"""The main Client class for the Python client.""" from __future__ import annotations import concurrent.futures import hashlib import json import math import os import re import secrets import shutil import tempfile import threading import time import urllib.parse import uuid import warnings from collections.abc import Callable from concurrent.futures import Future from dataclasses import dataclass from datetime import datetime from functools import partial from pathlib import Path from threading import Lock from typing import Any, Literal import httpx import huggingface_hub from huggingface_hub import CommitOperationAdd, SpaceHardware, SpaceStage from huggingface_hub.utils import ( RepositoryNotFoundError, build_hf_headers, send_telemetry, ) from packaging import version from gradio_client import utils from gradio_client.compatibility import EndpointV3Compatibility from gradio_client.data_classes import ParameterInfo from gradio_client.documentation import document from gradio_client.exceptions import AppError, AuthenticationError from gradio_client.utils import ( Communicator, JobStatus, Message, QueueError, ServerMessage, Status, StatusUpdate, ) DEFAULT_TEMP_DIR = os.environ.get("GRADIO_TEMP_DIR") or str( Path(tempfile.gettempdir()) / "gradio" ) @document("predict", "submit", "view_api", "duplicate", "deploy_discord") class Client: """ The main Client class for the Python client. This class is used to connect to a remote Gradio app and call its API endpoints. Example: from gradio_client import Client client = Client("abidlabs/whisper-large-v2") # connecting to a Hugging Face Space client.predict("test.mp4", api_name="/predict") >> What a nice recording! # returns the result of the remote API call client = Client("https://bec81a83-5b5c-471e.gradio.live") # connecting to a temporary Gradio share URL job = client.submit("hello", api_name="/predict") # runs the prediction in a background thread job.result() >> 49 # returns the result of the remote API call (blocking call) """ def __init__( self, src: str, hf_token: str | Literal[False] | None = False, max_workers: int = 40, verbose: bool = True, auth: tuple[str, str] | None = None, httpx_kwargs: dict[str, Any] | None = None, *, headers: dict[str, str] | None = None, download_files: str | Path | Literal[False] = DEFAULT_TEMP_DIR, ssl_verify: bool = True, _skip_components: bool = True, # internal parameter to skip values certain components (e.g. State) that do not need to be displayed to users. analytics_enabled: bool = True, ): """ Parameters: src: either the name of the Hugging Face Space to load, (e.g. "abidlabs/whisper-large-v2") or the full URL (including "http" or "https") of the hosted Gradio app to load (e.g. "http://mydomain.com/app" or "https://bec81a83-5b5c-471e.gradio.live/"). hf_token: optional Hugging Face token to use to access private Spaces. By default, no token is sent to the server. Set `hf_token=None` to use the locally saved token if there is one (warning: 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 tokens here: https://huggingface.co/settings/tokens. max_workers: maximum number of thread workers that can be used to make requests to the remote Gradio app simultaneously. verbose: whether the client should print statements to the console. headers: additional headers to send to the remote Gradio app on every request. By default only the HF authorization and user-agent headers are sent. This parameter will override the default headers if they have the same keys. download_files: directory where the client should download output files on the local machine from the remote API. By default, uses the value of the GRADIO_TEMP_DIR environment variable which, if not set by the user, is a temporary directory on your machine. If False, the client does not download files and returns a FileData dataclass object with the filepath on the remote machine instead. ssl_verify: if False, skips certificate validation which allows the client to connect to Gradio apps that are using self-signed certificates. httpx_kwargs: additional keyword arguments to pass to `httpx.Client`, `httpx.stream`, `httpx.get` and `httpx.post`. This can be used to set timeouts, proxies, http auth, etc. analytics_enabled: Whether to allow basic telemetry. If None, will use GRADIO_ANALYTICS_ENABLED environment variable or default to True. """ self.verbose = verbose self.hf_token = hf_token self.download_files = download_files self._skip_components = _skip_components self.headers = build_hf_headers( token=hf_token, library_name="gradio_client", library_version=utils.__version__, ) if headers: self.headers.update(headers) self.ssl_verify = ssl_verify self.space_id = None self.cookies: dict[str, str] = {} if isinstance(self.download_files, (str, Path)): if not os.path.exists(self.download_files): os.makedirs(self.download_files, exist_ok=True) if not os.path.isdir(self.download_files): raise ValueError(f"Path: {self.download_files} is not a directory.") self.output_dir = str(self.download_files) else: self.output_dir = DEFAULT_TEMP_DIR if src.startswith("http://") or src.startswith("https://"): _src = src if src.endswith("/") else src + "/" else: _src = self._space_name_to_src(src) if _src is None: raise ValueError( f"Could not find Space: {src}. If it is a private Space, please provide an hf_token." ) self.space_id = src self.src = _src state = self._get_space_state() if state == SpaceStage.BUILDING: if self.verbose: print("Space is still building. Please wait...") while self._get_space_state() == SpaceStage.BUILDING: time.sleep(2) # so we don't get rate limited by the API pass if state in utils.INVALID_RUNTIME: raise ValueError( f"The current space is in the invalid state: {state}. " "Please contact the owner to fix this." ) if self.verbose: print(f"Loaded as API: {self.src} ✔") self.httpx_kwargs = {} if httpx_kwargs is None else httpx_kwargs if auth is not None: self._login(auth) self.config = self._get_config() self.protocol: Literal["ws", "sse", "sse_v1", "sse_v2", "sse_v2.1"] = ( self.config.get("protocol", "ws") ) api_prefix: str = self.config.get("api_prefix", "") self.api_prefix = api_prefix.lstrip("/") + "/" self.src_prefixed = ( urllib.parse.urljoin(self.src, self.api_prefix).rstrip("/") + "/" ) self.api_url = urllib.parse.urljoin(self.src_prefixed, utils.API_URL) self.sse_url = urllib.parse.urljoin( self.src_prefixed, utils.SSE_URL_V0 if self.protocol == "sse" else utils.SSE_URL, ) self.heartbeat_url = urllib.parse.urljoin( self.src_prefixed, utils.HEARTBEAT_URL ) self.sse_data_url = urllib.parse.urljoin( self.src_prefixed, utils.SSE_DATA_URL_V0 if self.protocol == "sse" else utils.SSE_DATA_URL, ) self.ws_url = urllib.parse.urljoin( self.src_prefixed.replace("http", "ws", 1), utils.WS_URL ) self.upload_url = urllib.parse.urljoin(self.src_prefixed, utils.UPLOAD_URL) self.reset_url = urllib.parse.urljoin(self.src_prefixed, utils.RESET_URL) self.app_version = version.parse(self.config.get("version", "2.0")) self._info = self._get_api_info() self.session_hash = str(uuid.uuid4()) endpoint_class = ( Endpoint if self.protocol.startswith("sse") else EndpointV3Compatibility ) self.endpoints = { dependency.get("id", fn_index): endpoint_class( self, dependency.get("id", fn_index), dependency, self.protocol ) for fn_index, dependency in enumerate(self.config["dependencies"]) } # Create a pool of threads to handle the requests self.executor = concurrent.futures.ThreadPoolExecutor(max_workers=max_workers) self.analytics_enabled = ( analytics_enabled or os.getenv("GRADIO_ANALYTICS_ENABLED", "True") == "True" ) if self.analytics_enabled: threading.Thread(target=self._telemetry_thread, daemon=True).start() self._refresh_heartbeat = threading.Event() self._kill_heartbeat = threading.Event() self.heartbeat = threading.Thread(target=self._stream_heartbeat, daemon=True) self.heartbeat.start() self.stream_open = False self.streaming_future: Future | None = None self.pending_messages_per_event: dict[str, list[Message | None]] = {} self.pending_event_ids: set[str] = set() def close(self): self._kill_heartbeat.set() self.heartbeat.join(timeout=1) def _stream_heartbeat(self): while True: url = self.heartbeat_url.format(session_hash=self.session_hash) try: httpx_kwargs = self.httpx_kwargs.copy() httpx_kwargs.setdefault("timeout", 20) with httpx.stream( "GET", url, headers=self.headers, cookies=self.cookies, verify=self.ssl_verify, **httpx_kwargs, ) as response: for _ in response.iter_lines(): if self._refresh_heartbeat.is_set(): self._refresh_heartbeat.clear() break if self._kill_heartbeat.is_set(): return except httpx.TransportError: return def stream_messages( self, protocol: Literal["sse_v1", "sse_v2", "sse_v2.1", "sse_v3"] ) -> None: try: httpx_kwargs = self.httpx_kwargs.copy() httpx_kwargs.setdefault("timeout", httpx.Timeout(timeout=None)) with httpx.Client( verify=self.ssl_verify, **httpx_kwargs, ) as client: with client.stream( "GET", self.sse_url, params={"session_hash": self.session_hash}, headers=self.headers, cookies=self.cookies, ) as response: buffer = b"" for chunk in response.iter_bytes(): buffer += chunk while b"\n\n" in buffer: line, buffer = buffer.split(b"\n\n", 1) line = line.decode("utf-8").rstrip("\n") if not len(line): continue if line.startswith("data:"): resp = json.loads(line[5:]) if resp["msg"] == ServerMessage.heartbeat: continue elif ( resp.get("message", "") == ServerMessage.server_stopped ): for ( pending_messages ) in self.pending_messages_per_event.values(): pending_messages.append(resp) return elif resp["msg"] == ServerMessage.close_stream: self.stream_open = False return event_id = resp["event_id"] if event_id not in self.pending_messages_per_event: self.pending_messages_per_event[event_id] = [] self.pending_messages_per_event[event_id].append(resp) if resp["msg"] == ServerMessage.process_completed: self.pending_event_ids.remove(event_id) if ( len(self.pending_event_ids) == 0 and protocol != "sse_v3" ): self.stream_open = False return else: raise ValueError(f"Unexpected SSE line: '{line}'") except BaseException as e: # If the job is cancelled the stream will close so we # should not raise this httpx exception that comes from the # stream abruply closing if isinstance(e, httpx.RemoteProtocolError): return import traceback traceback.print_exc() raise e def send_data(self, data, hash_data, protocol): headers = self.add_zero_gpu_headers(self.headers) req = httpx.post( self.sse_data_url, json={**data, **hash_data}, headers=headers, cookies=self.cookies, verify=self.ssl_verify, **self.httpx_kwargs, ) if req.status_code == 503: raise QueueError("Queue is full! Please try again.") req.raise_for_status() resp = req.json() event_id = resp["event_id"] if not self.stream_open: self.stream_open = True def open_stream(): return self.stream_messages(protocol) def close_stream(_): self.stream_open = False for _, pending_messages in self.pending_messages_per_event.items(): pending_messages.append(None) if self.streaming_future is None or self.streaming_future.done(): self.streaming_future = self.executor.submit(open_stream) self.streaming_future.add_done_callback(close_stream) return event_id @classmethod def duplicate( cls, from_id: str, to_id: str | None = None, hf_token: str | Literal[False] | None = False, private: bool = True, hardware: Literal[ "cpu-basic", "cpu-upgrade", "t4-small", "t4-medium", "a10g-small", "a10g-large", "a100-large", ] | SpaceHardware | None = None, secrets: dict[str, str] | None = None, sleep_timeout: int = 5, max_workers: int = 40, verbose: bool = True, ): """ Duplicates a Hugging Face Space under your account and returns a Client object for the new Space. No duplication is created if the Space already exists in your account (to override this, provide a new name for the new Space using `to_id`). To use this method, you must provide an `hf_token` or be logged in via the Hugging Face Hub CLI. The new Space will be private by default and use the same hardware as the original Space. This can be changed by using the `private` and `hardware` parameters. For hardware upgrades (beyond the basic CPU tier), you may be required to provide billing information on Hugging Face: https://huggingface.co/settings/billing Parameters: from_id: The name of the Hugging Face Space to duplicate in the format "{username}/{space_id}", e.g. "gradio/whisper". to_id: The name of the new Hugging Face Space to create, e.g. "abidlabs/whisper-duplicate". If not provided, the new Space will be named "{your_HF_username}/{space_id}". hf_token: optional Hugging Face token to use to duplicating private Spaces. By default, no token is sent to the server. Set `hf_token=None` to use the locally saved token if there is one. Find your tokens here: https://huggingface.co/settings/tokens. private: Whether the new Space should be private (True) or public (False). Defaults to True. hardware: The hardware tier to use for the new Space. Defaults to the same hardware tier as the original Space. Options include "cpu-basic", "cpu-upgrade", "t4-small", "t4-medium", "a10g-small", "a10g-large", "a100-large", subject to availability. secrets: A dictionary of (secret key, secret value) to pass to the new Space. Defaults to None. Secrets are only used when the Space is duplicated for the first time, and are not updated if the duplicated Space already exists. sleep_timeout: The number of minutes after which the duplicate Space will be puased if no requests are made to it (to minimize billing charges). Defaults to 5 minutes. max_workers: The maximum number of thread workers that can be used to make requests to the remote Gradio app simultaneously. verbose: Whether the client should print statements to the console. Example: import os from gradio_client import Client HF_TOKEN = os.environ.get("HF_TOKEN") client = Client.duplicate("abidlabs/whisper", hf_token=HF_TOKEN) client.predict("audio_sample.wav") >> "This is a test of the whisper speech recognition model." """ try: original_info = huggingface_hub.get_space_runtime(from_id, token=hf_token) except RepositoryNotFoundError as rnfe: raise ValueError( f"Could not find Space: {from_id}. If it is a private Space, please provide an `hf_token`." ) from rnfe if to_id: if "/" in to_id: to_id = to_id.split("/")[1] space_id = huggingface_hub.get_full_repo_name(to_id, token=hf_token) else: space_id = huggingface_hub.get_full_repo_name( from_id.split("/")[1], token=hf_token ) try: huggingface_hub.get_space_runtime(space_id, token=hf_token) if verbose: print( f"Using your existing Space: {utils.SPACE_URL.format(space_id)} 🤗" ) if secrets is not None: warnings.warn( "Secrets are only used when the Space is duplicated for the first time, and are not updated if the duplicated Space already exists." ) except RepositoryNotFoundError: if verbose: print(f"Creating a duplicate of {from_id} for your own use... 🤗") huggingface_hub.duplicate_space( from_id=from_id, to_id=space_id, token=hf_token, exist_ok=True, private=private, ) if secrets is not None: for key, value in secrets.items(): huggingface_hub.add_space_secret( space_id, key, value, token=hf_token ) if verbose: print(f"Created new Space: {utils.SPACE_URL.format(space_id)}") current_info = huggingface_hub.get_space_runtime(space_id, token=hf_token) current_hardware = ( current_info.hardware or huggingface_hub.SpaceHardware.CPU_BASIC ) hardware = hardware or original_info.hardware if current_hardware != hardware: huggingface_hub.request_space_hardware(space_id, hardware, token=hf_token) # type: ignore print( f"-------\nNOTE: this Space uses upgraded hardware: {hardware}... see billing info at https://huggingface.co/settings/billing\n-------" ) # Setting a timeout only works if the hardware is not basic # so set it here after the hardware has been requested if hardware != huggingface_hub.SpaceHardware.CPU_BASIC: utils.set_space_timeout( space_id, hf_token=hf_token, timeout_in_seconds=sleep_timeout * 60 ) if verbose: print("") client = cls( space_id, hf_token=hf_token, max_workers=max_workers, verbose=verbose ) return client def _get_space_state(self): if not self.space_id: return None info = huggingface_hub.get_space_runtime(self.space_id, token=self.hf_token) return info.stage def predict( self, *args, api_name: str | None = None, fn_index: int | None = None, **kwargs, ) -> Any: """ Calls the Gradio API and returns the result (this is a blocking call). Arguments can be provided as positional arguments or as keyword arguments (latter is recommended). Parameters: args: The positional arguments to pass to the remote API endpoint. The order of the arguments must match the order of the inputs in the Gradio app. api_name: The name of the API endpoint to call starting with a leading slash, e.g. "/predict". Does not need to be provided if the Gradio app has only one named API endpoint. fn_index: As an alternative to api_name, this parameter takes the index of the API endpoint to call, e.g. 0. Both api_name and fn_index can be provided, but if they conflict, api_name will take precedence. kwargs: The keyword arguments to pass to the remote API endpoint. Returns: The result of the API call. Will be a Tuple if the API has multiple outputs. Example: from gradio_client import Client client = Client(src="gradio/calculator") client.predict(5, "add", 4, api_name="/predict") >> 9.0 """ self._infer_fn_index(api_name, fn_index) return self.submit( *args, api_name=api_name, fn_index=fn_index, **kwargs ).result() def new_helper(self, fn_index: int) -> Communicator: return Communicator( Lock(), JobStatus(), self.endpoints[fn_index].process_predictions, self.reset_url, ) def submit( self, *args, api_name: str | None = None, fn_index: int | None = None, result_callbacks: Callable | list[Callable] | None = None, **kwargs, ) -> Job: """ Creates and returns a Job object which calls the Gradio API in a background thread. The job can be used to retrieve the status and result of the remote API call. Arguments can be provided as positional arguments or as keyword arguments (latter is recommended). Parameters: args: The arguments to pass to the remote API. The order of the arguments must match the order of the inputs in the Gradio app. api_name: The name of the API endpoint to call starting with a leading slash, e.g. "/predict". Does not need to be provided if the Gradio app has only one named API endpoint. fn_index: As an alternative to api_name, this parameter takes the index of the API endpoint to call, e.g. 0. Both api_name and fn_index can be provided, but if they conflict, api_name will take precedence. result_callbacks: A callback function, or list of callback functions, to be called when the result is ready. If a list of functions is provided, they will be called in order. The return values from the remote API are provided as separate parameters into the callback. If None, no callback will be called. kwargs: The keyword arguments to pass to the remote API endpoint. Returns: A Job object that can be used to retrieve the status and result of the remote API call. Example: from gradio_client import Client client = Client(src="gradio/calculator") job = client.submit(5, "add", 4, api_name="/predict") job.status() >> job.result() # blocking call >> 9.0 """ inferred_fn_index = self._infer_fn_index(api_name, fn_index) endpoint = self.endpoints[inferred_fn_index] if isinstance(endpoint, Endpoint): args = utils.construct_args(endpoint.parameters_info, args, kwargs) helper = None if endpoint.protocol in ( "ws", "sse", "sse_v1", "sse_v2", "sse_v2.1", "sse_v3", ): helper = self.new_helper(inferred_fn_index) end_to_end_fn = endpoint.make_end_to_end_fn(helper) future = self.executor.submit(end_to_end_fn, *args) cancel_fn = endpoint.make_cancel(helper) job = Job( future, communicator=helper, verbose=self.verbose, space_id=self.space_id, _cancel_fn=cancel_fn, ) if result_callbacks: if isinstance(result_callbacks, Callable): result_callbacks = [result_callbacks] def create_fn(callback) -> Callable: def fn(future): if isinstance(future.result(), tuple): callback(*future.result()) else: callback(future.result()) return fn for callback in result_callbacks: job.add_done_callback(create_fn(callback)) return job def _get_api_info(self): api_info_url = urllib.parse.urljoin(self.src_prefixed, utils.RAW_API_INFO_URL) if self.app_version > version.Version("3.36.1"): r = httpx.get( api_info_url, headers=self.headers, cookies=self.cookies, verify=self.ssl_verify, **self.httpx_kwargs, ) if r.is_success: info = r.json() else: raise ValueError(f"Could not fetch api info for {self.src}: {r.text}") else: fetch = httpx.post( utils.SPACE_FETCHER_URL, json={ "config": json.dumps(self.config), "serialize": False, }, **self.httpx_kwargs, ) if fetch.is_success: info = fetch.json()["api"] else: raise ValueError( f"Could not fetch api info for {self.src}: {fetch.text}" ) info["named_endpoints"] = { a: e for a, e in info["named_endpoints"].items() if e.pop("show_api", True) } info["unnamed_endpoints"] = { a: e for a, e in info["unnamed_endpoints"].items() if e.pop("show_api", True) } return info def view_api( self, all_endpoints: bool | None = None, print_info: bool = True, return_format: Literal["dict", "str"] | None = None, ) -> dict | str | None: """ Prints the usage info for the API. If the Gradio app has multiple API endpoints, the usage info for each endpoint will be printed separately. If return_format="dict" the info is returned in dictionary format, as shown in the example below. Parameters: all_endpoints: If True, prints information for both named and unnamed endpoints in the Gradio app. If False, will only print info about named endpoints. If None (default), will print info about named endpoints, unless there aren't any -- in which it will print info about unnamed endpoints. print_info: If True, prints the usage info to the console. If False, does not print the usage info. return_format: If None, nothing is returned. If "str", returns the same string that would be printed to the console. If "dict", returns the usage info as a dictionary that can be programmatically parsed, and *all endpoints are returned in the dictionary* regardless of the value of `all_endpoints`. The format of the dictionary is in the docstring of this method. Example: from gradio_client import Client client = Client(src="gradio/calculator") client.view_api(return_format="dict") >> { 'named_endpoints': { '/predict': { 'parameters': [ { 'label': 'num1', 'python_type': 'int | float', 'type_description': 'numeric value', 'component': 'Number', 'example_input': '5' }, { 'label': 'operation', 'python_type': 'str', 'type_description': 'string value', 'component': 'Radio', 'example_input': 'add' }, { 'label': 'num2', 'python_type': 'int | float', 'type_description': 'numeric value', 'component': 'Number', 'example_input': '5' }, ], 'returns': [ { 'label': 'output', 'python_type': 'int | float', 'type_description': 'numeric value', 'component': 'Number', }, ] }, '/flag': { 'parameters': [ ... ], 'returns': [ ... ] } } 'unnamed_endpoints': { 2: { 'parameters': [ ... ], 'returns': [ ... ] } } } } """ num_named_endpoints = len(self._info["named_endpoints"]) num_unnamed_endpoints = len(self._info["unnamed_endpoints"]) if num_named_endpoints == 0 and all_endpoints is None: all_endpoints = True human_info = "Client.predict() Usage Info\n---------------------------\n" human_info += f"Named API endpoints: {num_named_endpoints}\n" for api_name, endpoint_info in self._info["named_endpoints"].items(): human_info += self._render_endpoints_info(api_name, endpoint_info) if all_endpoints: human_info += f"\nUnnamed API endpoints: {num_unnamed_endpoints}\n" for fn_index, endpoint_info in self._info["unnamed_endpoints"].items(): # When loading from json, the fn_indices are read as strings # because json keys can only be strings human_info += self._render_endpoints_info(int(fn_index), endpoint_info) elif num_unnamed_endpoints > 0: human_info += f"\nUnnamed API endpoints: {num_unnamed_endpoints}, to view, run Client.view_api(all_endpoints=True)\n" if print_info: print(human_info) if return_format == "str": return human_info elif return_format == "dict": return self._info def reset_session(self) -> None: self.session_hash = str(uuid.uuid4()) self._refresh_heartbeat.set() def add_zero_gpu_headers(self, headers: dict[str, str]) -> dict[str, str]: """ Adds the x-ip-token header to the headers dictionary to pass it to a Zero-GPU Space. This allows a user's ZeroGPU quota to be tracked and used by the underlying Space. For the x-ip-token header to be present, this method needs to be called when a Gradio app's LocalContext is defined. i.e. this method cannot be called when the Gradio Client is instantiated, but must be called from inside a Gradio app's prediction function. """ if not self.space_id: return headers try: from gradio.context import LocalContext except ( ImportError ): # this is not running within a Gradio app as Gradio is not installed return headers request = LocalContext.request.get() if request and hasattr(request, "headers") and "x-ip-token" in request.headers: headers["x-ip-token"] = request.headers["x-ip-token"] return headers def _render_endpoints_info( self, name_or_index: str | int, endpoints_info: dict[str, list[ParameterInfo]], ) -> str: parameter_info = endpoints_info["parameters"] parameter_names = [ p.get("parameter_name") or p["label"] for p in parameter_info ] parameter_names = [utils.sanitize_parameter_names(p) for p in parameter_names] rendered_parameters = ", ".join(parameter_names) if rendered_parameters: rendered_parameters = rendered_parameters + ", " return_values = [p["label"] for p in endpoints_info["returns"]] return_values = [utils.sanitize_parameter_names(r) for r in return_values] rendered_return_values = ", ".join(return_values) if len(return_values) > 1: rendered_return_values = f"({rendered_return_values})" if isinstance(name_or_index, str): final_param = f'api_name="{name_or_index}"' elif isinstance(name_or_index, int): final_param = f"fn_index={name_or_index}" else: raise ValueError("name_or_index must be a string or integer") human_info = f"\n - predict({rendered_parameters}{final_param}) -> {rendered_return_values}\n" human_info += " Parameters:\n" if parameter_info: for info in parameter_info: desc = ( f" ({info['python_type']['description']})" if info["python_type"].get("description") else "" ) default_value = info.get("parameter_default") default_value = utils.traverse( default_value, lambda x: f'handle_file("{x["url"]}")', utils.is_file_obj_with_meta, ) default_info = ( "(required)" if not info.get("parameter_has_default", False) else f"(not required, defaults to: {default_value})" ) type_ = info["python_type"]["type"] if info.get("parameter_has_default", False) and default_value is None: type_ += " | None" human_info += f" - [{info['component']}] {utils.sanitize_parameter_names(info.get('parameter_name') or info['label'])}: {type_} {default_info} {desc} \n" else: human_info += " - None\n" human_info += " Returns:\n" if endpoints_info["returns"]: for info in endpoints_info["returns"]: desc = ( f" ({info['python_type']['description']})" if info["python_type"].get("description") else "" ) type_ = info["python_type"]["type"] human_info += f" - [{info['component']}] {utils.sanitize_parameter_names(info['label'])}: {type_}{desc} \n" else: human_info += " - None\n" return human_info def __repr__(self): return self.view_api(print_info=False, return_format="str") def __str__(self): return self.view_api(print_info=False, return_format="str") def _telemetry_thread(self) -> None: # Disable telemetry by setting the env variable HF_HUB_DISABLE_TELEMETRY=1 data = { "src": self.src, } try: send_telemetry( topic="py_client/initiated", library_name="gradio_client", library_version=utils.__version__, user_agent=data, ) except Exception: pass def _infer_fn_index(self, api_name: str | None, fn_index: int | None) -> int: inferred_fn_index = None if api_name is not None: for i, d in enumerate(self.config["dependencies"]): config_api_name = d.get("api_name") if config_api_name is None or config_api_name is False: continue if "/" + config_api_name == api_name: inferred_fn_index = d.get("id", i) break else: error_message = f"Cannot find a function with `api_name`: {api_name}." if not api_name.startswith("/"): error_message += " Did you mean to use a leading slash?" raise ValueError(error_message) elif fn_index is not None: inferred_fn_index = fn_index if ( inferred_fn_index not in self.endpoints or not self.endpoints[inferred_fn_index].is_valid ): raise ValueError(f"Invalid function index: {fn_index}.") else: valid_endpoints = [ e for e in self.endpoints.values() if e.is_valid and e.api_name is not None and e.backend_fn is not None and e.show_api ] if len(valid_endpoints) == 1: inferred_fn_index = valid_endpoints[0].fn_index else: raise ValueError( "This Gradio app might have multiple endpoints. Please specify an `api_name` or `fn_index`" ) return inferred_fn_index def __del__(self): if hasattr(self, "executor"): self.executor.shutdown(wait=True) def _space_name_to_src(self, space) -> str | None: return huggingface_hub.space_info(space, token=self.hf_token).host # type: ignore def _login(self, auth: tuple[str, str]): resp = httpx.post( urllib.parse.urljoin(self.src, utils.LOGIN_URL), data={"username": auth[0], "password": auth[1]}, verify=self.ssl_verify, **self.httpx_kwargs, ) if not resp.is_success: if resp.status_code == 401: raise AuthenticationError( f"Could not login to {self.src}. Invalid credentials." ) else: raise ValueError(f"Could not login to {self.src}.") self.cookies = { name: value for name, value in resp.cookies.items() if value is not None } def _get_config(self) -> dict: r = httpx.get( urllib.parse.urljoin(self.src, utils.CONFIG_URL), headers=self.headers, cookies=self.cookies, verify=self.ssl_verify, **self.httpx_kwargs, ) if r.is_success: return r.json() elif r.status_code == 401: raise AuthenticationError( f"Could not load {self.src} as credentials were not provided. Please login." ) elif r.status_code == 429: raise utils.TooManyRequestsError( "Too many requests to the API, please try again later." ) from None else: # to support older versions of Gradio r = httpx.get( self.src, headers=self.headers, cookies=self.cookies, verify=self.ssl_verify, **self.httpx_kwargs, ) if not r.is_success: raise ValueError(f"Could not fetch config for {self.src}") # some basic regex to extract the config result = re.search(r"window.gradio_config = (.*?);[\s]*", r.text) try: config = json.loads(result.group(1)) # type: ignore except AttributeError as ae: raise ValueError( f"Could not get Gradio config from: {self.src}" ) from ae if "allow_flagging" in config: raise ValueError( "Gradio 2.x is not supported by this client. Please upgrade your Gradio app to Gradio 3.x or higher." ) return config def deploy_discord( self, discord_bot_token: str | None = None, api_names: list[str | tuple[str, str]] | None = None, to_id: str | None = None, hf_token: str | Literal[False] | None = False, private: bool = False, ): """ Deploy the upstream app as a discord bot. Currently only supports gr.ChatInterface. Parameters: discord_bot_token: This is the "password" needed to be able to launch the bot. Users can get a token by creating a bot app on the discord website. If run the method without specifying a token, the space will explain how to get one. See here: https://huggingface.co/spaces/freddyaboulton/test-discord-bot-v1. api_names: The api_names of the app to turn into bot commands. This parameter currently has no effect as ChatInterface only has one api_name ('/chat'). to_id: The name of the space hosting the discord bot. If None, the name will be gradio-discord-bot-{random-substring} hf_token: HF api token with write priviledges in order to upload the files to HF space. Can be ommitted if logged in via the HuggingFace CLI, unless the upstream space is private. Obtain from: https://huggingface.co/settings/token private: Whether the space hosting the discord bot is private. The visibility of the discord bot itself is set via the discord website. See https://huggingface.co/spaces/freddyaboulton/test-discord-bot-v1 """ warnings.warn( "This method is deprecated and may be removed in the future. Please see the documentation on how to create a discord bot with Gradio: https://www.gradio.app/guides/creating-a-discord-bot-from-a-gradio-app" ) if self.config["mode"] == "chat_interface" and not api_names: api_names = [("chat", "chat")] valid_list = isinstance(api_names, list) and ( isinstance(n, str) or ( isinstance(n, tuple) and isinstance(n[0], str) and isinstance(n[1], str) ) for n in api_names ) if api_names is None or not valid_list: raise ValueError( f"Each entry in api_names must be either a string or a tuple of strings. Received {api_names}" ) if len(api_names) != 1: raise ValueError("Currently only one api_name can be deployed to discord.") for i, name in enumerate(api_names): if isinstance(name, str): api_names[i] = (name, name) fn = next( ( ep for ep in self.endpoints.values() if ep.api_name == f"/{api_names[0][0]}" ), None, ) if not fn: raise ValueError( f"api_name {api_names[0][0]} not present in {self.space_id or self.src}" ) inputs = [inp for inp in fn.input_component_types if not inp.skip] outputs = [inp for inp in fn.input_component_types if not inp.skip] if not inputs == ["textbox"] and outputs == ["textbox"]: raise ValueError( "Currently only api_names with a single textbox as input and output are supported. " f"Received {inputs} and {outputs}" ) is_private = False if self.space_id: is_private = huggingface_hub.space_info(self.space_id).private if is_private and not hf_token: raise ValueError( f"Since {self.space_id} is private, you must explicitly pass in hf_token " "so that it can be added as a secret in the discord bot space." ) if to_id: if "/" in to_id: to_id = to_id.split("/")[1] space_id = huggingface_hub.get_full_repo_name(to_id, token=hf_token) else: if self.space_id: space_id = f"{self.space_id.split('/')[1]}-gradio-discord-bot" else: space_id = f"gradio-discord-bot-{secrets.token_hex(4)}" space_id = huggingface_hub.get_full_repo_name(space_id, token=hf_token) api = huggingface_hub.HfApi() try: huggingface_hub.space_info(space_id) first_upload = False except huggingface_hub.utils.RepositoryNotFoundError: first_upload = True huggingface_hub.create_repo( space_id, repo_type="space", space_sdk="gradio", token=hf_token, exist_ok=True, private=private, ) if first_upload: huggingface_hub.metadata_update( repo_id=space_id, repo_type="space", metadata={"tags": ["gradio-discord-bot"]}, ) with open( str(Path(__file__).parent / "templates" / "discord_chat.py"), encoding="utf-8", ) as f: app = f.read() app = app.replace("<>", self.src) app = app.replace("<>", api_names[0][0]) app = app.replace("<>", api_names[0][1]) with tempfile.NamedTemporaryFile( mode="w", delete=False, encoding="utf-8" ) as app_file: with tempfile.NamedTemporaryFile(mode="w", delete=False) as requirements: app_file.write(app) requirements.write("\n".join(["discord.py==2.3.1"])) operations = [ CommitOperationAdd(path_in_repo="app.py", path_or_fileobj=app_file.name), CommitOperationAdd( path_in_repo="requirements.txt", path_or_fileobj=requirements.name ), ] api.create_commit( repo_id=space_id, commit_message="Deploy Discord Bot", repo_type="space", operations=operations, token=hf_token, ) if discord_bot_token: huggingface_hub.add_space_secret( space_id, "DISCORD_TOKEN", discord_bot_token, token=hf_token ) if is_private: huggingface_hub.add_space_secret( space_id, "HF_TOKEN", hf_token, # type: ignore token=hf_token, ) url = f"https://huggingface.co/spaces/{space_id}" print(f"See your discord bot here! {url}") return url @dataclass class ComponentApiType: skip: bool value_is_file: bool is_state: bool @dataclass class ReplaceMe: index: int class Endpoint: """Helper class for storing all the information about a single API endpoint.""" def __init__( self, client: Client, fn_index: int, dependency: dict, protocol: str = "sse_v1" ): self.client: Client = client self.fn_index = fn_index self.dependency = dependency api_name = dependency.get("api_name") self.api_name: str | Literal[False] | None = ( "/" + api_name if isinstance(api_name, str) else api_name ) self._info = self.client._info self.protocol = protocol self.input_component_types = [ self._get_component_type(id_) for id_ in dependency["inputs"] ] self.output_component_types = [ self._get_component_type(id_) for id_ in dependency["outputs"] ] self.parameters_info = self._get_parameters_info() self.root_url = self.client.src_prefixed # Disallow hitting endpoints that the Gradio app has disabled self.is_valid = self.api_name is not False self.backend_fn = dependency.get("backend_fn") self.show_api = dependency.get("show_api") def _get_component_type(self, component_id: int): component = next( i for i in self.client.config["components"] if i["id"] == component_id ) skip_api = component.get("skip_api", component["type"] in utils.SKIP_COMPONENTS) return ComponentApiType( skip_api, self.value_is_file(component), component["type"] == "state", ) def _get_parameters_info(self) -> list[ParameterInfo] | None: if self.api_name in self._info["named_endpoints"]: return self._info["named_endpoints"][self.api_name]["parameters"] return None @staticmethod def value_is_file(component: dict) -> bool: # This is still hacky as it does not tell us which part of the payload is a file. # If a component has a complex payload, part of which is a file, this will simply # return True, which means that all parts of the payload will be uploaded as files # if they are valid file paths. We will deprecate this 1.0. if "api_info" not in component: return False return utils.value_is_file(component["api_info"]) def __repr__(self): return f"Endpoint src: {self.client.src}, api_name: {self.api_name}, fn_index: {self.fn_index}" def __str__(self): return self.__repr__() def make_end_to_end_fn(self, helper: Communicator | None = None): _predict = self.make_predict(helper) def _inner(*data): if not self.is_valid: raise utils.InvalidAPIEndpointError() if self.client._skip_components: data = self.insert_empty_state(*data) data = self.process_input_files(*data) predictions = _predict(*data) predictions = self.process_predictions(*predictions) # Append final output only if not already present # for consistency between generators and not generators if helper: with helper.lock: if not helper.job.outputs: helper.job.outputs.append(predictions) return predictions return _inner def make_cancel( self, helper: Communicator | None, ): if helper is None: return if self.client.app_version > version.Version("4.29.0"): url = urllib.parse.urljoin(self.client.src_prefixed, utils.CANCEL_URL) # The event_id won't be set on the helper until later # so need to create the data in a function that's run at cancel time def post_data(): return { "fn_index": self.fn_index, "session_hash": self.client.session_hash, "event_id": helper.event_id, } cancel_msg = None cancellable = True else: candidates: list[tuple[int, list[int]]] = [] for i, dep in enumerate(self.client.config["dependencies"]): if self.fn_index in dep["cancels"]: candidates.append( (i, [d for d in dep["cancels"] if d != self.fn_index]) ) fn_index, other_cancelled = ( min(candidates, key=lambda x: len(x[1])) if candidates else (None, None) ) cancellable = fn_index is not None cancel_msg = None if cancellable and other_cancelled: other_api_names = [ "/" + self.client.config["dependencies"][i].get("api_name") for i in other_cancelled ] cancel_msg = ( f"Cancelled this job will also cancel any jobs for {', '.join(other_api_names)} " "that are currently running." ) elif not cancellable: cancel_msg = ( "Cancelling this job will not stop the server from running. " "To fix this, an event must be added to the upstream app that explicitly cancels this one or " "the upstream app must be running Gradio 4.29.0 and greater." ) def post_data(): return { "data": [], "fn_index": fn_index, "session_hash": self.client.session_hash, } url = self.client.api_url def _cancel(): if cancel_msg: warnings.warn(cancel_msg) if cancellable: httpx.post( url, json=post_data(), headers=self.client.headers, cookies=self.client.cookies, verify=self.client.ssl_verify, **self.client.httpx_kwargs, ) return _cancel def make_predict(self, helper: Communicator | None = None): def _predict(*data) -> tuple: data = { "data": data, "fn_index": self.fn_index, "session_hash": self.client.session_hash, } hash_data = { "fn_index": self.fn_index, "session_hash": self.client.session_hash, } if self.protocol == "sse": result = self._sse_fn_v0(data, hash_data, helper) # type: ignore elif self.protocol in ("sse_v1", "sse_v2", "sse_v2.1", "sse_v3"): event_id = self.client.send_data(data, hash_data, self.protocol) self.client.pending_event_ids.add(event_id) self.client.pending_messages_per_event[event_id] = [] helper.event_id = event_id result = self._sse_fn_v1plus(helper, event_id, self.protocol) else: raise ValueError(f"Unsupported protocol: {self.protocol}") if "error" in result: if result["error"] is None: raise AppError( "The upstream Gradio app has raised an exception but has not enabled " "verbose error reporting. To enable, set show_error=True in launch()." ) else: message = result.pop("error") raise AppError(message=message, **result) try: output = result["data"] except KeyError as ke: is_public_space = ( self.client.space_id and not huggingface_hub.space_info(self.client.space_id).private ) if "error" in result and "429" in result["error"] and is_public_space: raise utils.TooManyRequestsError( f"Too many requests to the API, please try again later. To avoid being rate-limited, " f"please duplicate the Space using Client.duplicate({self.client.space_id}) " f"and pass in your Hugging Face token." ) from None elif "error" in result: raise ValueError(result["error"]) from None raise KeyError( f"Could not find 'data' key in response. Response received: {result}" ) from ke return tuple(output) return _predict def insert_empty_state(self, *data) -> tuple: data = list(data) for i, input_component_type in enumerate(self.input_component_types): if input_component_type.is_state: data.insert(i, None) return tuple(data) def process_input_files(self, *data) -> tuple: data_ = [] for i, d in enumerate(data): d = utils.traverse( d, partial(self._upload_file, data_index=i), utils.is_file_obj_with_meta, ) data_.append(d) return tuple(data_) def process_predictions(self, *predictions): # If self.download_file is True, we assume that that the user is using the Client directly (as opposed # within gr.load) and therefore, download any files generated by the server and skip values for # components that the user likely does not want to see (e.g. gr.State, gr.Tab). if self.client.download_files: predictions = self.download_files(*predictions) if self.client._skip_components: predictions = self.remove_skipped_components(*predictions) predictions = self.reduce_singleton_output(*predictions) return predictions def download_files(self, *data) -> tuple: data_ = list(data) if self.client.protocol == "sse_v2.1": data_ = utils.traverse( data_, self._download_file, utils.is_file_obj_with_meta ) else: data_ = utils.traverse(data_, self._download_file, utils.is_file_obj) return tuple(data_) def remove_skipped_components(self, *data) -> tuple: """""" data = [ d for d, oct in zip(data, self.output_component_types, strict=False) if not oct.skip ] return tuple(data) def reduce_singleton_output(self, *data) -> Any: if self.client._skip_components: effective_output_components = [ o for o in self.output_component_types if not o.skip ] else: effective_output_components = self.output_component_types if len(effective_output_components) == 1: return data[0] else: return data def _upload_file(self, f: dict, data_index: int) -> dict[str, str]: file_path = f["path"] orig_name = Path(file_path) if not utils.is_http_url_like(file_path): component_id = self.dependency["inputs"][data_index] component_config = next( ( c for c in self.client.config["components"] if c["id"] == component_id ), {}, ) max_file_size = self.client.config.get("max_file_size", None) max_file_size = math.inf if max_file_size is None else max_file_size if os.path.getsize(file_path) > max_file_size: raise ValueError( f"File {file_path} exceeds the maximum file size of {max_file_size} bytes " f"set in {component_config.get('label', '') + ''} component." ) with open(file_path, "rb") as f_: files = [("files", (orig_name.name, f_))] r = httpx.post( self.client.upload_url, headers=self.client.headers, cookies=self.client.cookies, verify=self.client.ssl_verify, files=files, **self.client.httpx_kwargs, ) r.raise_for_status() result = r.json() file_path = result[0] # Only return orig_name if has a suffix because components # use the suffix of the original name to determine format to save it to in cache. return { "path": file_path, "orig_name": utils.strip_invalid_filename_characters(orig_name.name), "meta": {"_type": "gradio.FileData"}, } def _download_file(self, x: dict) -> str: url_path = self.root_url + "file=" + x["path"] if self.client.output_dir is not None: os.makedirs(self.client.output_dir, exist_ok=True) sha = hashlib.sha256() temp_dir = Path(tempfile.gettempdir()) / secrets.token_hex(20) temp_dir.mkdir(exist_ok=True, parents=True) with httpx.stream( "GET", url_path, headers=self.client.headers, cookies=self.client.cookies, verify=self.client.ssl_verify, follow_redirects=True, **self.client.httpx_kwargs, ) as response: response.raise_for_status() with open(temp_dir / Path(url_path).name, "wb") as f: for chunk in response.iter_bytes(chunk_size=128 * sha.block_size): sha.update(chunk) f.write(chunk) directory = Path(self.client.output_dir) / sha.hexdigest() directory.mkdir(exist_ok=True, parents=True) dest = directory / Path(url_path).name shutil.move(temp_dir / Path(url_path).name, dest) return str(dest.resolve()) def _sse_fn_v0(self, data: dict, hash_data: dict, helper: Communicator): with httpx.Client( timeout=httpx.Timeout(timeout=None), verify=self.client.ssl_verify, **self.client.httpx_kwargs, ) as client: return utils.get_pred_from_sse_v0( client, data, hash_data, helper, self.client.sse_url, self.client.sse_data_url, self.client.headers, self.client.cookies, self.client.ssl_verify, self.client.executor, ) def _sse_fn_v1plus( self, helper: Communicator, event_id: str, protocol: Literal["sse_v1", "sse_v2", "sse_v2.1", "sse_v3"], ): return utils.get_pred_from_sse_v1plus( helper, self.client.headers, self.client.cookies, self.client.pending_messages_per_event, event_id, protocol, self.client.ssl_verify, self.client.executor, ) @document("result", "outputs", "status") class Job(Future): """ A Job is a wrapper over the Future class that represents a prediction call that has been submitted by the Gradio client. This class is not meant to be instantiated directly, but rather is created by the Client.submit() method. A Job object includes methods to get the status of the prediction call, as well to get the outputs of the prediction call. Job objects are also iterable, and can be used in a loop to get the outputs of prediction calls as they become available for generator endpoints. """ def __init__( self, future: Future, communicator: Communicator | None = None, verbose: bool = True, space_id: str | None = None, _cancel_fn: Callable[[], None] | None = None, ): """ Parameters: future: The future object that represents the prediction call, created by the Client.submit() method communicator: The communicator object that is used to communicate between the client and the background thread running the job verbose: Whether to print any status-related messages to the console space_id: The space ID corresponding to the Client object that created this Job object """ self.future = future self.communicator = communicator self._counter = 0 self.verbose = verbose self.space_id = space_id self.cancel_fn = _cancel_fn def __iter__(self) -> Job: return self def __next__(self) -> tuple | Any: if not self.communicator: raise StopIteration() while True: with self.communicator.lock: if len(self.communicator.job.outputs) >= self._counter + 1: o = self.communicator.job.outputs[self._counter] self._counter += 1 return o if ( self.communicator.job.latest_status.code == Status.FINISHED and self._counter >= len(self.communicator.job.outputs) ): raise StopIteration() time.sleep(0.001) def result(self, timeout: float | None = None) -> Any: """ Return the result of the call that the future represents. Raises CancelledError: If the future was cancelled, TimeoutError: If the future didn't finish executing before the given timeout, and Exception: If the call raised then that exception will be raised. Parameters: timeout: The number of seconds to wait for the result if the future isn't done. If None, then there is no limit on the wait time. Returns: The result of the call that the future represents. For generator functions, it will return the final iteration. Example: from gradio_client import Client calculator = Client(src="gradio/calculator") job = calculator.submit("foo", "add", 4, fn_index=0) job.result(timeout=5) >> 9 """ return super().result(timeout=timeout) def outputs(self) -> list[tuple | Any]: """ Returns a list containing the latest outputs from the Job. If the endpoint has multiple output components, the list will contain a tuple of results. Otherwise, it will contain the results without storing them in tuples. For endpoints that are queued, this list will contain the final job output even if that endpoint does not use a generator function. Example: from gradio_client import Client client = Client(src="gradio/count_generator") job = client.submit(3, api_name="/count") while not job.done(): time.sleep(0.1) job.outputs() >> ['0', '1', '2'] """ if not self.communicator: return [] else: with self.communicator.lock: return self.communicator.job.outputs def status(self) -> StatusUpdate: """ Returns the latest status update from the Job in the form of a StatusUpdate object, which contains the following fields: code, rank, queue_size, success, time, eta, and progress_data. progress_data is a list of updates emitted by the gr.Progress() tracker of the event handler. Each element of the list has the following fields: index, length, unit, progress, desc. If the event handler does not have a gr.Progress() tracker, the progress_data field will be None. Example: from gradio_client import Client client = Client(src="gradio/calculator") job = client.submit(5, "add", 4, api_name="/predict") job.status() >> job.status().eta >> 43.241 # seconds """ time = datetime.now() cancelled = False if self.communicator: with self.communicator.lock: cancelled = self.communicator.should_cancel if cancelled: return StatusUpdate( code=Status.CANCELLED, rank=0, queue_size=None, success=False, time=time, eta=None, progress_data=None, ) if self.done(): if not self.future._exception: # type: ignore return StatusUpdate( code=Status.FINISHED, rank=0, queue_size=None, success=True, time=time, eta=None, progress_data=None, ) else: return StatusUpdate( code=Status.FINISHED, rank=0, queue_size=None, success=False, time=time, eta=None, progress_data=None, ) elif not self.communicator: return StatusUpdate( code=Status.PROCESSING, rank=0, queue_size=None, success=None, time=time, eta=None, progress_data=None, ) else: with self.communicator.lock: eta = self.communicator.job.latest_status.eta if self.verbose and self.space_id and eta and eta > 30: print( f"Due to heavy traffic on this app, the prediction will take approximately {int(eta)} seconds." f"For faster predictions without waiting in queue, you may duplicate the space using: Client.duplicate({self.space_id})" ) return self.communicator.job.latest_status def cancel(self) -> bool: """Cancels the job as best as possible. If the app you are connecting to has the gradio queue enabled, the job will be cancelled locally as soon as possible. For apps that do not use the queue, the job cannot be cancelled if it's been sent to the local executor (for the time being). Note: In general, this DOES not stop the process from running in the upstream server except for the following situations: 1. If the job is queued upstream, it will be removed from the queue and the server will not run the job 2. If the job has iterative outputs, the job will finish as soon as the current iteration finishes running 3. If the job has not been picked up by the queue yet, the queue will not pick up the job """ if self.communicator: with self.communicator.lock: self.communicator.should_cancel = True if self.cancel_fn: self.cancel_fn() return True return self.future.cancel() def __getattr__(self, name): """Forwards any properties to the Future class.""" return getattr(self.future, name)