File size: 42,285 Bytes
e1a9145 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 | """This module should not be used directly as its API is subject to change. Instead,
use the `gr.Blocks.load()` or `gr.load()` functions."""
from __future__ import annotations
import json
import os
import re
import tempfile
import warnings
from collections.abc import Callable, Generator
from pathlib import Path
from typing import TYPE_CHECKING, Literal
import httpx
import huggingface_hub
from gradio_client import Client
from gradio_client.client import Endpoint
from gradio_client.documentation import document
from gradio_client.utils import encode_url_or_file_to_base64, is_http_url_like
from packaging import version
import gradio as gr
from gradio import components, external_utils, utils
from gradio.components.multimodal_textbox import MultimodalValue
from gradio.context import Context
from gradio.exceptions import (
GradioVersionIncompatibleError,
TooManyRequestsError,
)
from gradio.processing_utils import save_base64_to_cache, to_binary
if TYPE_CHECKING:
from huggingface_hub.inference._providers import PROVIDER_T
from gradio.blocks import Blocks
from gradio.chat_interface import ChatInterface
from gradio.components.chatbot import NormalizedMessageDict
from gradio.components.login_button import LoginButton
from gradio.interface import Interface
@document()
def load(
name: str,
src: Callable[[str, str | None], Blocks]
| Literal["models", "spaces", "huggingface"]
| None = None,
token: str | None = None,
accept_token: bool | LoginButton = False,
provider: PROVIDER_T | None = None,
**kwargs,
) -> Blocks:
"""
Constructs a Gradio app automatically from a Hugging Face model/Space repo name or a 3rd-party API provider. Note that if a Space repo is loaded, certain high-level attributes of the Blocks (e.g. custom `css`, `js`, and `head` attributes) will not be loaded.
Parameters:
name: the name of the model (e.g. "google/vit-base-patch16-224") or Space (e.g. "flax-community/spanish-gpt2"). This is the first parameter passed into the `src` function. Can also be formatted as {src}/{repo name} (e.g. "models/google/vit-base-patch16-224") if `src` is not provided.
src: function that accepts a string model `name` and a string or None `token` and returns a Gradio app. Alternatively, this parameter takes one of two strings for convenience: "models" (for loading a Hugging Face model through the Inference API) or "spaces" (for loading a Hugging Face Space). If None, uses the prefix of the `name` parameter to determine `src`.
token: optional token that is passed as the second parameter to the `src` function. If not explicitly provided, will use the HF_TOKEN environment variable or fallback to the locally-saved HF token when loading models but not Spaces (when loading Spaces, only provide a token if you are loading a trusted private Space as the token can be read by the Space you are loading). Find your HF tokens here: https://huggingface.co/settings/tokens.
accept_token: if True, a Textbox component is first rendered to allow the user to provide a token, which will be used instead of the `token` parameter when calling the loaded model or Space. Can also provide an instance of a gr.LoginButton in the same Blocks scope, which allows the user to login with a Hugging Face account whose token will be used instead of the `token` parameter when calling the loaded model or Space.
kwargs: additional keyword parameters to pass into the `src` function. If `src` is "models" or "Spaces", these parameters are passed into the `gr.Interface` or `gr.ChatInterface` constructor.
provider: the name of the third-party (non-Hugging Face) providers to use for model inference (e.g. "replicate", "sambanova", "fal-ai", etc). Should be one of the providers supported by `huggingface_hub.InferenceClient`. This parameter is only used when `src` is "models"
Returns:
a Gradio Blocks app for the given model
Example:
import gradio as gr
demo = gr.load("gradio/question-answering", src="spaces")
demo.launch()
"""
if src is None:
# Separate the repo type (e.g. "model") from repo name (e.g. "google/vit-base-patch16-224")
parts = name.split("/")
if len(parts) <= 1:
raise ValueError(
"Either `src` parameter must be provided, or `name` must be formatted as {src}/{repo name}"
)
src = parts[0] # type: ignore
name = "/".join(parts[1:])
assert src is not None # noqa: S101
if not isinstance(src, Callable) and src not in ["models", "spaces", "huggingface"]:
raise ValueError(
"The `src` parameter must be one of 'huggingface', 'models', 'spaces', or a function that accepts a model name (and optionally, a token), and returns a Gradio app."
)
if (
token is None
and src in ["models", "huggingface"]
and os.environ.get("HF_TOKEN") is not None
):
token = os.environ.get("HF_TOKEN")
if isinstance(src, Callable):
return src(name, token, **kwargs) # type: ignore
if not accept_token:
return load_blocks_from_huggingface(
name=name,
src=src, # type: ignore
token=token,
provider=provider,
**kwargs, # type: ignore
)
elif isinstance(accept_token, gr.LoginButton):
with gr.Blocks(fill_height=True) as demo:
if not accept_token.is_rendered:
accept_token.render()
@gr.render(triggers=[demo.load])
def create_blocks(oauth_token: gr.OAuthToken | None):
token_value = None if oauth_token is None else oauth_token.token
return load_blocks_from_huggingface(
name=name,
src=src, # type: ignore
token=token_value,
provider=provider,
**kwargs,
)
return demo
else:
with gr.Blocks(fill_height=True) as demo:
with gr.Accordion("Enter your token and press enter") as accordion:
textbox = gr.Textbox(
type="password",
show_label=False,
container=False,
)
remember_token = gr.Checkbox(
label="Remember me on this device", value=False, container=False
)
browser_state = gr.BrowserState()
@gr.on([textbox.submit], outputs=accordion)
def hide_accordion():
return gr.Accordion("Token settings", open=False)
@gr.on(
[textbox.submit, remember_token.change],
inputs=[textbox, remember_token],
outputs=[browser_state, remember_token],
)
def save_token(token_value, remember_token_value):
if remember_token_value and token_value:
return token_value, gr.Checkbox(
label="Remember me on this device (saved!)", value=True
)
else:
return "", gr.Checkbox(label="Remember me on this device")
@gr.on(demo.load, inputs=[browser_state], outputs=[textbox, remember_token])
def load_token(token_value):
if token_value:
return token_value, True
else:
return "", False
@gr.render(inputs=[textbox], triggers=[textbox.submit])
def create(token_value):
return load_blocks_from_huggingface(
name=name,
src=src, # type: ignore
token=token_value,
provider=provider,
**kwargs,
)
return demo
def load_blocks_from_huggingface(
name: str,
src: str,
token: str | None = None,
alias: str | None = None,
provider: PROVIDER_T | None = None,
**kwargs,
) -> Blocks:
"""Creates and returns a Blocks instance from a Hugging Face model or Space repo."""
if token is not None:
if Context.token is not None and Context.token != token:
warnings.warn(
"""You are loading a model/Space with a different access token than the one you used to load a previous model/Space. This is not recommended, as it may cause unexpected behavior."""
)
Context.token = token
if src == "spaces":
blocks = from_spaces(
name, token=token, alias=alias, provider=provider, **kwargs
)
else:
blocks = from_model(name, token=token, alias=alias, provider=provider, **kwargs)
return blocks
def from_model(
model_name: str,
token: str | None,
alias: str | None,
provider: PROVIDER_T | None = None,
**kwargs,
) -> Blocks:
headers = {"X-Wait-For-Model": "true"}
client = huggingface_hub.InferenceClient(
model=model_name, headers=headers, token=token, provider=provider
)
p, tags = external_utils.get_model_info(model_name, token)
# For tasks that are not yet supported by the InferenceClient
api_url = f"https://api-inference.huggingface.co/models/{model_name}"
GRADIO_CACHE = os.environ.get("GRADIO_TEMP_DIR") or str( # noqa: N806
Path(tempfile.gettempdir()) / "gradio"
)
def custom_post_binary(data):
data = to_binary({"path": data})
response = httpx.request("POST", api_url, headers=headers, content=data)
return save_base64_to_cache(
external_utils.encode_to_base64(response), cache_dir=GRADIO_CACHE
)
preprocess = None
postprocess = None
examples = None
# example model: ehcalabres/wav2vec2-lg-xlsr-en-speech-emotion-recognition
if p == "audio-classification":
inputs = components.Audio(type="filepath", label="Input")
outputs = components.Label(label="Class")
postprocess = external_utils.postprocess_label
examples = [
"https://gradio-builds.s3.amazonaws.com/demo-files/audio_sample.wav"
]
fn = client.audio_classification
# example model: facebook/xm_transformer_sm_all-en
elif p == "audio-to-audio":
inputs = components.Audio(type="filepath", label="Input")
outputs = components.Audio(label="Output")
examples = [
"https://gradio-builds.s3.amazonaws.com/demo-files/audio_sample.wav"
]
fn = custom_post_binary
# example model: facebook/wav2vec2-base-960h
elif p == "automatic-speech-recognition":
inputs = components.Audio(type="filepath", label="Input")
outputs = components.Textbox(label="Output")
examples = [
"https://gradio-builds.s3.amazonaws.com/demo-files/audio_sample.wav"
]
fn = client.automatic_speech_recognition
postprocess = lambda x: x.text # noqa: E731
# example model: julien-c/distilbert-feature-extraction
elif p == "feature-extraction":
inputs = components.Textbox(label="Input")
outputs = components.Dataframe(label="Output")
fn = client.feature_extraction
postprocess = utils.resolve_singleton
# example model: distilbert/distilbert-base-uncased
elif p == "fill-mask":
inputs = components.Textbox(label="Input")
outputs = components.Label(label="Classification")
examples = [
"Hugging Face is the AI community, working together, to [MASK] the future."
]
postprocess = external_utils.postprocess_mask_tokens
fn = client.fill_mask
# Example: google/vit-base-patch16-224
elif p == "image-classification":
inputs = components.Image(type="filepath", label="Input Image")
outputs = components.Label(label="Classification")
postprocess = external_utils.postprocess_label
examples = ["https://gradio-builds.s3.amazonaws.com/demo-files/cheetah-002.jpg"]
fn = client.image_classification
# Example: deepset/xlm-roberta-base-squad2
elif p == "question-answering":
inputs = [
components.Textbox(label="Question"),
components.Textbox(lines=7, label="Context"),
]
outputs = [
components.Textbox(label="Answer"),
components.Label(label="Score"),
]
examples = [
[
"What entity was responsible for the Apollo program?",
"The Apollo program, also known as Project Apollo, was the third United States human spaceflight"
" program carried out by the National Aeronautics and Space Administration (NASA), which accomplished"
" landing the first humans on the Moon from 1969 to 1972.",
]
]
postprocess = external_utils.postprocess_question_answering
fn = client.question_answering
# Example: facebook/bart-large-cnn
elif p == "summarization":
inputs = components.Textbox(label="Input")
outputs = components.Textbox(label="Summary")
examples = [
[
"The tower is 324 metres (1,063 ft) tall, about the same height as an 81-storey building, and the tallest structure in Paris. Its base is square, measuring 125 metres (410 ft) on each side. During its construction, the Eiffel Tower surpassed the Washington Monument to become the tallest man-made structure in the world, a title it held for 41 years until the Chrysler Building in New York City was finished in 1930. It was the first structure to reach a height of 300 metres. Due to the addition of a broadcasting aerial at the top of the tower in 1957, it is now taller than the Chrysler Building by 5.2 metres (17 ft). Excluding transmitters, the Eiffel Tower is the second tallest free-standing structure in France after the Millau Viaduct."
]
]
postprocess = lambda x: x.summary_text # noqa: E731
fn = client.summarization
# Example: distilbert-base-uncased-finetuned-sst-2-english
elif p == "text-classification":
inputs = components.Textbox(label="Input")
outputs = components.Label(label="Classification")
examples = ["I feel great"]
postprocess = external_utils.postprocess_label
fn = client.text_classification
# Example: gpt2
elif p == "text-generation":
# Example: meta-llama/Meta-Llama-3-8B-Instruct
if tags and "conversational" in tags:
from gradio import ChatInterface
from gradio.components import Chatbot
fn = external_utils.conversational_wrapper(client)
examples = [
"What is the capital of Pakistan?",
"Tell me a joke about calculus.",
"Explain gravity to a 5-year-old.",
"What were the main causes of World War I?",
]
chat_interface_kwargs = {
"examples": examples,
}
kwargs = dict(chat_interface_kwargs, **kwargs)
chatbot = Chatbot(scale=1, allow_tags=True)
return ChatInterface(fn, chatbot=chatbot, **kwargs) # type: ignore
inputs = components.Textbox(label="Text")
outputs = inputs
examples = ["Once upon a time"]
fn = external_utils.text_generation_wrapper(client)
# Example: valhalla/t5-small-qa-qg-hl
elif p == "text2text-generation":
inputs = components.Textbox(label="Input")
outputs = components.Textbox(label="Generated Text")
examples = ["Translate English to Arabic: How are you?"]
fn = client.text_generation
# Example: Helsinki-NLP/opus-mt-en-ar
elif p == "translation":
inputs = components.Textbox(label="Input")
outputs = components.Textbox(label="Translation")
postprocess = lambda x: x.translation_text # noqa: E731
examples = ["Hello, how are you?"]
fn = client.translation
# Example: facebook/bart-large-mnli
elif p == "zero-shot-classification":
inputs = [
components.Textbox(label="Input"),
components.Textbox(label="Possible class names (comma-separated)"),
components.Checkbox(label="Allow multiple true classes"),
]
outputs = components.Label(label="Classification")
postprocess = external_utils.postprocess_label
examples = [["I feel great", "happy, sad", False]]
fn = external_utils.zero_shot_classification_wrapper(client)
# Example: sentence-transformers/distilbert-base-nli-stsb-mean-tokens
elif p == "sentence-similarity":
inputs = [
components.Textbox(
label="Source Sentence",
placeholder="Enter an original sentence",
),
components.Textbox(
lines=7,
placeholder="Sentences to compare to -- separate each sentence by a newline",
label="Sentences to compare to",
),
]
outputs = components.JSON(label="Similarity scores")
examples = [["That is a happy person", "That person is very happy"]]
fn = external_utils.sentence_similarity_wrapper(client)
# Example: julien-c/ljspeech_tts_train_tacotron2_raw_phn_tacotron_g2p_en_no_space_train
elif p == "text-to-speech":
inputs = components.Textbox(label="Input")
outputs = components.Audio(label="Audio")
examples = ["Hello, how are you?"]
fn = client.text_to_speech
# example model: osanseviero/BigGAN-deep-128
elif p == "text-to-image":
inputs = components.Textbox(label="Input")
outputs = components.Image(label="Output")
examples = ["A beautiful sunset"]
fn = client.text_to_image
# example model: huggingface-course/bert-finetuned-ner
elif p == "token-classification":
inputs = components.Textbox(label="Input")
outputs = components.HighlightedText(label="Output")
examples = [
"Hugging Face is a company based in Paris and New York City that acquired Gradio in 2021."
]
fn = external_utils.token_classification_wrapper(client)
# example model: impira/layoutlm-document-qa
elif p == "document-question-answering":
inputs = [
components.Image(type="filepath", label="Input Document"),
components.Textbox(label="Question"),
]
postprocess = external_utils.postprocess_label
outputs = components.Label(label="Label")
fn = client.document_question_answering
# example model: dandelin/vilt-b32-finetuned-vqa
elif p == "visual-question-answering":
inputs = [
components.Image(type="filepath", label="Input Image"),
components.Textbox(label="Question"),
]
outputs = components.Label(label="Label")
postprocess = external_utils.postprocess_visual_question_answering
examples = [
[
"https://gradio-builds.s3.amazonaws.com/demo-files/cheetah-002.jpg",
"What animal is in the image?",
]
]
fn = client.visual_question_answering
# example model: Salesforce/blip-image-captioning-base
elif p == "image-to-text":
inputs = components.Image(type="filepath", label="Input Image")
outputs = components.Textbox(label="Generated Text")
examples = ["https://gradio-builds.s3.amazonaws.com/demo-files/cheetah-002.jpg"]
fn = client.image_to_text
# example model: rajistics/autotrain-Adult-934630783
elif p in ["tabular-classification", "tabular-regression"]:
examples = external_utils.get_tabular_examples(model_name)
col_names, examples = external_utils.cols_to_rows(examples) # type: ignore
examples = [[examples]] if examples else None
inputs = components.Dataframe(
label="Input Rows",
type="pandas",
headers=col_names,
column_count=len(col_names),
column_limits=(len(col_names), len(col_names)),
render=False,
)
outputs = components.Dataframe(
label="Predictions", type="array", headers=["prediction"]
)
fn = external_utils.tabular_wrapper
# example model: microsoft/table-transformer-detection
elif p == "object-detection":
inputs = components.Image(type="filepath", label="Input Image")
outputs = components.AnnotatedImage(label="Annotations")
fn = external_utils.object_detection_wrapper(client)
# example model: stabilityai/stable-diffusion-xl-refiner-1.0
elif p == "image-to-image":
inputs = [
components.Image(type="filepath", label="Input Image"),
components.Textbox(label="Input"),
]
outputs = components.Image(label="Output")
examples = [
[
"https://gradio-builds.s3.amazonaws.com/demo-files/cheetah-002.jpg",
"Photo of a cheetah with green eyes",
]
]
fn = client.image_to_image
# example model: meta-llama/Llama-3.2-11B-Vision-Instruct
elif p == "image-text-to-text":
inputs = [
components.Image(type="filepath", label="Input Image"),
components.Textbox(
label="Input Text", placeholder="Ask a question about the image"
),
]
outputs = components.Textbox(label="Generated Text")
examples = [
[
"https://gradio-builds.s3.amazonaws.com/demo-files/cheetah-002.jpg",
"What animal is in the image?",
]
]
fn = external_utils.image_text_to_text_wrapper(client)
else:
raise ValueError(f"Unsupported pipeline type: {p}")
def query_huggingface_inference_endpoints(*data):
if preprocess is not None:
data = preprocess(*data)
try:
data = fn(*data)
except Exception as e:
external_utils.handle_hf_error(e)
if postprocess is not None:
data = postprocess(data) # type: ignore
return data
query_huggingface_inference_endpoints.__name__ = alias or model_name
interface_info = {
"fn": query_huggingface_inference_endpoints,
"inputs": inputs,
"outputs": outputs,
"title": model_name,
"examples": examples,
"cache_examples": False,
}
kwargs = dict(interface_info, **kwargs)
interface_fn = kwargs.pop("fn", None)
inputs = kwargs.pop("inputs", None)
outputs = kwargs.pop("outputs", None)
interface = gr.Interface(
interface_fn, inputs, outputs, **kwargs, api_name="predict"
)
return interface
def from_spaces(
space_name: str,
token: str | None,
alias: str | None,
provider: PROVIDER_T | None = None,
**kwargs,
) -> Blocks:
if provider is not None:
warnings.warn(
"The `provider` parameter is not supported when loading Spaces. It will be ignored."
)
space_url = f"https://huggingface.co/spaces/{space_name}"
print(f"Fetching Space from: {space_url}")
headers = {}
if token not in [False, None]:
headers["Authorization"] = f"Bearer {token}"
iframe_url = (
httpx.get(
f"https://huggingface.co/api/spaces/{space_name}/host", headers=headers
)
.json()
.get("host")
)
if iframe_url is None:
raise ValueError(
f"Could not find Space: {space_name}. If it is a private or gated Space, please provide your Hugging Face access token (https://huggingface.co/settings/tokens) as the argument for the `token` parameter."
)
config_request = httpx.get(iframe_url + "/config", headers=headers)
if config_request.status_code == 404:
r = httpx.get(iframe_url, headers=headers)
result = re.search(
r"window.gradio_config = (.*?);[\s]*</script>", r.text
) # some basic regex to extract the config
try:
config = json.loads(result.group(1)) # type: ignore
except AttributeError as ae:
raise ValueError(f"Could not load the Space: {space_name}") from ae
elif config_request.status_code == 200:
config = config_request.json()
else:
raise ValueError(
f"Could not load the Space: {space_name} because the config could not be fetched."
)
if "allow_flagging" in config: # Create an Interface for Gradio 2.x Spaces
return from_spaces_interface(
space_name, config, alias, token, iframe_url, **kwargs
)
else: # Create a Blocks for Gradio 3.x Spaces
if kwargs:
warnings.warn(
"You cannot override parameters for this Space by passing in kwargs. "
"Instead, please load the Space as a function and use it to create a "
"Blocks or Interface locally. You may find this Guide helpful: "
"https://gradio.app/using_blocks_like_functions/"
)
return from_spaces_blocks(space=space_name, token=token)
def make_event_data_fn(client, endpoint):
"""Create a function that accepts EventData.
The event_data_fn has to be created in this closure so that the value of endpoint
is correctly captured."""
helper = client.new_helper(endpoint.fn_index)
def event_data_fn(event_data: gr.EventData, *args):
fn = endpoint.make_end_to_end_fn(helper)
return fn(*args, event_data=event_data._data)
return event_data_fn
def from_spaces_blocks(space: str, token: str | None) -> Blocks:
client = Client(
space,
token=token,
download_files=False,
_skip_components=False,
)
# We set deserialize to False to avoid downloading output files from the server.
# Instead, we serve them as URLs using the /proxy/ endpoint directly from the server.
if client.app_version < version.Version("4.0.0b14"):
raise GradioVersionIncompatibleError(
f"Gradio version 4.x cannot load spaces with versions less than 4.x ({client.app_version})."
"Please downgrade to version 3 to load this space."
)
# Use end_to_end_fn here to properly upload/download all files
predict_fns = []
for endpoint in client.endpoints.values():
if not isinstance(endpoint, Endpoint):
raise TypeError(
f"Expected endpoint to be an Endpoint, but got {type(endpoint)}"
)
if endpoint.backend_fn:
dep_config = next(
(
d
for d in client.config["dependencies"]
if d.get("id") == endpoint.fn_index
),
{},
)
if dep_config.get("collects_event_data"):
event_data_fn = make_event_data_fn(client, endpoint)
predict_fns.append(event_data_fn)
else:
helper = client.new_helper(endpoint.fn_index)
fn = endpoint.make_end_to_end_fn(helper)
predict_fns.append(fn)
else:
predict_fns.append(None)
blocks = gr.Blocks.from_config(client.config, predict_fns, client.src) # type: ignore
with blocks:
# Reset the session_hash when page loads
blocks.load(lambda: client.reset_session(), None, None)
return blocks
def from_spaces_interface(
model_name: str,
config: dict,
alias: str | None,
token: str | None,
iframe_url: str,
**kwargs,
) -> Interface:
config = external_utils.streamline_spaces_interface(config)
api_url = f"{iframe_url}/api/predict/"
headers = {"Content-Type": "application/json"}
if token not in [False, None]:
headers["Authorization"] = f"Bearer {token}"
# The function should call the API with preprocessed data
def fn(*data):
data = json.dumps({"data": data})
response = httpx.post(api_url, headers=headers, data=data) # type: ignore
result = json.loads(response.content.decode("utf-8"))
if "error" in result and "429" in result["error"]:
raise TooManyRequestsError("Too many requests to the Hugging Face API")
try:
output = result["data"]
except KeyError as ke:
raise KeyError(
f"Could not find 'data' key in response from external Space. Response received: {result}"
) from ke
if (
len(config["outputs"]) == 1
): # if the fn is supposed to return a single value, pop it
output = output[0]
if (
len(config["outputs"]) == 1 and isinstance(output, list)
): # Needed to support Output.Image() returning bounding boxes as well (TODO: handle different versions of gradio since they have slightly different APIs)
output = output[0]
return output
fn.__name__ = alias if (alias is not None) else model_name
config["fn"] = fn
kwargs = dict(config, **kwargs)
kwargs["_api_mode"] = True
fn = kwargs.pop("fn", None)
inputs = kwargs.pop("inputs", None)
outputs = kwargs.pop("outputs", None)
interface = gr.Interface(fn, inputs, outputs, **kwargs)
return interface
TEXT_FILE_EXTENSIONS = (
".doc",
".docx",
".rtf",
".epub",
".odt",
".odp",
".pptx",
".txt",
".md",
".py",
".ipynb",
".js",
".jsx",
".html",
".css",
".java",
".cs",
".php",
".c",
".cc",
".cpp",
".cxx",
".cts",
".h",
".hh",
".hpp",
".rs",
".R",
".Rmd",
".swift",
".go",
".rb",
".kt",
".kts",
".ts",
".tsx",
".m",
".mm",
".mts",
".scala",
".dart",
".lua",
".pl",
".pm",
".t",
".sh",
".bash",
".zsh",
".bat",
".coffee",
".csv",
".log",
".ini",
".cfg",
".config",
".json",
".proto",
".yaml",
".yml",
".toml",
".sql",
)
IMAGE_FILE_EXTENSIONS = (".png", ".jpg", ".jpeg", ".gif", ".webp")
def format_conversation(
history: list[NormalizedMessageDict], new_message: str | MultimodalValue
) -> list[dict]:
conversation = []
for message in history:
new_content = []
for content in message["content"]:
if "type" not in content:
raise ValueError(
f"Invalid message format: {message['content']}. Each element must have a type key."
)
elif content["type"] == "file":
new_content.append(
{
"type": "image_url",
"image_url": {
"url": encode_url_or_file_to_base64(content["file"]["path"]) # type: ignore
},
}
)
else:
new_content.append(content)
message["content"] = new_content
conversation.append(message)
if isinstance(new_message, str):
text = new_message
files = []
else:
text = new_message.get("text", None)
files = new_message.get("files", [])
image_files, text_encoded = [], []
for file in files:
if file.lower().endswith(TEXT_FILE_EXTENSIONS):
text_encoded.append(file)
else:
image_files.append(file)
for image in image_files:
conversation.append(
{
"role": "user",
"content": [
{
"type": "image_url",
"image_url": {"url": encode_url_or_file_to_base64(image)},
}
],
}
)
if text or text_encoded:
text = text or ""
text += "\n".join(
[
f"\n## {Path(file).name}\n{Path(file).read_text()}"
for file in text_encoded
]
)
conversation.append(
{"role": "user", "content": [{"type": "text", "text": text}]}
)
return conversation
@document()
def load_chat(
base_url: str,
model: str,
token: str | None = None,
*,
file_types: Literal["text_encoded", "image"]
| list[Literal["text_encoded", "image"]]
| None = "text_encoded",
system_message: str | None = None,
streaming: bool = True,
**kwargs,
) -> ChatInterface:
"""
Load a chat interface from an OpenAI API chat compatible endpoint.
Parameters:
base_url: The base URL of the endpoint, e.g. "http://localhost:11434/v1/"
model: The name of the model you are loading, e.g. "llama3.2"
token: The API token or a placeholder string if you are using a local model, e.g. "ollama"
file_types: The file types allowed to be uploaded by the user. "text_encoded" allows uploading any text-encoded file (which is simply appended to the prompt), and "image" adds image upload support. Set to None to disable file uploads.
system_message: The system message to use for the conversation, if any.
streaming: Whether the response should be streamed.
kwargs: Additional keyword arguments to pass into ChatInterface for customization.
Example:
import gradio as gr
gr.load_chat(
"http://localhost:11434/v1/",
model="qwen2.5",
token="***",
file_types=["text_encoded", "image"],
system_message="You are a silly assistant.",
).launch()
"""
try:
from openai import OpenAI
except ImportError as e:
raise ImportError(
"To use OpenAI API Client, you must install the `openai` package. You can install it with `pip install openai`."
) from e
from gradio.chat_interface import ChatInterface
client = OpenAI(api_key=token or "***", base_url=base_url)
start_message = (
[{"role": "system", "content": system_message}] if system_message else []
)
file_types = utils.none_or_singleton_to_list(file_types)
def open_api(message: str | MultimodalValue, history: list | None) -> str | None:
history = history or start_message
conversation = format_conversation(history, message) # type: ignore
return (
client.chat.completions.create(
model=model,
messages=conversation, # type: ignore
)
.choices[0]
.message.content
)
def open_api_stream(
message: str | MultimodalValue, history: list | None
) -> Generator[str, None, None]:
history = history or start_message
conversation = format_conversation(history, message) # type: ignore
stream = client.chat.completions.create(
model=model,
messages=conversation, # type: ignore
stream=True,
)
response = ""
for chunk in stream:
if chunk.choices and chunk.choices[0].delta.content is not None:
response += chunk.choices[0].delta.content
yield response
supported_extensions = []
for file_type in file_types:
if file_type == "text_encoded":
supported_extensions += TEXT_FILE_EXTENSIONS
elif file_type == "image":
supported_extensions += IMAGE_FILE_EXTENSIONS
else:
raise ValueError(
f"Invalid file type: {file_type}. Must be 'text_encoded' or 'image'."
)
if "chatbot" not in kwargs:
from gradio.components import Chatbot
kwargs["chatbot"] = Chatbot(scale=1, allow_tags=True)
textbox_arg = kwargs.pop("textbox", None)
if textbox_arg is not None:
textbox = textbox_arg
else:
textbox = (
gr.MultimodalTextbox(file_types=supported_extensions)
if file_types
else None
)
return ChatInterface(
open_api_stream if streaming else open_api,
multimodal=bool(file_types),
textbox=textbox,
**kwargs,
)
@document()
def load_openapi(
openapi_spec: str | dict,
base_url: str,
*,
paths: list[str] | None = None,
exclude_paths: list[str] | None = None,
methods: list[Literal["get", "post", "put", "patch", "delete"]] | None = None,
auth_token: str | None = None,
**interface_kwargs,
) -> Blocks:
"""
Load a Gradio app from an OpenAPI v3 specification.
Parameters:
openapi_spec: URL, file path, or dictionary containing the OpenAPI specification (v3, JSON format only)
base_url: Base URL for the API endpoints, e.g. "https://api.example.com/v1". This is used to construct the full URL for each endpoint.
paths: Optional list of specific API paths to create Gradio endpoints from. Supports regex patterns, e.g. ["/api/v1/books", ".*user.*"]. If None, all paths in the OpenAPI spec will be included.
exclude_paths: Optional list of API paths to exclude from the Gradio endpoints. Supports regex patterns and takes precedence over `paths`. For example, [".*internal.*"] will exclude all paths containing "internal".
methods: Optional list of HTTP methods to include in the Gradio endpoints. If None, all methods will be included.
auth_token: Optional authentication token to be sent as a Bearer token in the Authorization header for all API requests.
interface_kwargs: Additional keyword arguments to pass to each generated gr.Interface instance (e.g., title, description, article, examples_per_page, etc.)
Returns:
A Gradio Blocks app with endpoints generated from the OpenAPI spec
"""
if isinstance(openapi_spec, dict):
spec = openapi_spec
elif isinstance(openapi_spec, str):
if is_http_url_like(openapi_spec):
response = httpx.get(openapi_spec)
response.raise_for_status()
content = response.text
else:
with open(openapi_spec) as f:
content = f.read()
try:
spec = json.loads(content)
except json.JSONDecodeError as e:
raise ValueError("openapi_spec must be a JSON string or dictionary") from e
else:
raise ValueError("openapi_spec must be a string (URL/file path) or dictionary")
api_paths = spec.get("paths", {})
if paths is not None or exclude_paths is not None:
filtered_paths = {}
for path in api_paths:
if exclude_paths:
excluded = False
for exclude_pattern in exclude_paths:
if re.match(exclude_pattern, path):
excluded = True
break
if excluded:
continue
if paths is not None:
for pattern in paths:
if re.match(pattern, path):
filtered_paths[path] = api_paths[path]
break
else:
filtered_paths[path] = api_paths[path]
api_paths = filtered_paths
if not api_paths:
raise ValueError("No valid paths found in the OpenAPI specification")
else:
print(f"* Loaded {len(api_paths)} paths from the OpenAPI specification")
valid_api_paths = []
for path, path_item in api_paths.items():
for method, operation in path_item.items():
if methods and method.lower() not in [m.lower() for m in methods]:
continue
valid_api_paths.append((path, method, operation))
with gr.Blocks(
title=spec.get("info", {}).get("title", "OpenAPI Interface")
) as demo:
with gr.Sidebar():
gr.Markdown(f"## {spec.get('info', {}).get('title', 'OpenAPI Interface')}")
gr.Markdown(spec.get("info", {}).get("description", ""))
gr.Markdown("### API Endpoints")
api_path_str = "<div style='overflow-x: auto; overflow-y: auto; max-height: 500px;'><ul>"
for path, method, _ in valid_api_paths:
api_path_str += (
f"<li style='white-space: nowrap;'>"
f"<a href='#{method.upper()}_{path.replace('/', '_').replace('{', '').replace('}', '')}' style='white-space: nowrap;'>"
f"{method.upper()} <code style='font-size: inherit; white-space: nowrap;'>{path}</code>"
f"</a></li>\n"
)
api_path_str += "</ul></div>"
gr.Markdown(api_path_str)
for path, method, operation in valid_api_paths:
components_list = []
for param in operation.get("parameters", []):
component = external_utils.component_from_parameter_schema(param)
components_list.append(component)
request_body = operation.get("requestBody")
if request_body:
body_component = external_utils.component_from_request_body_schema(
request_body, spec
)
if body_component:
components_list.append(body_component)
endpoint_fn = external_utils.create_endpoint_fn(
path, method, operation, base_url, auth_token
)
endpoint_fn.__name__ = (
f"{method}_{path.replace('/', '_').replace('{', '').replace('}', '')}"
)
gr.Markdown(
f"<h2 id='{method.upper()}_{path.replace('/', '_').replace('{', '').replace('}', '')}'>"
f"{external_utils.method_box(method)} <code style='font-size: inherit;'>{path}</code></h2>"
)
if operation.get("summary"):
gr.Markdown(operation["summary"])
inputs = components_list if components_list else []
output = gr.JSON(label="Response")
gr.Interface(
fn=endpoint_fn,
inputs=inputs,
outputs=output,
**interface_kwargs,
)
return demo
|