta-llama/Meta-Llama-3-8B-Instruct",
messages=[
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "Count to 10"},
],
stream=True,
max_tokens=1024,
)
for chunk in output:
print(chunk.choices[0].delta.content)
```
Example using Image + Text as input:
```py
>>> from huggingface_hub import InferenceClient
# provide a remote URL
>>> image_url ="https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
# or a base64-encoded image
>>> image_path = "/path/to/image.jpeg"
>>> with open(image_path, "rb") as f:
... base64_image = base64.b64encode(f.read()).decode("utf-8")
>>> image_url = f"data:image/jpeg;base64,{base64_image}"
>>> client = InferenceClient("meta-llama/Llama-3.2-11B-Vision-Instruct")
>>> output = client.chat.completions.create(
... messages=[
... {
... "role": "user",
... "content": [
... {
... "type": "image_url",
... "image_url": {"url": image_url},
... },
... {
... "type": "text",
... "text": "Describe this image in one sentence.",
... },
... ],
... },
... ],
... )
>>> output
The image depicts the iconic Statue of Liberty situated in New York Harbor, New York, on a clear day.
```
Example using tools:
```py
>>> client = InferenceClient("meta-llama/Meta-Llama-3-70B-Instruct")
>>> messages = [
... {
... "role": "system",
... "content": "Don't make assumptions about what values to plug into functions. Ask for clarification if a user request is ambiguous.",
... },
... {
... "role": "user",
... "content": "What's the weather like the next 3 days in San Francisco, CA?",
... },
... ]
>>> tools = [
... {
... "type": "function",
... "function": {
... "name": "get_current_weather",
... "description": "Get the current weather",
... "parameters": {
... "type": "object",
... "properties": {
... "location": {
... "type": "string",
... "description": "The city and state, e.g. San Francisco, CA",
... },
... "format": {
... "type": "string",
... "enum": ["celsius", "fahrenheit"],
... "description": "The temperature unit to use. Infer this from the users location.",
... },
... },
... "required": ["location", "format"],
... },
... },
... },
... {
... "type": "function",
... "function": {
... "name": "get_n_day_weather_forecast",
... "description": "Get an N-day weather forecast",
... "parameters": {
... "type": "object",
... "properties": {
... "location": {
... "type": "string",
... "description": "The city and state, e.g. San Francisco, CA",
... },
... "format": {
... "type": "string",
... "enum": ["celsius", "fahrenheit"],
... "description": "The temperature unit to use. Infer this from the users location.",
... },
... "num_days": {
... "type": "integer",
... "description": "The number of days to forecast",
... },
... },
... "required": ["location", "format", "num_days"],
... },
... },
... },
... ]
>>> response = client.chat_completion(
... model="meta-llama/Meta-Llama-3-70B-Instruct",
... messages=messages,
... tools=tools,
... tool_choice="auto",
... max_tokens=500,
... )
>>> response.choices[0].message.tool_calls[0].function
ChatCompletionOutputFunctionDefinition(
arguments={
'location': 'San Francisco, CA',
'format': 'fahrenheit',
'num_days': 3
},
name='get_n_day_weather_forecast',
description=None
)
```
Example using response_format:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient("meta-llama/Meta-Llama-3-70B-Instruct")
>>> messages = [
... {
... "role": "user",
... "content": "I saw a puppy a cat and a raccoon during my bike ride in the park. What did I saw and when?",
... },
... ]
>>> response_format = {
... "type": "json",
... "value": {
... "properties": {
... "location": {"type": "string"},
... "activity": {"type": "string"},
... "animals_seen": {"type": "integer", "minimum": 1, "maximum": 5},
... "animals": {"type": "array", "items": {"type": "string"}},
... },
... "required": ["location", "activity", "animals_seen", "animals"],
... },
... }
>>> response = client.chat_completion(
... messages=messages,
... response_format=response_format,
... max_tokens=500,
)
>>> response.choices[0].message.content
'{\n\n"activity": "bike ride",\n"animals": ["puppy", "cat", "raccoon"],\n"animals_seen": 3,\n"location": "park"}'
```
"""
model_url = self._resolve_chat_completion_url(model)
# `model` is sent in the payload. Not used by the server but can be useful for debugging/routing.
# If it's a ID on the Hub => use it. Otherwise, we use a random string.
model_id = model or self.model or "tgi"
if model_id.startswith(("http://", "https://")):
model_id = "tgi" # dummy value
payload = dict(
model=model_id,
messages=messages,
frequency_penalty=frequency_penalty,
logit_bias=logit_bias,
logprobs=logprobs,
max_tokens=max_tokens,
n=n,
presence_penalty=presence_penalty,
response_format=response_format,
seed=seed,
stop=stop,
temperature=temperature,
tool_choice=tool_choice,
tool_prompt=tool_prompt,
tools=tools,
top_logprobs=top_logprobs,
top_p=top_p,
stream=stream,
stream_options=stream_options,
)
payload = {key: value for key, value in payload.items() if value is not None}
data = self.post(model=model_url, json=payload, stream=stream)
if stream:
return _stream_chat_completion_response(data) # type: ignore[arg-type]
return ChatCompletionOutput.parse_obj_as_instance(data) # type: ignore[arg-type]
def _resolve_chat_completion_url(self, model: Optional[str] = None) -> str:
# Since `chat_completion(..., model=xxx)` is also a payload parameter for the server, we need to handle 'model' differently.
# `self.base_url` and `self.model` takes precedence over 'model' argument only in `chat_completion`.
model_id_or_url = self.base_url or self.model or model or self.get_recommended_model("text-generation")
# Resolve URL if it's a model ID
model_url = (
model_id_or_url
if model_id_or_url.startswith(("http://", "https://"))
else self._resolve_url(model_id_or_url, task="text-generation")
)
# Strip trailing /
model_url = model_url.rstrip("/")
# Append /chat/completions if not already present
if model_url.endswith("/v1"):
model_url += "/chat/completions"
# Append /v1/chat/completions if not already present
if not model_url.endswith("/chat/completions"):
model_url += "/v1/chat/completions"
return model_url
def document_question_answering(
self,
image: ContentT,
question: str,
*,
model: Optional[str] = None,
doc_stride: Optional[int] = None,
handle_impossible_answer: Optional[bool] = None,
lang: Optional[str] = None,
max_answer_len: Optional[int] = None,
max_question_len: Optional[int] = None,
max_seq_len: Optional[int] = None,
top_k: Optional[int] = None,
word_boxes: Optional[List[Union[List[float], str]]] = None,
) -> List[DocumentQuestionAnsweringOutputElement]:
"""
Answer questions on document images.
Args:
image (`Union[str, Path, bytes, BinaryIO]`):
The input image for the context. It can be raw bytes, an image file, or a URL to an online image.
question (`str`):
Question to be answered.
model (`str`, *optional*):
The model to use for the document question answering task. Can be a model ID hosted on the Hugging Face Hub or a URL to
a deployed Inference Endpoint. If not provided, the default recommended document question answering model will be used.
Defaults to None.
doc_stride (`int`, *optional*):
If the words in the document are too long to fit with the question for the model, it will
be split in several chunks with some overlap. This argument controls the size of that
overlap.
handle_impossible_answer (`bool`, *optional*):
Whether to accept impossible as an answer.
lang (`str`, *optional*):
Language to use while running OCR.
max_answer_len (`int`, *optional*):
The maximum length of predicted answers (e.g., only answers with a shorter length are
considered).
max_question_len (`int`, *optional*):
The maximum length of the question after tokenization. It will be truncated if needed.
max_seq_len (`int`, *optional*):
The maximum length of the total sentence (context + question) in tokens of each chunk
passed to the model. The context will be split in several chunks (using doc_stride as
overlap) if needed.
top_k (`int`, *optional*):
The number of answers to return (will be chosen by order of likelihood). Can return less
than top_k answers if there are not enough options available within the context.
word_boxes (`List[Union[List[float], str]]`, *optional*):
A list of words and bounding boxes (normalized 0->1000). If provided, the inference will
skip the OCR step and use the provided bounding boxes instead.
Returns:
`List[DocumentQuestionAnsweringOutputElement]`: a list of [`DocumentQuestionAnsweringOutputElement`] items containing the predicted label, associated probability, word ids, and page number.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.document_question_answering(image="https://huggingface.co/spaces/impira/docquery/resolve/2359223c1837a7587402bda0f2643382a6eefeab/invoice.png", question="What is the invoice number?")
[DocumentQuestionAnsweringOutputElement(answer='us-001', end=16, score=0.9999666213989258, start=16, words=None)]
```
"""
inputs: Dict[str, Any] = {"question": question, "image": _b64_encode(image)}
parameters = {
"doc_stride": doc_stride,
"handle_impossible_answer": handle_impossible_answer,
"lang": lang,
"max_answer_len": max_answer_len,
"max_question_len": max_question_len,
"max_seq_len": max_seq_len,
"top_k": top_k,
"word_boxes": word_boxes,
}
payload = _prepare_payload(inputs, parameters=parameters)
response = self.post(**payload, model=model, task="document-question-answering")
return DocumentQuestionAnsweringOutputElement.parse_obj_as_list(response)
def feature_extraction(
self,
text: str,
*,
normalize: Optional[bool] = None,
prompt_name: Optional[str] = None,
truncate: Optional[bool] = None,
truncation_direction: Optional[Literal["Left", "Right"]] = None,
model: Optional[str] = None,
) -> "np.ndarray":
"""
Generate embeddings for a given text.
Args:
text (`str`):
The text to embed.
model (`str`, *optional*):
The model to use for the conversational task. Can be a model ID hosted on the Hugging Face Hub or a URL to
a deployed Inference Endpoint. If not provided, the default recommended conversational model will be used.
Defaults to None.
normalize (`bool`, *optional*):
Whether to normalize the embeddings or not.
Only available on server powered by Text-Embedding-Inference.
prompt_name (`str`, *optional*):
The name of the prompt that should be used by for encoding. If not set, no prompt will be applied.
Must be a key in the `Sentence Transformers` configuration `prompts` dictionary.
For example if ``prompt_name`` is "query" and the ``prompts`` is {"query": "query: ",...},
then the sentence "What is the capital of France?" will be encoded as "query: What is the capital of France?"
because the prompt text will be prepended before any text to encode.
truncate (`bool`, *optional*):
Whether to truncate the embeddings or not.
Only available on server powered by Text-Embedding-Inference.
truncation_direction (`Literal["Left", "Right"]`, *optional*):
Which side of the input should be truncated when `truncate=True` is passed.
Returns:
`np.ndarray`: The embedding representing the input text as a float32 numpy array.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.feature_extraction("Hi, who are you?")
array([[ 2.424802 , 2.93384 , 1.1750331 , ..., 1.240499, -0.13776633, -0.7889173 ],
[-0.42943227, -0.6364878 , -1.693462 , ..., 0.41978157, -2.4336355 , 0.6162071 ],
...,
[ 0.28552425, -0.928395 , -1.2077185 , ..., 0.76810825, -2.1069427 , 0.6236161 ]], dtype=float32)
```
"""
parameters = {
"normalize": normalize,
"prompt_name": prompt_name,
"truncate": truncate,
"truncation_direction": truncation_direction,
}
payload = _prepare_payload(text, parameters=parameters)
response = self.post(**payload, model=model, task="feature-extraction")
np = _import_numpy()
return np.array(_bytes_to_dict(response), dtype="float32")
def fill_mask(
self,
text: str,
*,
model: Optional[str] = None,
targets: Optional[List[str]] = None,
top_k: Optional[int] = None,
) -> List[FillMaskOutputElement]:
"""
Fill in a hole with a missing word (token to be precise).
Args:
text (`str`):
a string to be filled from, must contain the [MASK] token (check model card for exact name of the mask).
model (`str`, *optional*):
The model to use for the fill mask task. Can be a model ID hosted on the Hugging Face Hub or a URL to
a deployed Inference Endpoint. If not provided, the default recommended fill mask model will be used.
targets (`List[str]`, *optional*):
When passed, the model will limit the scores to the passed targets instead of looking up
in the whole vocabulary. If the provided targets are not in the model vocab, they will be
tokenized and the first resulting token will be used (with a warning, and that might be
slower).
top_k (`int`, *optional*):
When passed, overrides the number of predictions to return.
Returns:
`List[FillMaskOutputElement]`: a list of [`FillMaskOutputElement`] items containing the predicted label, associated
probability, token reference, and completed text.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.fill_mask("The goal of life is <mask>.")
[
FillMaskOutputElement(score=0.06897063553333282, token=11098, token_str=' happiness', sequence='The goal of life is happiness.'),
FillMaskOutputElement(score=0.06554922461509705, token=45075, token_str=' immortality', sequence='The goal of life is immortality.')
]
```
"""
parameters = {"targets": targets, "top_k": top_k}
payload = _prepare_payload(text, parameters=parameters)
response = self.post(**payload, model=model, task="fill-mask")
return FillMaskOutputElement.parse_obj_as_list(response)
def image_classification(
self,
image: ContentT,
*,
model: Optional[str] = None,
function_to_apply: Optional[Literal["sigmoid", "softmax", "none"]] = None,
top_k: Optional[int] = None,
) -> List[ImageClassificationOutputElement]:
"""
Perform image classification on the given image using the specified model.
Args:
image (`Union[str, Path, bytes, BinaryIO]`):
The image to classify. It can be raw bytes, an image file, or a URL to an online image.
model (`str`, *optional*):
The model to use for image classification. Can be a model ID hosted on the Hugging Face Hub or a URL to a
deployed Inference Endpoint. If not provided, the default recommended model for image classification will be used.
function_to_apply (`Literal["sigmoid", "softmax", "none"]`, *optional*):
The function to apply to the output scores.
top_k (`int`, *optional*):
When specified, limits the output to the top K most probable classes.
Returns:
`List[ImageClassificationOutputElement]`: a list of [`ImageClassificationOutputElement`] items containing the predicted label and associated probability.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.image_classification("https://upload.wikimedia.org/wikipedia/commons/thumb/4/43/Cute_dog.jpg/320px-Cute_dog.jpg")
[ImageClassificationOutputElement(label='Blenheim spaniel', score=0.9779096841812134), ...]
```
"""
parameters = {"function_to_apply": function_to_apply, "top_k": top_k}
payload = _prepare_payload(image, parameters=parameters, expect_binary=True)
response = self.post(**payload, model=model, task="image-classification")
return ImageClassificationOutputElement.parse_obj_as_list(response)
def image_segmentation(
self,
image: ContentT,
*,
model: Optional[str] = None,
mask_threshold: Optional[float] = None,
overlap_mask_area_threshold: Optional[float] = None,
subtask: Optional[Literal["instance", "panoptic", "semantic"]] = None,
threshold: Optional[float] = None,
) -> List[ImageSegmentationOutputElement]:
"""
Perform image segmentation on the given image using the specified model.
<Tip warning={true}>
You must have `PIL` installed if you want to work with images (`pip install Pillow`).
</Tip>
Args:
image (`Union[str, Path, bytes, BinaryIO]`):
The image to segment. It can be raw bytes, an image file, or a URL to an online image.
model (`str`, *optional*):
The model to use for image segmentation. Can be a model ID hosted on the Hugging Face Hub or a URL to a
deployed Inference Endpoint. If not provided, the default recommended model for image segmentation will be used.
mask_threshold (`float`, *optional*):
Threshold to use when turning the predicted masks into binary values.
overlap_mask_area_threshold (`float`, *optional*):
Mask overlap threshold to eliminate small, disconnected segments.
subtask (`Literal["instance", "panoptic", "semantic"]`, *optional*):
Segmentation task to be performed, depending on model capabilities.
threshold (`float`, *optional*):
Probability threshold to filter out predicted masks.
Returns:
`List[ImageSegmentationOutputElement]`: A list of [`ImageSegmentationOutputElement`] items containing the segmented masks and associated attributes.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.image_segmentation("cat.jpg")
[ImageSegmentationOutputElement(score=0.989008, label='LABEL_184', mask=<PIL.PngImagePlugin.PngImageFile image mode=L size=400x300 at 0x7FDD2B129CC0>), ...]
```
"""
parameters = {
"mask_threshold": mask_threshold,
"overlap_mask_area_threshold": overlap_mask_area_threshold,
"subtask": subtask,
"threshold": threshold,
}
payload = _prepare_payload(image, parameters=parameters, expect_binary=True)
response = self.post(**payload, model=model, task="image-segmentation")
output = ImageSegmentationOutputElement.parse_obj_as_list(response)
for item in output:
item.mask = _b64_to_image(item.mask) # type: ignore [assignment]
return output
def image_to_image(
self,
image: ContentT,
prompt: Optional[str] = None,
*,
negative_prompt: Optional[str] = None,
height: Optional[int] = None,
width: Optional[int] = None,
num_inference_steps: Optional[int] = None,
guidance_scale: Optional[float] = None,
model: Optional[str] = None,
**kwargs,
) -> "Image":
"""
Perform image-to-image translation using a specified model.
<Tip warning={true}>
You must have `PIL` installed if you want to work with images (`pip install Pillow`).
</Tip>
Args:
image (`Union[str, Path, bytes, BinaryIO]`):
The input image for translation. It can be raw bytes, an image file, or a URL to an online image.
prompt (`str`, *optional*):
The text prompt to guide the image generation.
negative_prompt (`str`, *optional*):
A negative prompt to guide the translation process.
height (`int`, *optional*):
The height in pixels of the generated image.
width (`int`, *optional*):
The width in pixels of the generated image.
num_inference_steps (`int`, *optional*):
The number of denoising steps. More denoising steps usually lead to a higher quality image at the
expense of slower inference.
guidance_scale (`float`, *optional*):
Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
usually at the expense of lower image quality.
model (`str`, *optional*):
The model to use for inference. Can be a model ID hosted on the Hugging Face Hub or a URL to a deployed
Inference Endpoint. This parameter overrides the model defined at the instance level. Defaults to None.
Returns:
`Image`: The translated image.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> image = client.image_to_image("cat.jpg", prompt="turn the cat into a tiger")
>>> image.save("tiger.jpg")
```
"""
parameters = {
"prompt": prompt,
"negative_prompt": negative_prompt,
"height": height,
"width": width,
"num_inference_steps": num_inference_steps,
"guidance_scale": guidance_scale,
**kwargs,
}
payload = _prepare_payload(image, parameters=parameters, expect_binary=True)
response = self.post(**payload, model=model, task="image-to-image")
return _bytes_to_image(response)
def image_to_text(self, image: ContentT, *, model: Optional[str] = None) -> ImageToTextOutput:
"""
Takes an input image and return text.
Models can have very different outputs depending on your use case (image captioning, optical character recognition
(OCR), Pix2Struct, etc). Please have a look to the model card to learn more about a model's specificities.
Args:
image (`Union[str, Path, bytes, BinaryIO]`):
The input image to caption. It can be raw bytes, an image file, or a URL to an online image..
model (`str`, *optional*):
The model to use for inference. Can be a model ID hosted on the Hugging Face Hub or a URL to a deployed
Inference Endpoint. This parameter overrides the model defined at the instance level. Defaults to None.
Returns:
[`ImageToTextOutput`]: The generated text.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.image_to_text("cat.jpg")
'a cat standing in a grassy field '
>>> client.image_to_text("https://upload.wikimedia.org/wikipedia/commons/thumb/4/43/Cute_dog.jpg/320px-Cute_dog.jpg")
'a dog laying on the grass next to a flower pot '
```
"""
response = self.post(data=image, model=model, task="image-to-text")
output = ImageToTextOutput.parse_obj(response)
return output[0] if isinstance(output, list) else output
def list_deployed_models(
self, frameworks: Union[None, str, Literal["all"], List[str]] = None
) -> Dict[str, List[str]]:
"""
List models deployed on the Serverless Inference API service.
This helper checks deployed models framework by framework. By default, it will check the 4 main frameworks that
are supported and account for 95% of the hosted models. However, if you want a complete list of models you can
specify `frameworks="all"` as input. Alternatively, if you know before-hand which framework you are interested
in, you can also restrict to search to this one (e.g. `frameworks="text-generation-inference"`). The more
frameworks are checked, the more time it will take.
<Tip warning={true}>
This endpoint method does not return a live list of all models available for the Serverless Inference API service.
It searches over a cached list of models that were recently available and the list may not be up to date.
If you want to know the live status of a specific model, use [`~InferenceClient.get_model_status`].
</Tip>
<Tip>
This endpoint method is mostly useful for discoverability. If you already know which model you want to use and want to
check its availability, you can directly use [`~InferenceClient.get_model_status`].
</Tip>
Args:
frameworks (`Literal["all"]` or `List[str]` or `str`, *optional*):
The frameworks to filter on. By default only a subset of the available frameworks are tested. If set to
"all", all available frameworks will be tested. It is also possible to provide a single framework or a
custom set of frameworks to check.
Returns:
`Dict[str, List[str]]`: A dictionary mapping task names to a sorted list of model IDs.
Example:
```python
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
# Discover zero-shot-classification models currently deployed
>>> models = client.list_deployed_models()
>>> models["zero-shot-classification"]
['Narsil/deberta-large-mnli-zero-cls', 'facebook/bart-large-mnli', ...]
# List from only 1 framework
>>> client.list_deployed_models("text-generation-inference")
{'text-generation': ['bigcode/starcoder', 'meta-llama/Llama-2-70b-chat-hf', ...], ...}
```
"""
# Resolve which frameworks to check
if frameworks is None:
frameworks = MAIN_INFERENCE_API_FRAMEWORKS
elif frameworks == "all":
frameworks = ALL_INFERENCE_API_FRAMEWORKS
elif isinstance(frameworks, str):
frameworks = [frameworks]
frameworks = list(set(frameworks))
# Fetch them iteratively
models_by_task: Dict[str, List[str]] = {}
def _unpack_response(framework: str, items: List[Dict]) -> None:
for model in items:
if framework == "sentence-transformers":
# Model running with the `sentence-transformers` framework can work with both tasks even if not
# branded as such in the API response
models_by_task.setdefault("feature-extraction", []).append(model["model_id"])
models_by_task.setdefault("sentence-similarity", []).append(model["model_id"])
else:
models_by_task.setdefault(model["task"], []).append(model["model_id"])
for framework in frameworks:
response = get_session().get(f"{INFERENCE_ENDPOINT}/framework/{framework}", headers=self.headers)
hf_raise_for_status(response)
_unpack_response(framework, response.json())
# Sort alphabetically for discoverability and return
for task, models in models_by_task.items():
models_by_task[task] = sorted(set(models), key=lambda x: x.lower())
return models_by_task
def object_detection(
self, image: ContentT, *, model: Optional[str] = None, threshold: Optional[float] = None
) -> List[ObjectDetectionOutputElement]:
"""
Perform object detection on the given image using the specified model.
<Tip warning={true}>
You must have `PIL` installed if you want to work with images (`pip install Pillow`).
</Tip>
Args:
image (`Union[str, Path, bytes, BinaryIO]`):
The image to detect objects on. It can be raw bytes, an image file, or a URL to an online image.
model (`str`, *optional*):
The model to use for object detection. Can be a model ID hosted on the Hugging Face Hub or a URL to a
deployed Inference Endpoint. If not provided, the default recommended model for object detection (DETR) will be used.
threshold (`float`, *optional*):
The probability necessary to make a prediction.
Returns:
`List[ObjectDetectionOutputElement]`: A list of [`ObjectDetectionOutputElement`] items containing the bounding boxes and associated attributes.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
`ValueError`:
If the request output is not a List.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.object_detection("people.jpg")
[ObjectDetectionOutputElement(score=0.9486683011054993, label='person', box=ObjectDetectionBoundingBox(xmin=59, ymin=39, xmax=420, ymax=510)), ...]
```
"""
parameters = {
"threshold": threshold,
}
payload = _prepare_payload(image, parameters=parameters, expect_binary=True)
response = self.post(**payload, model=model, task="object-detection")
return ObjectDetectionOutputElement.parse_obj_as_list(response)
def question_answering(
self,
question: str,
context: str,
*,
model: Optional[str] = None,
align_to_words: Optional[bool] = None,
doc_stride: Optional[int] = None,
handle_impossible_answer: Optional[bool] = None,
max_answer_len: Optional[int] = None,
max_question_len: Optional[int] = None,
max_seq_len: Optional[int] = None,
top_k: Optional[int] = None,
) -> Union[QuestionAnsweringOutputElement, List[QuestionAnsweringOutputElement]]:
"""
Retrieve the answer to a question from a given text.
Args:
question (`str`):
Question to be answered.
context (`str`):
The context of the question.
model (`str`):
The model to use for the question answering task. Can be a model ID hosted on the Hugging Face Hub or a URL to
a deployed Inference Endpoint.
align_to_words (`bool`, *optional*):
Attempts to align the answer to real words. Improves quality on space separated
languages. Might hurt on non-space-separated languages (like Japanese or Chinese).
doc_stride (`int`, *optional*):
If the context is too long to fit with the question for the model, it will be split in
several chunks with some overlap. This argument controls the size of that overlap.
handle_impossible_answer (`bool`, *optional*):
Whether to accept impossible as an answer.
max_answer_len (`int`, *optional*):
The maximum length of predicted answers (e.g., only answers with a shorter length are
considered).
max_question_len (`int`, *optional*):
The maximum length of the question after tokenization. It will be truncated if needed.
max_seq_len (`int`, *optional*):
The maximum length of the total sentence (context + question) in tokens of each chunk
passed to the model. The context will be split in several chunks (using docStride as
overlap) if needed.
top_k (`int`, *optional*):
The number of answers to return (will be chosen by order of likelihood). Note that we
return less than topk answers if there are not enough options available within the
context.
Returns:
Union[`QuestionAnsweringOutputElement`, List[`QuestionAnsweringOutputElement`]]:
When top_k is 1 or not provided, it returns a single `QuestionAnsweringOutputElement`.
When top_k is greater than 1, it returns a list of `QuestionAnsweringOutputElement`.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.question_answering(question="What's my name?", context="My name is Clara and I live in Berkeley.")
QuestionAnsweringOutputElement(answer='Clara', end=16, score=0.9326565265655518, start=11)
```
"""
parameters = {
"align_to_words": align_to_words,
"doc_stride": doc_stride,
"handle_impossible_answer": handle_impossible_answer,
"max_answer_len": max_answer_len,
"max_question_len": max_question_len,
"max_seq_len": max_seq_len,
"top_k": top_k,
}
inputs: Dict[str, Any] = {"question": question, "context": context}
payload = _prepare_payload(inputs, parameters=parameters)
response = self.post(
**payload,
model=model,
task="question-answering",
)
# Parse the response as a single `QuestionAnsweringOutputElement` when top_k is 1 or not provided, or a list of `QuestionAnsweringOutputElement` to ensure backward compatibility.
output = QuestionAnsweringOutputElement.parse_obj(response)
return output
def sentence_similarity(
self, sentence: str, other_sentences: List[str], *, model: Optional[str] = None
) -> List[float]:
"""
Compute the semantic similarity between a sentence and a list of other sentences by comparing their embeddings.
Args:
sentence (`str`):
The main sentence to compare to others.
other_sentences (`List[str]`):
The list of sentences to compare to.
model (`str`, *optional*):
The model to use for the conversational task. Can be a model ID hosted on the Hugging Face Hub or a URL to
a deployed Inference Endpoint. If not provided, the default recommended conversational model will be used.
Defaults to None.
Returns:
`List[float]`: The embedding representing the input text.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.sentence_similarity(
... "Machine learning is so easy.",
... other_sentences=[
... "Deep learning is so straightforward.",
... "This is so difficult, like rocket science.",
... "I can't believe how much I struggled with this.",
... ],
... )
[0.7785726189613342, 0.45876261591911316, 0.2906220555305481]
```
"""
response = self.post(
json={"inputs": {"source_sentence": sentence, "sentences": other_sentences}},
model=model,
task="sentence-similarity",
)
return _bytes_to_list(response)
@_deprecate_arguments(
version="0.29",
deprecated_args=["parameters"],
custom_message=(
"The `parameters` argument is deprecated and will be removed in a future version. "
"Provide individual parameters instead: `clean_up_tokenization_spaces`, `generate_parameters`, and `truncation`."
),
)
def summarization(
self,
text: str,
*,
parameters: Optional[Dict[str, Any]] = None,
model: Optional[str] = None,
clean_up_tokenization_spaces: Optional[bool] = None,
generate_parameters: Optional[Dict[str, Any]] = None,
truncation: Optional[Literal["do_not_truncate", "longest_first", "only_first", "only_second"]] = None,
) -> SummarizationOutput:
"""
Generate a summary of a given text using a specified model.
Args:
text (`str`):
The input text to summarize.
parameters (`Dict[str, Any]`, *optional*):
Additional parameters for summarization. Check out this [page](https://huggingface.co/docs/api-inference/detailed_parameters#summarization-task)
for more details.
model (`str`, *optional*):
The model to use for inference. Can be a model ID hosted on the Hugging Face Hub or a URL to a deployed
Inference Endpoint. If not provided, the default recommended model for summarization will be used.
clean_up_tokenization_spaces (`bool`, *optional*):
Whether to clean up the potential extra spaces in the text output.
generate_parameters (`Dict[str, Any]`, *optional*):
Additional parametrization of the text generation algorithm.
truncation (`Literal["do_not_truncate", "longest_first", "only_first", "only_second"]`, *optional*):
The truncation strategy to use.
Returns:
[`SummarizationOutput`]: The generated summary text.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.summarization("The Eiffel tower...")
SummarizationOutput(generated_text="The Eiffel tower is one of the most famous landmarks in the world....")
```
"""
if parameters is None:
parameters = {
"clean_up_tokenization_spaces": clean_up_tokenization_spaces,
"generate_parameters": generate_parameters,
"truncation": truncation,
}
payload = _prepare_payload(text, parameters=parameters)
response = self.post(**payload, model=model, task="summarization")
return SummarizationOutput.parse_obj_as_list(response)[0]
def table_question_answering(
self,
table: Dict[str, Any],
query: str,
*,
model: Optional[str] = None,
parameters: Optional[Dict[str, Any]] = None,
) -> TableQuestionAnsweringOutputElement:
"""
Retrieve the answer to a question from information given in a table.
Args:
table (`str`):
A table of data represented as a dict of lists where entries are headers and the lists are all the
values, all lists must have the same size.
query (`str`):
The query in plain text that you want to ask the table.
model (`str`):
The model to use for the table-question-answering task. Can be a model ID hosted on the Hugging Face
Hub or a URL to a deployed Inference Endpoint.
parameters (`Dict[str, Any]`, *optional*):
Additional inference parameters. Defaults to None.
Returns:
[`TableQuestionAnsweringOutputElement`]: a table question answering output containing the answer, coordinates, cells and the aggregator used.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> query = "How many stars does the transformers repository have?"
>>> table = {"Repository": ["Transformers", "Datasets", "Tokenizers"], "Stars": ["36542", "4512", "3934"]}
>>> client.table_question_answering(table, query, model="google/tapas-base-finetuned-wtq")
TableQuestionAnsweringOutputElement(answer='36542', coordinates=[[0, 1]], cells=['36542'], aggregator='AVERAGE')
```
"""
inputs = {
"query": query,
"table": table,
}
payload = _prepare_payload(inputs, parameters=parameters)
response = self.post(
**payload,
model=model,
task="table-question-answering",
)
return TableQuestionAnsweringOutputElement.parse_obj_as_instance(response)
def tabular_classification(self, table: Dict[str, Any], *, model: Optional[str] = None) -> List[str]:
"""
Classifying a target category (a group) based on a set of attributes.
Args:
table (`Dict[str, Any]`):
Set of attributes to classify.
model (`str`, *optional*):
The model to use for the tabular classification task. Can be a model ID hosted on the Hugging Face Hub or a URL to
a deployed Inference Endpoint. If not provided, the default recommended tabular classification model will be used.
Defaults to None.
Returns:
`List`: a list of labels, one per row in the initial table.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> table = {
... "fixed_acidity": ["7.4", "7.8", "10.3"],
... "volatile_acidity": ["0.7", "0.88", "0.32"],
... "citric_acid": ["0", "0", "0.45"],
... "residual_sugar": ["1.9", "2.6", "6.4"],
... "chlorides": ["0.076", "0.098", "0.073"],
... "free_sulfur_dioxide": ["11", "25", "5"],
... "total_sulfur_dioxide": ["34", "67", "13"],
... "density": ["0.9978", "0.9968", "0.9976"],
... "pH": ["3.51", "3.2", "3.23"],
... "sulphates": ["0.56", "0.68", "0.82"],
... "alcohol": ["9.4", "9.8", "12.6"],
... }
>>> client.tabular_classification(table=table, model="julien-c/wine-quality")
["5", "5", "5"]
```
"""
response = self.post(
json={"table": table},
model=model,
task="tabular-classification",
)
return _bytes_to_list(response)
def tabular_regression(self, table: Dict[str, Any], *, model: Optional[str] = None) -> List[float]:
"""
Predicting a numerical target value given a set of attributes/features in a table.
Args:
table (`Dict[str, Any]`):
Set of attributes stored in a table. The attributes used to predict the target can be both numerical and categorical.
model (`str`, *optional*):
The model to use for the tabular regression task. Can be a model ID hosted on the Hugging Face Hub or a URL to
a deployed Inference Endpoint. If not provided, the default recommended tabular regression model will be used.
Defaults to None.
Returns:
`List`: a list of predicted numerical target values.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> table = {
... "Height": ["11.52", "12.48", "12.3778"],
... "Length1": ["23.2", "24", "23.9"],
... "Length2": ["25.4", "26.3", "26.5"],
... "Length3": ["30", "31.2", "31.1"],
... "Species": ["Bream", "Bream", "Bream"],
... "Width": ["4.02", "4.3056", "4.6961"],
... }
>>> client.tabular_regression(table, model="scikit-learn/Fish-Weight")
[110, 120, 130]
```
"""
response = self.post(json={"table": table}, model=model, task="tabular-regression")
return _bytes_to_list(response)
def text_classification(
self,
text: str,
*,
model: Optional[str] = None,
top_k: Optional[int] = None,
function_to_apply: Optional["TextClassificationOutputTransform"] = None,
) -> List[TextClassificationOutputElement]:
"""
Perform text classification (e.g. sentiment-analysis) on the given text.
Args:
text (`str`):
A string to be classified.
model (`str`, *optional*):
The model to use for the text classification task. Can be a model ID hosted on the Hugging Face Hub or a URL to
a deployed Inference Endpoint. If not provided, the default recommended text classification model will be used.
Defaults to None.
top_k (`int`, *optional*):
When specified, limits the output to the top K most probable classes.
function_to_apply (`"TextClassificationOutputTransform"`, *optional*):
The function to apply to the output.
Returns:
`List[TextClassificationOutputElement]`: a list of [`TextClassificationOutputElement`] items containing the predicted label and associated probability.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.text_classification("I like you")
[
TextClassificationOutputElement(label='POSITIVE', score=0.9998695850372314),
TextClassificationOutputElement(label='NEGATIVE', score=0.0001304351753788069),
]
```
"""
parameters = {
"function_to_apply": function_to_apply,
"top_k": top_k,
}
payload = _prepare_payload(text, parameters=parameters)
response = self.post(
**payload,
model=model,
task="text-classification",
)
return TextClassificationOutputElement.parse_obj_as_list(response)[0] # type: ignore [return-value]
@overload
def text_generation( # type: ignore
self,
prompt: str,
*,
details: Literal[False] = ...,
stream: Literal[False] = ...,
model: Optional[str] = None,
# Parameters from `TextGenerationInputGenerateParameters` (maintained manually)
adapter_id: Optional[str] = None,
best_of: Optional[int] = None,
decoder_input_details: Optional[bool] = None,
do_sample: Optional[bool] = False, # Manual default value
frequency_penalty: Optional[float] = None,
grammar: Optional[TextGenerationInputGrammarType] = None,
max_new_tokens: Optional[int] = None,
repetition_penalty: Optional[float] = None,
return_full_text: Optional[bool] = False, # Manual default value
seed: Optional[int] = None,
stop: Optional[List[str]] = None,
stop_sequences: Optional[List[str]] = None, # Deprecated, use `stop` instead
temperature: Optional[float] = None,
top_k: Optional[int] = None,
top_n_tokens: Optional[int] = None,
top_p: Optional[float] = None,
truncate: Optional[int] = None,
typical_p: Optional[float] = None,
watermark: Optional[bool] = None,
) -> str: ...
@overload
def text_generation( # type: ignore
self,
prompt: str,
*,
details: Literal[True] = ...,
stream: Literal[False] = ...,
model: Optional[str] = None,
# Parameters from `TextGenerationInputGenerateParameters` (maintained manually)
adapter_id: Optional[str] = None,
best_of: Optional[int] = None,
decoder_input_details: Optional[bool] = None,
do_sample: Optional[bool] = False, # Manual default value
frequency_penalty: Optional[float] = None,
grammar: Optional[TextGenerationInputGrammarType] = None,
max_new_tokens: Optional[int] = None,
repetition_penalty: Optional[float] = None,
return_full_text: Optional[bool] = False, # Manual default value
seed: Optional[int] = None,
stop: Optional[List[str]] = None,
stop_sequences: Optional[List[str]] = None, # Deprecated, use `stop` instead
temperature: Optional[float] = None,
top_k: Optional[int] = None,
top_n_tokens: Optional[int] = None,
top_p: Optional[float] = None,
truncate: Optional[int] = None,
typical_p: Optional[float] = None,
watermark: Optional[bool] = None,
) -> TextGenerationOutput: ...
@overload
def text_generation( # type: ignore
self,
prompt: str,
*,
details: Literal[False] = ...,
stream: Literal[True] = ...,
model: Optional[str] = None,
# Parameters from `TextGenerationInputGenerateParameters` (maintained manually)
adapter_id: Optional[str] = None,
best_of: Optional[int] = None,
decoder_input_details: Optional[bool] = None,
do_sample: Optional[bool] = False, # Manual default value
frequency_penalty: Optional[float] = None,
grammar: Optional[TextGenerationInputGrammarType] = None,
max_new_tokens: Optional[int] = None,
repetition_penalty: Optional[float] = None,
return_full_text: Optional[bool] = False, # Manual default value
seed: Optional[int] = None,
stop: Optional[List[str]] = None,
stop_sequences: Optional[List[str]] = None, # Deprecated, use `stop` instead
temperature: Optional[float] = None,
top_k: Optional[int] = None,
top_n_tokens: Optional[int] = None,
top_p: Optional[float] = None,
truncate: Optional[int] = None,
typical_p: Optional[float] = None,
watermark: Optional[bool] = None,
) -> Iterable[str]: ...
@overload
def text_generation( # type: ignore
self,
prompt: str,
*,
details: Literal[True] = ...,
stream: Literal[True] = ...,
model: Optional[str] = None,
# Parameters from `TextGenerationInputGenerateParameters` (maintained manually)
adapter_id: Optional[str] = None,
best_of: Optional[int] = None,
decoder_input_details: Optional[bool] = None,
do_sample: Optional[bool] = False, # Manual default value
frequency_penalty: Optional[float] = None,
grammar: Optional[TextGenerationInputGrammarType] = None,
max_new_tokens: Optional[int] = None,
repetition_penalty: Optional[float] = None,
return_full_text: Optional[bool] = False, # Manual default value
seed: Optional[int] = None,
stop: Optional[List[str]] = None,
stop_sequences: Optional[List[str]] = None, # Deprecated, use `stop` instead
temperature: Optional[float] = None,
top_k: Optional[int] = None,
top_n_tokens: Optional[int] = None,
top_p: Optional[float] = None,
truncate: Optional[int] = None,
typical_p: Optional[float] = None,
watermark: Optional[bool] = None,
) -> Iterable[TextGenerationStreamOutput]: ...
@overload
def text_generation(
self,
prompt: str,
*,
details: Literal[True] = ...,
stream: bool = ...,
model: Optional[str] = None,
# Parameters from `TextGenerationInputGenerateParameters` (maintained manually)
adapter_id: Optional[str] = None,
best_of: Optional[int] = None,
decoder_input_details: Optional[bool] = None,
do_sample: Optional[bool] = False, # Manual default value
frequency_penalty: Optional[float] = None,
grammar: Optional[TextGenerationInputGrammarType] = None,
max_new_tokens: Optional[int] = None,
repetition_penalty: Optional[float] = None,
return_full_text: Optional[bool] = False, # Manual default value
seed: Optional[int] = None,
stop: Optional[List[str]] = None,
stop_sequences: Optional[List[str]] = None, # Deprecated, use `stop` instead
temperature: Optional[float] = None,
top_k: Optional[int] = None,
top_n_tokens: Optional[int] = None,
top_p: Optional[float] = None,
truncate: Optional[int] = None,
typical_p: Optional[float] = None,
watermark: Optional[bool] = None,
) -> Union[TextGenerationOutput, Iterable[TextGenerationStreamOutput]]: ...
def text_generation(
self,
prompt: str,
*,
details: bool = False,
stream: bool = False,
model: Optional[str] = None,
# Parameters from `TextGenerationInputGenerateParameters` (maintained manually)
adapter_id: Optional[str] = None,
best_of: Optional[int] = None,
decoder_input_details: Optional[bool] = None,
do_sample: Optional[bool] = False, # Manual default value
frequency_penalty: Optional[float] = None,
grammar: Optional[TextGenerationInputGrammarType] = None,
max_new_tokens: Optional[int] = None,
repetition_penalty: Optional[float] = None,
return_full_text: Optional[bool] = False, # Manual default value
seed: Optional[int] = None,
stop: Optional[List[str]] = None,
stop_sequences: Optional[List[str]] = None, # Deprecated, use `stop` instead
temperature: Optional[float] = None,
top_k: Optional[int] = None,
top_n_tokens: Optional[int] = None,
top_p: Optional[float] = None,
truncate: Optional[int] = None,
typical_p: Optional[float] = None,
watermark: Optional[bool] = None,
) -> Union[str, TextGenerationOutput, Iterable[str], Iterable[TextGenerationStreamOutput]]:
"""
Given a prompt, generate the following text.
API endpoint is supposed to run with the `text-generation-inference` backend (TGI). This backend is the
go-to solution to run large language models at scale. However, for some smaller models (e.g. "gpt2") the
default `transformers` + `api-inference` solution is still in use. Both approaches have very similar APIs, but
not exactly the same. This method is compatible with both approaches but some parameters are only available for
`text-generation-inference`. If some parameters are ignored, a warning message is triggered but the process
continues correctly.
To learn more about the TGI project, please refer to https://github.com/huggingface/text-generation-inference.
<Tip>
If you want to generate a response from chat messages, you should use the [`InferenceClient.chat_completion`] method.
It accepts a list of messages instead of a single text prompt and handles the chat templating for you.
</Tip>
Args:
prompt (`str`):
Input text.
details (`bool`, *optional*):
By default, text_generation returns a string. Pass `details=True` if you want a detailed output (tokens,
probabilities, seed, finish reason, etc.). Only available for models running on with the
`text-generation-inference` backend.
stream (`bool`, *optional*):
By default, text_generation returns the full generated text. Pass `stream=True` if you want a stream of
tokens to be returned. Only available for models running on with the `text-generation-inference`
backend.
model (`str`, *optional*):
The model to use for inference. Can be a model ID hosted on the Hugging Face Hub or a URL to a deployed
Inference Endpoint. This parameter overrides the model defined at the instance level. Defaults to None.
adapter_id (`str`, *optional*):
Lora adapter id.
best_of (`int`, *optional*):
Generate best_of sequences and return the one if the highest token logprobs.
decoder_input_details (`bool`, *optional*):
Return the decoder input token logprobs and ids. You must set `details=True` as well for it to be taken
into account. Defaults to `False`.
do_sample (`bool`, *optional*):
Activate logits sampling
frequency_penalty (`float`, *optional*):
Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in
the text so far, decreasing the model's likelihood to repeat the same line verbatim.
grammar ([`TextGenerationInputGrammarType`], *optional*):
Grammar constraints. Can be either a JSONSchema or a regex.
max_new_tokens (`int`, *optional*):
Maximum number of generated tokens
repetition_penalty (`float`, *optional*):
The parameter for repetition penalty. 1.0 means no penalty. See [this
paper](https://arxiv.org/pdf/1909.05858.pdf) for more details.
return_full_text (`bool`, *optional*):
Whether to prepend the prompt to the generated text
seed (`int`, *optional*):
Random sampling seed
stop (`List[str]`, *optional*):
Stop generating tokens if a member of `stop` is generated.
stop_sequences (`List[str]`, *optional*):
Deprecated argument. Use `stop` instead.
temperature (`float`, *optional*):
The value used to module the logits distribution.
top_n_tokens (`int`, *optional*):
Return information about the `top_n_tokens` most likely tokens at each generation step, instead of
just the sampled token.
top_k (`int`, *optional`):
The number of highest probability vocabulary tokens to keep for top-k-filtering.
top_p (`float`, *optional`):
If set to < 1, only the smallest set of most probable tokens with probabilities that add up to `top_p` or
higher are kept for generation.
truncate (`int`, *optional`):
Truncate inputs tokens to the given size.
typical_p (`float`, *optional`):
Typical Decoding mass
See [Typical Decoding for Natural Language Generation](https://arxiv.org/abs/2202.00666) for more information
watermark (`bool`, *optional`):
Watermarking with [A Watermark for Large Language Models](https://arxiv.org/abs/2301.10226)
Returns:
`Union[str, TextGenerationOutput, Iterable[str], Iterable[TextGenerationStreamOutput]]`:
Generated text returned from the server:
- if `stream=False` and `details=False`, the generated text is returned as a `str` (default)
- if `stream=True` and `details=False`, the generated text is returned token by token as a `Iterable[str]`
- if `stream=False` and `details=True`, the generated text is returned with more details as a [`~huggingface_hub.TextGenerationOutput`]
- if `details=True` and `stream=True`, the generated text is returned token by token as a iterable of [`~huggingface_hub.TextGenerationStreamOutput`]
Raises:
`ValidationError`:
If input values are not valid. No HTTP call is made to the server.
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
# Case 1: generate text
>>> client.text_generation("The huggingface_hub library is ", max_new_tokens=12)
'100% open source and built to be easy to use.'
# Case 2: iterate over the generated tokens. Useful for large generation.
>>> for token in client.text_generation("The huggingface_hub library is ", max_new_tokens=12, stream=True):
... print(token)
100
%
open
source
and
built
to
be
easy
to
use
.
# Case 3: get more details about the generation process.
>>> client.text_generation("The huggingface_hub library is ", max_new_tokens=12, details=True)
TextGenerationOutput(
generated_text='100% open source and built to be easy to use.',
details=TextGenerationDetails(
finish_reason='length',
generated_tokens=12,
seed=None,
prefill=[
TextGenerationPrefillOutputToken(id=487, text='The', logprob=None),
TextGenerationPrefillOutputToken(id=53789, text=' hugging', logprob=-13.171875),
(...)
TextGenerationPrefillOutputToken(id=204, text=' ', logprob=-7.0390625)
],
tokens=[
TokenElement(id=1425, text='100', logprob=-1.0175781, special=False),
TokenElement(id=16, text='%', logprob=-0.0463562, special=False),
(...)
TokenElement(id=25, text='.', logprob=-0.5703125, special=False)
],
best_of_sequences=None
)
)
# Case 4: iterate over the generated tokens with more details.
# Last object is more complete, containing the full generated text and the finish reason.
>>> for details in client.text_generation("The huggingface_hub library is ", max_new_tokens=12, details=True, stream=True):
... print(details)
...
TextGenerationStreamOutput(token=TokenElement(id=1425, text='100', logprob=-1.0175781, special=False), generated_text=None, details=None)
TextGenerationStreamOutput(token=TokenElement(id=16, text='%', logprob=-0.0463562, special=False), generated_text=None, details=None)
TextGenerationStreamOutput(token=TokenElement(id=1314, text=' open', logprob=-1.3359375, special=False), generated_text=None, details=None)
TextGenerationStreamOutput(token=TokenElement(id=3178, text=' source', logprob=-0.28100586, special=False), generated_text=None, details=None)
TextGenerationStreamOutput(token=TokenElement(id=273, text=' and', logprob=-0.5961914, special=False), generated_text=None, details=None)
TextGenerationStreamOutput(token=TokenElement(id=3426, text=' built', logprob=-1.9423828, special=False), generated_text=None, details=None)
TextGenerationStreamOutput(token=TokenElement(id=271, text=' to', logprob=-1.4121094, special=False), generated_text=None, details=None)
TextGenerationStreamOutput(token=TokenElement(id=314, text=' be', logprob=-1.5224609, special=False), generated_text=None, details=None)
TextGenerationStreamOutput(token=TokenElement(id=1833, text=' easy', logprob=-2.1132812, special=False), generated_text=None, details=None)
TextGenerationStreamOutput(token=TokenElement(id=271, text=' to', logprob=-0.08520508, special=False), generated_text=None, details=None)
TextGenerationStreamOutput(token=TokenElement(id=745, text=' use', logprob=-0.39453125, special=False), generated_text=None, details=None)
TextGenerationStreamOutput(token=TokenElement(
id=25,
text='.',
logprob=-0.5703125,
special=False),
generated_text='100% open source and built to be easy to use.',
details=TextGenerationStreamOutputStreamDetails(finish_reason='length', generated_tokens=12, seed=None)
)
# Case 5: generate constrained output using grammar
>>> response = client.text_generation(
... prompt="I saw a puppy a cat and a raccoon during my bike ride in the park",
... model="HuggingFaceH4/zephyr-orpo-141b-A35b-v0.1",
... max_new_tokens=100,
... repetition_penalty=1.3,
... grammar={
... "type": "json",
... "value": {
... "properties": {
... "location": {"type": "string"},
... "activity": {"type": "string"},
... "animals_seen": {"type": "integer", "minimum": 1, "maximum": 5},
... "animals": {"type": "array", "items": {"type": "string"}},
... },
... "required": ["location", "activity", "animals_seen", "animals"],
... },
... },
... )
>>> json.loads(response)
{
"activity": "bike riding",
"animals": ["puppy", "cat", "raccoon"],
"animals_seen": 3,
"location": "park"
}
```
"""
if decoder_input_details and not details:
warnings.warn(
"`decoder_input_details=True` has been passed to the server but `details=False` is set meaning that"
" the output from the server will be truncated."
)
decoder_input_details = False
if stop_sequences is not None:
warnings.warn(
"`stop_sequences` is a deprecated argument for `text_generation` task"
" and will be removed in version '0.28.0'. Use `stop` instead.",
FutureWarning,
)
if stop is None:
stop = stop_sequences # use deprecated arg if provided
# Build payload
parameters = {
"adapter_id": adapter_id,
"best_of": best_of,
"decoder_input_details": decoder_input_details,
"details": details,
"do_sample": do_sample,
"frequency_penalty": frequency_penalty,
"grammar": grammar,
"max_new_tokens": max_new_tokens,
"repetition_penalty": repetition_penalty,
"return_full_text": return_full_text,
"seed": seed,
"stop": stop if stop is not None else [],
"temperature": temperature,
"top_k": top_k,
"top_n_tokens": top_n_tokens,
"top_p": top_p,
"truncate": truncate,
"typical_p": typical_p,
"watermark": watermark,
}
parameters = {k: v for k, v in parameters.items() if v is not None}
payload = {
"inputs": prompt,
"parameters": parameters,
"stream": stream,
}
# Remove some parameters if not a TGI server
unsupported_kwargs = _get_unsupported_text_generation_kwargs(model)
if len(unsupported_kwargs) > 0:
# The server does not support some parameters
# => means it is not a TGI server
# => remove unsupported parameters and warn the user
ignored_parameters = []
for key in unsupported_kwargs:
if parameters.get(key):
ignored_parameters.append(key)
parameters.pop(key, None)
if len(ignored_parameters) > 0:
warnings.warn(
"API endpoint/model for text-generation is not served via TGI. Ignoring following parameters:"
f" {', '.join(ignored_parameters)}.",
UserWarning,
)
if details:
warnings.warn(
"API endpoint/model for text-generation is not served via TGI. Parameter `details=True` will"
" be ignored meaning only the generated text will be returned.",
UserWarning,
)
details = False
if stream:
raise ValueError(
"API endpoint/model for text-generation is not served via TGI. Cannot return output as a stream."
" Please pass `stream=False` as input."
)
# Handle errors separately for more precise error messages
try:
bytes_output = self.post(json=payload, model=model, task="text-generation", stream=stream) # type: ignore
except HTTPError as e:
match = MODEL_KWARGS_NOT_USED_REGEX.search(str(e))
if isinstance(e, BadRequestError) and match:
unused_params = [kwarg.strip("' ") for kwarg in match.group(1).split(",")]
_set_unsupported_text_generation_kwargs(model, unused_params)
return self.text_generation( # type: ignore
prompt=prompt,
details=details,
stream=stream,
model=model,
adapter_id=adapter_id,
best_of=best_of,
decoder_input_details=decoder_input_details,
do_sample=do_sample,
frequency_penalty=frequency_penalty,
grammar=grammar,
max_new_tokens=max_new_tokens,
repetition_penalty=repetition_penalty,
return_full_text=return_full_text,
seed=seed,
stop=stop,
temperature=temperature,
top_k=top_k,
top_n_tokens=top_n_tokens,
top_p=top_p,
truncate=truncate,
typical_p=typical_p,
watermark=watermark,
)
raise_text_generation_error(e)
# Parse output
if stream:
return _stream_text_generation_response(bytes_output, details) # type: ignore
data = _bytes_to_dict(bytes_output) # type: ignore[arg-type]
# Data can be a single element (dict) or an iterable of dicts where we select the first element of.
if isinstance(data, list):
data = data[0]
return TextGenerationOutput.parse_obj_as_instance(data) if details else data["generated_text"]
def text_to_image(
self,
prompt: str,
*,
negative_prompt: Optional[str] = None,
height: Optional[float] = None,
width: Optional[float] = None,
num_inference_steps: Optional[float] = None,
guidance_scale: Optional[float] = None,
model: Optional[str] = None,
scheduler: Optional[str] = None,
target_size: Optional[TextToImageTargetSize] = None,
seed: Optional[int] = None,
**kwargs,
) -> "Image":
"""
Generate an image based on a given text using a specified model.
<Tip warning={true}>
You must have `PIL` installed if you want to work with images (`pip install Pillow`).
</Tip>
Args:
prompt (`str`):
The prompt to generate an image from.
negative_prompt (`str`, *optional*):
An optional negative prompt for the image generation.
height (`float`, *optional*):
The height in pixels of the image to generate.
width (`float`, *optional*):
The width in pixels of the image to generate.
num_inference_steps (`int`, *optional*):
The number of denoising steps. More denoising steps usually lead to a higher quality image at the
expense of slower inference.
guidance_scale (`float`, *optional*):
Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
usually at the expense of lower image quality.
model (`str`, *optional*):
The model to use for inference. Can be a model ID hosted on the Hugging Face Hub or a URL to a deployed
Inference Endpoint. If not provided, the default recommended text-to-image model will be used.
Defaults to None.
scheduler (`str`, *optional*):
Override the scheduler with a compatible one.
target_size (`TextToImageTargetSize`, *optional*):
The size in pixel of the output image
seed (`int`, *optional*):
Seed for the random number generator.
Returns:
`Image`: The generated image.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> image = client.text_to_image("An astronaut riding a horse on the moon.")
>>> image.save("astronaut.png")
>>> image = client.text_to_image(
... "An astronaut riding a horse on the moon.",
... negative_prompt="low resolution, blurry",
... model="stabilityai/stable-diffusion-2-1",
... )
>>> image.save("better_astronaut.png")
```
"""
parameters = {
"negative_prompt": negative_prompt,
"height": height,
"width": width,
"num_inference_steps": num_inference_steps,
"guidance_scale": guidance_scale,
"scheduler": scheduler,
"target_size": target_size,
"seed": seed,
**kwargs,
}
payload = _prepare_payload(prompt, parameters=parameters)
response = self.post(**payload, model=model, task="text-to-image")
return _bytes_to_image(response)
def text_to_speech(
self,
text: str,
*,
model: Optional[str] = None,
do_sample: Optional[bool] = None,
early_stopping: Optional[Union[bool, "TextToSpeechEarlyStoppingEnum"]] = None,
epsilon_cutoff: Optional[float] = None,
eta_cutoff: Optional[float] = None,
max_length: Optional[int] = None,
max_new_tokens: Optional[int] = None,
min_length: Optional[int] = None,
min_new_tokens: Optional[int] = None,
num_beam_groups: Optional[int] = None,
num_beams: Optional[int] = None,
penalty_alpha: Optional[float] = None,
temperature: Optional[float] = None,
top_k: Optional[int] = None,
top_p: Optional[float] = None,
typical_p: Optional[float] = None,
use_cache: Optional[bool] = None,
) -> bytes:
"""
Synthesize an audio of a voice pronouncing a given text.
Args:
text (`str`):
The text to synthesize.
model (`str`, *optional*):
The model to use for inference. Can be a model ID hosted on the Hugging Face Hub or a URL to a deployed
Inference Endpoint. If not provided, the default recommended text-to-speech model will be used.
Defaults to None.
do_sample (`bool`, *optional*):
Whether to use sampling instead of greedy decoding when generating new tokens.
early_stopping (`Union[bool, "TextToSpeechEarlyStoppingEnum"`, *optional*):
Controls the stopping condition for beam-based methods.
epsilon_cutoff (`float`, *optional*):
If set to float strictly between 0 and 1, only tokens with a conditional probability
greater than epsilon_cutoff will be sampled. In the paper, suggested values range from
3e-4 to 9e-4, depending on the size of the model. See [Truncation Sampling as Language
Model Desmoothing](https://hf.co/papers/2210.15191) for more details.
eta_cutoff (`float`, *optional*):
Eta sampling is a hybrid of locally typical sampling and epsilon sampling. If set to
float strictly between 0 and 1, a token is only considered if it is greater than either
eta_cutoff or sqrt(eta_cutoff) * exp(-entropy(softmax(next_token_logits))). The latter
term is intuitively the expected next token probability, scaled by sqrt(eta_cutoff). In
the paper, suggested values range from 3e-4 to 2e-3, depending on the size of the model.
See [Truncation Sampling as Language Model Desmoothing](https://hf.co/papers/2210.15191)
for more details.
max_length (`int`, *optional*):
The maximum length (in tokens) of the generated text, including the input.
max_new_tokens (`int`, *optional*):
The maximum number of tokens to generate. Takes precedence over maxLength.
min_length (`int`, *optional*):
The minimum length (in tokens) of the generated text, including the input.
min_new_tokens (`int`, *optional*):
The minimum number of tokens to generate. Takes precedence over maxLength.
num_beam_groups (`int`, *optional*):
Number of groups to divide num_beams into in order to ensure diversity among different
groups of beams. See [this paper](https://hf.co/papers/1610.02424) for more details.
num_beams (`int`, *optional*):
Number of beams to use for beam search.
penalty_alpha (`float`, *optional*):
The value balances the model confidence and the degeneration penalty in contrastive
search decoding.
temperature (`float`, *optional*):
The value used to modulate the next token probabilities.
top_k (`int`, *optional*):
The number of highest probability vocabulary tokens to keep for top-k-filtering.
top_p (`float`, *optional*):
If set to float < 1, only the smallest set of most probable tokens with probabilities
that add up to top_p or higher are kept for generation.
typical_p (`float`, *optional*):
Local typicality measures how similar the conditional probability of predicting a target token next is
to the expected conditional probability of predicting a random token next, given the partial text
already generated. If set to float < 1, the smallest set of the most locally typical tokens with
probabilities that add up to typical_p or higher are kept for generation. See [this
paper](https://hf.co/papers/2202.00666) for more details.
use_cache (`bool`, *optional*):
Whether the model should use the past last key/values attentions to speed up decoding
Returns:
`bytes`: The generated audio.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from pathlib import Path
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> audio = client.text_to_speech("Hello world")
>>> Path("hello_world.flac").write_bytes(audio)
```
"""
parameters = {
"do_sample": do_sample,
"early_stopping": early_stopping,
"epsilon_cutoff": epsilon_cutoff,
"eta_cutoff": eta_cutoff,
"max_length": max_length,
"max_new_tokens": max_new_tokens,
"min_length": min_length,
"min_new_tokens": min_new_tokens,
"num_beam_groups": num_beam_groups,
"num_beams": num_beams,
"penalty_alpha": penalty_alpha,
"temperature": temperature,
"top_k": top_k,
"top_p": top_p,
"typical_p": typical_p,
"use_cache": use_cache,
}
payload = _prepare_payload(text, parameters=parameters)
response = self.post(**payload, model=model, task="text-to-speech")
return response
def token_classification(
self,
text: str,
*,
model: Optional[str] = None,
aggregation_strategy: Optional[Literal["none", "simple", "first", "average", "max"]] = None,
ignore_labels: Optional[List[str]] = None,
stride: Optional[int] = None,
) -> List[TokenClassificationOutputElement]:
"""
Perform token classification on the given text.
Usually used for sentence parsing, either grammatical, or Named Entity Recognition (NER) to understand keywords contained within text.
Args:
text (`str`):
A string to be classified.
model (`str`, *optional*):
The model to use for the token classification task. Can be a model ID hosted on the Hugging Face Hub or a URL to
a deployed Inference Endpoint. If not provided, the default recommended token classification model will be used.
Defaults to None.
aggregation_strategy (`Literal["none", "simple", "first", "average", "max"]`, *optional*):
The strategy used to fuse tokens based on model predictions.
ignore_labels (`List[str]`, *optional*):
A list of labels to ignore.
stride (`int`, *optional*):
The number of overlapping tokens between chunks when splitting the input text.
Returns:
`List[TokenClassificationOutputElement]`: List of [`TokenClassificationOutputElement`] items containing the entity group, confidence score, word, start and end index.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.token_classification("My name is Sarah Jessica Parker but you can call me Jessica")
[
TokenClassificationOutputElement(
entity_group='PER',
score=0.9971321225166321,
word='Sarah Jessica Parker',
start=11,
end=31,
),
TokenClassificationOutputElement(
entity_group='PER',
score=0.9773476123809814,
word='Jessica',
start=52,
end=59,
)
]
```
"""
parameters = {
"aggregation_strategy": aggregation_strategy,
"ignore_labels": ignore_labels,
"stride": stride,
}
payload = _prepare_payload(text, parameters=parameters)
response = self.post(
**payload,
model=model,
task="token-classification",
)
return TokenClassificationOutputElement.parse_obj_as_list(response)
def translation(
self,
text: str,
*,
model: Optional[str] = None,
src_lang: Optional[str] = None,
tgt_lang: Optional[str] = None,
clean_up_tokenization_spaces: Optional[bool] = None,
truncation: Optional[Literal["do_not_truncate", "longest_first", "only_first", "only_second"]] = None,
generate_parameters: Optional[Dict[str, Any]] = None,
) -> TranslationOutput:
"""
Convert text from one language to another.
Check out https://huggingface.co/tasks/translation for more information on how to choose the best model for
your specific use case. Source and target languages usually depend on the model.
However, it is possible to specify source and target languages for certain models. If you are working with one of these models,
you can use `src_lang` and `tgt_lang` arguments to pass the relevant information.
Args:
text (`str`):
A string to be translated.
model (`str`, *optional*):
The model to use for the translation task. Can be a model ID hosted on the Hugging Face Hub or a URL to
a deployed Inference Endpoint. If not provided, the default recommended translation model will be used.
Defaults to None.
src_lang (`str`, *optional*):
The source language of the text. Required for models that can translate from multiple languages.
tgt_lang (`str`, *optional*):
Target language to translate to. Required for models that can translate to multiple languages.
clean_up_tokenization_spaces (`bool`, *optional*):
Whether to clean up the potential extra spaces in the text output.
truncation (`Literal["do_not_truncate", "longest_first", "only_first", "only_second"]`, *optional*):
The truncation strategy to use.
generate_parameters (`Dict[str, Any]`, *optional*):
Additional parametrization of the text generation algorithm.
Returns:
[`TranslationOutput`]: The generated translated text.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
`ValueError`:
If only one of the `src_lang` and `tgt_lang` arguments are provided.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.translation("My name is Wolfgang and I live in Berlin")
'Mein Name ist Wolfgang und ich lebe in Berlin.'
>>> client.translation("My name is Wolfgang and I live in Berlin", model="Helsinki-NLP/opus-mt-en-fr")
TranslationOutput(translation_text='Je m\'appelle Wolfgang et je vis à Berlin.')
```
Specifying languages:
```py
>>> client.translation("My name is Sarah Jessica Parker but you can call me Jessica", model="facebook/mbart-large-50-many-to-many-mmt", src_lang="en_XX", tgt_lang="fr_XX")
"Mon nom est Sarah Jessica Parker mais vous pouvez m\'appeler Jessica"
```
"""
# Throw error if only one of `src_lang` and `tgt_lang` was given
if src_lang is not None and tgt_lang is None:
raise ValueError("You cannot specify `src_lang` without specifying `tgt_lang`.")
if src_lang is None and tgt_lang is not None:
raise ValueError("You cannot specify `tgt_lang` without specifying `src_lang`.")
parameters = {
"src_lang": src_lang,
"tgt_lang": tgt_lang,
"clean_up_tokenization_spaces": clean_up_tokenization_spaces,
"truncation": truncation,
"generate_parameters": generate_parameters,
}
payload = _prepare_payload(text, parameters=parameters)
response = self.post(**payload, model=model, task="translation")
return TranslationOutput.parse_obj_as_list(response)[0]
def visual_question_answering(
self,
image: ContentT,
question: str,
*,
model: Optional[str] = None,
top_k: Optional[int] = None,
) -> List[VisualQuestionAnsweringOutputElement]:
"""
Answering open-ended questions based on an image.
Args:
image (`Union[str, Path, bytes, BinaryIO]`):
The input image for the context. It can be raw bytes, an image file, or a URL to an online image.
question (`str`):
Question to be answered.
model (`str`, *optional*):
The model to use for the visual question answering task. Can be a model ID hosted on the Hugging Face Hub or a URL to
a deployed Inference Endpoint. If not provided, the default recommended visual question answering model will be used.
Defaults to None.
top_k (`int`, *optional*):
The number of answers to return (will be chosen by order of likelihood). Note that we
return less than topk answers if there are not enough options available within the
context.
Returns:
`List[VisualQuestionAnsweringOutputElement]`: a list of [`VisualQuestionAnsweringOutputElement`] items containing the predicted label and associated probability.
Raises:
`InferenceTimeoutError`:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.visual_question_answering(
... image="https://huggingface.co/datasets/mishig/sample_images/resolve/main/tiger.jpg",
... question="What is the animal doing?"
... )
[
VisualQuestionAnsweringOutputElement(score=0.778609573841095, answer='laying down'),
VisualQuestionAnsweringOutputElement(score=0.6957435607910156, answer='sitting'),
]
```
"""
payload: Dict[str, Any] = {"question": question, "image": _b64_encode(image)}
if top_k is not None:
payload.setdefault("parameters", {})["top_k"] = top_k
response = self.post(json=payload, model=model, task="visual-question-answering")
return VisualQuestionAnsweringOutputElement.parse_obj_as_list(response)
def zero_shot_classification(
self,
text: str,
labels: List[str],
*,
multi_label: bool = False,
hypothesis_template: Optional[str] = None,
model: Optional[str] = None,
) -> List[ZeroShotClassificationOutputElement]:
"""
Provide as input a text and a set of candidate labels to classify the input text.
Args:
text (`str`):
The input text to classify.
labels (`List[str]`):
List of strings. Each string is the verbalization of a possible label for the input text.
multi_label (`bool`):
Boolean. If True, the probability for each label is evaluated independently and multiple labels can have a probability close to 1 simultaneously or all probabilities can be close to 0.
If False, the labels are considered mutually exclusive and the probability over all labels always sums to 1. Defaults to False.
hypothesis_template (`str`, *optional*):
A template sentence string with curly brackets to which the label strings are added. The label strings are added at the position of the curly brackets "{}".
Zero-shot classifiers are based on NLI models, which evaluate if a hypothesis is entailed in another text or not.
For example, with hypothesis_template="This text is about {}." and labels=["economics", "politics"], the system internally creates the two hypotheses "This text is about economics." and "This text is about politics.".
The model then evaluates for both hypotheses if they are entailed in the provided `text` or not.
model (`str`, *optional*):
The model to use for inference. Can be a model ID hosted on the Hugging Face Hub or a URL to a deployed
Inference Endpoint. This parameter overrides the model defined at the instance level. If not provided, the default recommended zero-shot classification model will be used.
Returns:
`List[ZeroShotClassificationOutputElement]`: List of [`ZeroShotClassificationOutputElement`] items containing the predicted labels and their confidence.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example with `multi_label=False`:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> text = (
... "A new model offers an explanation for how the Galilean satellites formed around the solar system's"
... "largest world. Konstantin Batygin did not set out to solve one of the solar system's most puzzling"
... " mysteries when he went for a run up a hill in Nice, France."
... )
>>> labels = ["space & cosmos", "scientific discovery", "microbiology", "robots", "archeology"]
>>> client.zero_shot_classification(text, labels)
[
ZeroShotClassificationOutputElement(label='scientific discovery', score=0.7961668968200684),
ZeroShotClassificationOutputElement(label='space & cosmos', score=0.18570658564567566),
ZeroShotClassificationOutputElement(label='microbiology', score=0.00730885099619627),
ZeroShotClassificationOutputElement(label='archeology', score=0.006258360575884581),
ZeroShotClassificationOutputElement(label='robots', score=0.004559356719255447),
]
>>> client.zero_shot_classification(text, labels, multi_label=True)
[
ZeroShotClassificationOutputElement(label='scientific discovery', score=0.9829297661781311),
ZeroShotClassificationOutputElement(label='space & cosmos', score=0.755190908908844),
ZeroShotClassificationOutputElement(label='microbiology', score=0.0005462635890580714),
ZeroShotClassificationOutputElement(label='archeology', score=0.00047131875180639327),
ZeroShotClassificationOutputElement(label='robots', score=0.00030448526376858354),
]
```
Example with `multi_label=True` and a custom `hypothesis_template`:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.zero_shot_classification(
... text="I really like our dinner and I'm very happy. I don't like the weather though.",
... labels=["positive", "negative", "pessimistic", "optimistic"],
... multi_label=True,
... hypothesis_template="This text is {} towards the weather"
... )
[
ZeroShotClassificationOutputElement(label='negative', score=0.9231801629066467),
ZeroShotClassificationOutputElement(label='pessimistic', score=0.8760990500450134),
ZeroShotClassificationOutputElement(label='optimistic', score=0.0008674879791215062),
ZeroShotClassificationOutputElement(label='positive', score=0.0005250611575320363)
]
```
"""
parameters = {
"candidate_labels": labels,
"multi_label": multi_label,
"hypothesis_template": hypothesis_template,
}
payload = _prepare_payload(text, parameters=parameters)
response = self.post(
**payload,
task="zero-shot-classification",
model=model,
)
output = _bytes_to_dict(response)
return [
ZeroShotClassificationOutputElement.parse_obj_as_instance({"label": label, "score": score})
for label, score in zip(output["labels"], output["scores"])
]
def zero_shot_image_classification(
self,
image: ContentT,
labels: List[str],
*,
model: Optional[str] = None,
hypothesis_template: Optional[str] = None,
) -> List[ZeroShotImageClassificationOutputElement]:
"""
Provide input image and text labels to predict text labels for the image.
Args:
image (`Union[str, Path, bytes, BinaryIO]`):
The input image to caption. It can be raw bytes, an image file, or a URL to an online image.
labels (`List[str]`):
List of string possible labels. There must be at least 2 labels.
model (`str`, *optional*):
The model to use for inference. Can be a model ID hosted on the Hugging Face Hub or a URL to a deployed
Inference Endpoint. This parameter overrides the model defined at the instance level. If not provided, the default recommended zero-shot image classification model will be used.
hypothesis_template (`str`, *optional*):
The sentence used in conjunction with `labels` to attempt the text classification by replacing the
placeholder with the candidate labels.
Returns:
`List[ZeroShotImageClassificationOutputElement]`: List of [`ZeroShotImageClassificationOutputElement`] items containing the predicted labels and their confidence.
Raises:
[`InferenceTimeoutError`]:
If the model is unavailable or the request times out.
`HTTPError`:
If the request fails with an HTTP error status code other than HTTP 503.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.zero_shot_image_classification(
... "https://upload.wikimedia.org/wikipedia/commons/thumb/4/43/Cute_dog.jpg/320px-Cute_dog.jpg",
... labels=["dog", "cat", "horse"],
... )
[ZeroShotImageClassificationOutputElement(label='dog', score=0.956),...]
```
"""
# Raise ValueError if input is less than 2 labels
if len(labels) < 2:
raise ValueError("You must specify at least 2 classes to compare.")
inputs = {"image": _b64_encode(image), "candidateLabels": ",".join(labels)}
parameters = {"hypothesis_template": hypothesis_template}
payload = _prepare_payload(inputs, parameters=parameters)
response = self.post(
**payload,
model=model,
task="zero-shot-image-classification",
)
return ZeroShotImageClassificationOutputElement.parse_obj_as_list(response)
def _resolve_url(self, model: Optional[str] = None, task: Optional[str] = None) -> str:
model = model or self.model or self.base_url
# If model is already a URL, ignore `task` and return directly
if model is not None and (model.startswith("http://") or model.startswith("https://")):
return model
# # If no model but task is set => fetch the recommended one for this task
if model is None:
if task is None:
raise ValueError(
"You must specify at least a model (repo_id or URL) or a task, either when instantiating"
" `InferenceClient` or when making a request."
)
model = self.get_recommended_model(task)
logger.info(
f"Using recommended model {model} for task {task}. Note that it is"
f" encouraged to explicitly set `model='{model}'` as the recommended"
" models list might get updated without prior notice."
)
# Compute InferenceAPI url
return (
# Feature-extraction and sentence-similarity are the only cases where we handle models with several tasks.
f"{INFERENCE_ENDPOINT}/pipeline/{task}/{model}"
if task in ("feature-extraction", "sentence-similarity")
# Otherwise, we use the default endpoint
else f"{INFERENCE_ENDPOINT}/models/{model}"
)
@staticmethod
def get_recommended_model(task: str) -> str:
"""
Get the model Hugging Face recommends for the input task.
Args:
task (`str`):
The Hugging Face task to get which model Hugging Face recommends.
All available tasks can be found [here](https://huggingface.co/tasks).
Returns:
`str`: Name of the model recommended for the input task.
Raises:
`ValueError`: If Hugging Face has no recommendation for the input task.
"""
model = _fetch_recommended_models().get(task)
if model is None:
raise ValueError(
f"Task {task} has no recommended model. Please specify a model"
" explicitly. Visit https://huggingface.co/tasks for more info."
)
return model
def get_endpoint_info(self, *, model: Optional[str] = None) -> Dict[str, Any]:
"""
Get information about the deployed endpoint.
This endpoint is only available on endpoints powered by Text-Generation-Inference (TGI) or Text-Embedding-Inference (TEI).
Endpoints powered by `transformers` return an empty payload.
Args:
model (`str`, *optional*):
The model to use for inference. Can be a model ID hosted on the Hugging Face Hub or a URL to a deployed
Inference Endpoint. This parameter overrides the model defined at the instance level. Defaults to None.
Returns:
`Dict[str, Any]`: Information about the endpoint.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient("meta-llama/Meta-Llama-3-70B-Instruct")
>>> client.get_endpoint_info()
{
'model_id': 'meta-llama/Meta-Llama-3-70B-Instruct',
'model_sha': None,
'model_dtype': 'torch.float16',
'model_device_type': 'cuda',
'model_pipeline_tag': None,
'max_concurrent_requests': 128,
'max_best_of': 2,
'max_stop_sequences': 4,
'max_input_length': 8191,
'max_total_tokens': 8192,
'waiting_served_ratio': 0.3,
'max_batch_total_tokens': 1259392,
'max_waiting_tokens': 20,
'max_batch_size': None,
'validation_workers': 32,
'max_client_batch_size': 4,
'version': '2.0.2',
'sha': 'dccab72549635c7eb5ddb17f43f0b7cdff07c214',
'docker_label': 'sha-dccab72'
}
```
"""
model = model or self.model
if model is None:
raise ValueError("Model id not provided.")
if model.startswith(("http://", "https://")):
url = model.rstrip("/") + "/info"
else:
url = f"{INFERENCE_ENDPOINT}/models/{model}/info"
response = get_session().get(url, headers=self.headers)
hf_raise_for_status(response)
return response.json()
def health_check(self, model: Optional[str] = None) -> bool:
"""
Check the health of the deployed endpoint.
Health check is only available with Inference Endpoints powered by Text-Generation-Inference (TGI) or Text-Embedding-Inference (TEI).
For Inference API, please use [`InferenceClient.get_model_status`] instead.
Args:
model (`str`, *optional*):
URL of the Inference Endpoint. This parameter overrides the model defined at the instance level. Defaults to None.
Returns:
`bool`: True if everything is working fine.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient("https://jzgu0buei5.us-east-1.aws.endpoints.huggingface.cloud")
>>> client.health_check()
True
```
"""
model = model or self.model
if model is None:
raise ValueError("Model id not provided.")
if not model.startswith(("http://", "https://")):
raise ValueError(
"Model must be an Inference Endpoint URL. For serverless Inference API, please use `InferenceClient.get_model_status`."
)
url = model.rstrip("/") + "/health"
response = get_session().get(url, headers=self.headers)
return response.status_code == 200
def get_model_status(self, model: Optional[str] = None) -> ModelStatus:
"""
Get the status of a model hosted on the Inference API.
<Tip>
This endpoint is mostly useful when you already know which model you want to use and want to check its
availability. If you want to discover already deployed models, you should rather use [`~InferenceClient.list_deployed_models`].
</Tip>
Args:
model (`str`, *optional*):
Identifier of the model for witch the status gonna be checked. If model is not provided,
the model associated with this instance of [`InferenceClient`] will be used. Only InferenceAPI service can be checked so the
identifier cannot be a URL.
Returns:
[`ModelStatus`]: An instance of ModelStatus dataclass, containing information,
about the state of the model: load, state, compute type and framework.
Example:
```py
>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient()
>>> client.get_model_status("meta-llama/Meta-Llama-3-8B-Instruct")
ModelStatus(loaded=True, state='Loaded', compute_type='gpu', framework='text-generation-inference')
```
"""
model = model or self.model
if model is None:
raise ValueError("Model id not provided.")
if model.startswith("https://"):
raise NotImplementedError("Model status is only available for Inference API endpoints.")
url = f"{INFERENCE_ENDPOINT}/status/{model}"
response = get_session().get(url, headers=self.headers)
hf_raise_for_status(response)
response_data = response.json()
if "error" in response_data:
raise ValueError(response_data["error"])
return ModelStatus(
loaded=response_data["loaded"],
state=response_data["state"],
compute_type=response_data["compute_type"],
framework=response_data["framework"],
)
@property
def chat(self) -> "ProxyClientChat":
return ProxyClientChat(self)
class _ProxyClient:
"""Proxy class to be able to call `client.chat.completion.create(...)` as OpenAI client."""
def __init__(self, client: InferenceClient):
self._client = client
class ProxyClientChat(_ProxyClient):
"""Proxy class to be able to call `client.chat.completion.create(...)` as OpenAI client."""
@property
def completions(self) -> "ProxyClientChatCompletions":
return ProxyClientChatCompletions(self._client)
class ProxyClientChatCompletions(_ProxyClient):
"""Proxy class to be able to call `client.chat.completion.create(...)` as OpenAI client."""
@property
def create(self):
return self._client.chat_completion
Reference in New Issue View Git Blame Copy Permalink