diff options
Diffstat (limited to 'g4f/Provider/needs_auth/OpenaiChat.py')
| -rw-r--r-- | g4f/Provider/needs_auth/OpenaiChat.py | 360 |
1 files changed, 152 insertions, 208 deletions
diff --git a/g4f/Provider/needs_auth/OpenaiChat.py b/g4f/Provider/needs_auth/OpenaiChat.py index f02121e3..13e15f1d 100644 --- a/g4f/Provider/needs_auth/OpenaiChat.py +++ b/g4f/Provider/needs_auth/OpenaiChat.py @@ -1,37 +1,39 @@ from __future__ import annotations +import re import asyncio import uuid import json import base64 import time +import requests from aiohttp import ClientWebSocketResponse from copy import copy try: - import webview - has_webview = True + import nodriver + from nodriver.cdp.network import get_response_body + has_nodriver = True except ImportError: - has_webview = False - + has_nodriver = False try: - from selenium.webdriver.common.by import By - from selenium.webdriver.support.ui import WebDriverWait - from selenium.webdriver.support import expected_conditions as EC + from platformdirs import user_config_dir + has_platformdirs = True except ImportError: - pass + has_platformdirs = False from ..base_provider import AsyncGeneratorProvider, ProviderModelMixin -from ...webdriver import get_browser from ...typing import AsyncResult, Messages, Cookies, ImageType, AsyncIterator -from ...requests import get_args_from_browser, raise_for_status +from ...requests.raise_for_status import raise_for_status from ...requests.aiohttp import StreamSession from ...image import ImageResponse, ImageRequest, to_image, to_bytes, is_accepted_format from ...errors import MissingAuthError, ResponseError from ...providers.conversation import BaseConversation from ..helper import format_cookies -from ..openai.har_file import getArkoseAndAccessToken, NoValidHarFileError +from ..openai.har_file import get_request_config, NoValidHarFileError +from ..openai.har_file import RequestConfig, arkReq, arkose_url, start_url, conversation_url, backend_url, backend_anon_url from ..openai.proofofwork import generate_proof_token +from ..openai.new import get_requirements_token from ... import debug DEFAULT_HEADERS = { @@ -55,24 +57,33 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin): label = "OpenAI ChatGPT" url = "https://chatgpt.com" working = True + needs_auth = True supports_gpt_4 = True supports_message_history = True supports_system_message = True - default_model = None + default_model = "auto" default_vision_model = "gpt-4o" - models = [ "auto", "gpt-4o-mini", "gpt-4o", "gpt-4", "gpt-4-gizmo"] - - model_aliases = { - #"gpt-4-turbo": "gpt-4", - #"gpt-4": "gpt-4-gizmo", - #"dalle": "gpt-4", - } + fallback_models = ["auto", "gpt-4", "gpt-4o", "gpt-4o-mini", "gpt-4o-canmore", "o1-preview", "o1-mini"] + vision_models = fallback_models + _api_key: str = None _headers: dict = None _cookies: Cookies = None _expires: int = None @classmethod + def get_models(cls): + if not cls.models: + try: + response = requests.get(f"{cls.url}/backend-anon/models") + response.raise_for_status() + data = response.json() + cls.models = [model.get("slug") for model in data.get("models")] + except Exception: + cls.models = cls.fallback_models + return cls.models + + @classmethod async def create( cls, prompt: str = None, @@ -195,7 +206,7 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin): async with session.get(url, headers=headers) as response: cls._update_request_args(session) if response.status == 401: - raise MissingAuthError('Add a "api_key" or a .har file' if cls._api_key is None else "Invalid api key") + raise MissingAuthError('Add a .har file for OpenaiChat' if cls._api_key is None else "Invalid api key") await raise_for_status(response) data = await response.json() if "categories" in data: @@ -218,9 +229,12 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin): """ # Create a message object with the user role and the content messages = [{ - "id": str(uuid.uuid4()), "author": {"role": message["role"]}, "content": {"content_type": "text", "parts": [message["content"]]}, + "id": str(uuid.uuid4()), + "create_time": int(time.time()), + "id": str(uuid.uuid4()), + "metadata": {"serialization_metadata": {"custom_symbol_offsets": []}} } for message in messages] # Check if there is an image response @@ -249,7 +263,7 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin): return messages @classmethod - async def get_generated_image(cls, session: StreamSession, headers: dict, line: dict) -> ImageResponse: + async def get_generated_image(cls, session: StreamSession, headers: dict, element: dict) -> ImageResponse: """ Retrieves the image response based on the message content. @@ -268,15 +282,8 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin): Raises: RuntimeError: If there'san error in downloading the image, including issues with the HTTP request or response. """ - if "parts" not in line["message"]["content"]: - return - first_part = line["message"]["content"]["parts"][0] - if "asset_pointer" not in first_part or "metadata" not in first_part: - return - if first_part["metadata"] is None or first_part["metadata"]["dalle"] is None: - return - prompt = first_part["metadata"]["dalle"]["prompt"] - file_id = first_part["asset_pointer"].split("file-service://", 1)[1] + prompt = element["metadata"]["dalle"]["prompt"] + file_id = element["asset_pointer"].split("file-service://", 1)[1] try: async with session.get(f"{cls.url}/backend-api/files/{file_id}/download", headers=headers) as response: cls._update_request_args(session) @@ -363,20 +370,12 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin): ) as session: if cls._expires is not None and cls._expires < time.time(): cls._headers = cls._api_key = None - arkose_token = None - proofTokens = None try: - arkose_token, api_key, cookies, headers, proofTokens = await getArkoseAndAccessToken(proxy) - cls._create_request_args(cookies, headers) - cls._set_api_key(api_key) + await get_request_config(proxy) + cls._create_request_args(RequestConfig.cookies, RequestConfig.headers) + cls._set_api_key(RequestConfig.access_token) except NoValidHarFileError as e: - if cls._api_key is None and cls.needs_auth: - raise e - cls._create_request_args() - - if cls.default_model is None: - cls.default_model = cls.get_model(await cls.get_default_model(session, cls._headers)) - + await cls.nodriver_auth(proxy) try: image_request = await cls.upload_image(session, cls._headers, image, image_name) if image else None except Exception as e: @@ -384,9 +383,7 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin): if debug.logging: print("OpenaiChat: Upload image failed") print(f"{e.__class__.__name__}: {e}") - model = cls.get_model(model) - model = "text-davinci-002-render-sha" if model == "gpt-3.5-turbo" else model if conversation is None: conversation = Conversation(conversation_id, str(uuid.uuid4()) if parent_id is None else parent_id) else: @@ -399,36 +396,32 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin): f"{cls.url}/backend-anon/sentinel/chat-requirements" if cls._api_key is None else f"{cls.url}/backend-api/sentinel/chat-requirements", - json={"p": generate_proof_token(True, user_agent=cls._headers["user-agent"], proofTokens=proofTokens)}, + json={"p": get_requirements_token(RequestConfig.proof_token)}, headers=cls._headers ) as response: cls._update_request_args(session) await raise_for_status(response) - requirements = await response.json() - text_data = json.loads(requirements.get("text", "{}")) - need_arkose = text_data.get("turnstile", {}).get("required", False) - if need_arkose: - arkose_token = text_data.get("turnstile", {}).get("dx") - else: - need_arkose = requirements.get("arkose", {}).get("required", False) - chat_token = requirements["token"] - - if need_arkose and arkose_token is None: - arkose_token, api_key, cookies, headers, proofTokens = await getArkoseAndAccessToken(proxy) - cls._create_request_args(cookies, headers) - cls._set_api_key(api_key) - if arkose_token is None: + chat_requirements = await response.json() + need_turnstile = chat_requirements.get("turnstile", {}).get("required", False) + need_arkose = chat_requirements.get("arkose", {}).get("required", False) + chat_token = chat_requirements.get("token") + + if need_arkose and RequestConfig.arkose_token is None: + await get_request_config(proxy) + cls._create_request_args(RequestConfig,cookies, RequestConfig.headers) + cls._set_api_key(RequestConfig.access_token) + if RequestConfig.arkose_token is None: raise MissingAuthError("No arkose token found in .har file") - if "proofofwork" in requirements: + if "proofofwork" in chat_requirements: proofofwork = generate_proof_token( - **requirements["proofofwork"], + **chat_requirements["proofofwork"], user_agent=cls._headers["user-agent"], - proofTokens=proofTokens + proof_token=RequestConfig.proof_token ) if debug.logging: print( - 'Arkose:', False if not need_arkose else arkose_token[:12]+"...", + 'Arkose:', False if not need_arkose else RequestConfig.arkose_token[:12]+"...", 'Proofofwork:', False if proofofwork is None else proofofwork[:12]+"...", ) ws = None @@ -437,18 +430,20 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin): wss_url = (await response.json()).get("wss_url") if wss_url: ws = await session.ws_connect(wss_url) - websocket_request_id = str(uuid.uuid4()) data = { "action": action, - "conversation_mode": {"kind": "primary_assistant"}, - "force_paragen": False, - "force_rate_limit": False, - "conversation_id": conversation.conversation_id, + "messages": None, "parent_message_id": conversation.message_id, "model": model, + "paragen_cot_summary_display_override": "allow", "history_and_training_disabled": history_disabled and not auto_continue and not return_conversation, - "websocket_request_id": websocket_request_id + "conversation_mode": {"kind":"primary_assistant"}, + "websocket_request_id": str(uuid.uuid4()), + "supported_encodings": ["v1"], + "supports_buffering": True } + if conversation.conversation_id is not None: + data["conversation_id"] = conversation.conversation_id if action != "continue": messages = messages if conversation_id is None else [messages[-1]] data["messages"] = cls.create_messages(messages, image_request) @@ -457,10 +452,12 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin): "Openai-Sentinel-Chat-Requirements-Token": chat_token, **cls._headers } - if need_arkose: - headers["Openai-Sentinel-Arkose-Token"] = arkose_token + if RequestConfig.arkose_token: + headers["Openai-Sentinel-Arkose-Token"] = RequestConfig.arkose_token if proofofwork is not None: headers["Openai-Sentinel-Proof-Token"] = proofofwork + if need_turnstile and RequestConfig.turnstile_token is not None: + headers['openai-sentinel-turnstile-token'] = RequestConfig.turnstile_token async with session.post( f"{cls.url}/backend-anon/conversation" if cls._api_key is None else @@ -509,7 +506,6 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin): fields: Conversation, ws = None ) -> AsyncIterator: - last_message: int = 0 async for message in messages: if message.startswith(b'{"wss_url":'): message = json.loads(message) @@ -526,10 +522,6 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin): async for chunk in cls.iter_messages_line(session, message, fields): if fields.finish_reason is not None: break - elif isinstance(chunk, str): - if len(chunk) > last_message: - yield chunk[last_message:] - last_message = len(chunk) else: yield chunk if fields.finish_reason is not None: @@ -547,148 +539,99 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin): line = json.loads(line[6:]) except: return - if "message" not in line: - return - if "error" in line and line["error"]: - raise RuntimeError(line["error"]) - if "message_type" not in line["message"]["metadata"]: - return - image_response = await cls.get_generated_image(session, cls._headers, line) - if image_response is not None: - yield image_response - if line["message"]["author"]["role"] != "assistant": - return - if line["message"]["content"]["content_type"] != "text": + if isinstance(line, dict) and "v" in line: + v = line.get("v") + if isinstance(v, str): + yield v + elif isinstance(v, list): + for m in v: + if m.get("p") == "/message/content/parts/0": + yield m.get("v") + elif m.get("p") == "/message/metadata": + fields.finish_reason = m.get("v", {}).get("finish_details", {}).get("type") + break + elif isinstance(v, dict): + if fields.conversation_id is None: + fields.conversation_id = v.get("conversation_id") + fields.message_id = v.get("message", {}).get("id") + c = v.get("message", {}).get("content", {}) + if c.get("content_type") == "multimodal_text": + generated_images = [] + for element in c.get("parts"): + if element.get("content_type") == "image_asset_pointer": + generated_images.append( + cls.get_generated_image(session, cls._headers, element) + ) + elif element.get("content_type") == "text": + for part in element.get("parts", []): + yield part + for image_response in await asyncio.gather(*generated_images): + yield image_response return - if line["message"]["metadata"]["message_type"] not in ("next", "continue", "variant"): - return - if line["message"]["recipient"] != "all": - return - if fields.conversation_id is None: - fields.conversation_id = line["conversation_id"] - fields.message_id = line["message"]["id"] - if "parts" in line["message"]["content"]: - yield line["message"]["content"]["parts"][0] - if "finish_details" in line["message"]["metadata"]: - fields.finish_reason = line["message"]["metadata"]["finish_details"]["type"] - - @classmethod - async def webview_access_token(cls) -> str: - window = webview.create_window("OpenAI Chat", cls.url) - await asyncio.sleep(3) - prompt_input = None - while not prompt_input: - try: - await asyncio.sleep(1) - prompt_input = window.dom.get_element("#prompt-textarea") - except: - ... - window.evaluate_js(""" -this._fetch = this.fetch; -this.fetch = async (url, options) => { - const response = await this._fetch(url, options); - if (url == "https://chatgpt.com/backend-api/conversation") { - this._headers = options.headers; - return response; - } - return response; -}; -""") - window.evaluate_js(""" - document.querySelector('.from-token-main-surface-secondary').click(); - """) - headers = None - while headers is None: - headers = window.evaluate_js("this._headers") - await asyncio.sleep(1) - headers["User-Agent"] = window.evaluate_js("this.navigator.userAgent") - cookies = [list(*cookie.items()) for cookie in window.get_cookies()] - window.destroy() - cls._cookies = dict([(name, cookie.value) for name, cookie in cookies]) - cls._headers = headers - cls._expires = int(time.time()) + 60 * 60 * 4 - cls._update_cookie_header() + if "error" in line and line.get("error"): + raise RuntimeError(line.get("error")) @classmethod - async def nodriver_access_token(cls, proxy: str = None): - try: - import nodriver as uc - except ImportError: + async def nodriver_auth(cls, proxy: str = None): + if not has_nodriver: return - try: - from platformdirs import user_config_dir + if has_platformdirs: user_data_dir = user_config_dir("g4f-nodriver") - except: + else: user_data_dir = None if debug.logging: print(f"Open nodriver with user_dir: {user_data_dir}") - browser = await uc.start( + browser = await nodriver.start( user_data_dir=user_data_dir, browser_args=None if proxy is None else [f"--proxy-server={proxy}"], ) - page = await browser.get("https://chatgpt.com/") - await page.select("[id^=headlessui-menu-button-]", 240) - api_key = await page.evaluate( - "(async () => {" - "let session = await fetch('/api/auth/session');" - "let data = await session.json();" - "let accessToken = data['accessToken'];" - "let expires = new Date(); expires.setTime(expires.getTime() + 60 * 60 * 4 * 1000);" - "document.cookie = 'access_token=' + accessToken + ';expires=' + expires.toUTCString() + ';path=/';" - "return accessToken;" - "})();", - await_promise=True - ) - cookies = {} - for c in await page.browser.cookies.get_all(): - if c.domain.endswith("chatgpt.com"): - cookies[c.name] = c.value - user_agent = await page.evaluate("window.navigator.userAgent") - await page.close() - cls._create_request_args(cookies, user_agent=user_agent) - cls._set_api_key(api_key) - - @classmethod - def browse_access_token(cls, proxy: str = None, timeout: int = 1200) -> None: - """ - Browse to obtain an access token. - - Args: - proxy (str): Proxy to use for browsing. - - Returns: - tuple[str, dict]: A tuple containing the access token and cookies. - """ - driver = get_browser(proxy=proxy) + page = browser.main_tab + def on_request(event: nodriver.cdp.network.RequestWillBeSent): + if event.request.url == start_url or event.request.url.startswith(conversation_url): + RequestConfig.access_request_id = event.request_id + RequestConfig.headers = event.request.headers + elif event.request.url in (backend_url, backend_anon_url): + if "OpenAI-Sentinel-Proof-Token" in event.request.headers: + RequestConfig.proof_token = json.loads(base64.b64decode( + event.request.headers["OpenAI-Sentinel-Proof-Token"].split("gAAAAAB", 1)[-1].encode() + ).decode()) + if "OpenAI-Sentinel-Turnstile-Token" in event.request.headers: + RequestConfig.turnstile_token = event.request.headers["OpenAI-Sentinel-Turnstile-Token"] + if "Authorization" in event.request.headers: + RequestConfig.access_token = event.request.headers["Authorization"].split()[-1] + elif event.request.url == arkose_url: + RequestConfig.arkose_request = arkReq( + arkURL=event.request.url, + arkBx=None, + arkHeader=event.request.headers, + arkBody=event.request.post_data, + userAgent=event.request.headers.get("user-agent") + ) + await page.send(nodriver.cdp.network.enable()) + page.add_handler(nodriver.cdp.network.RequestWillBeSent, on_request) + page = await browser.get(cls.url) try: - driver.get(f"{cls.url}/") - WebDriverWait(driver, timeout).until(EC.presence_of_element_located((By.ID, "prompt-textarea"))) - access_token = driver.execute_script( - "let session = await fetch('/api/auth/session');" - "let data = await session.json();" - "let accessToken = data['accessToken'];" - "let expires = new Date(); expires.setTime(expires.getTime() + 60 * 60 * 4 * 1000);" - "document.cookie = 'access_token=' + accessToken + ';expires=' + expires.toUTCString() + ';path=/';" - "return accessToken;" - ) - args = get_args_from_browser(f"{cls.url}/", driver, do_bypass_cloudflare=False) - cls._headers = args["headers"] - cls._cookies = args["cookies"] - cls._update_cookie_header() - cls._set_api_key(access_token) - finally: - driver.close() - - @classmethod - async def fetch_access_token(cls, session: StreamSession, headers: dict): - async with session.get( - f"{cls.url}/api/auth/session", - headers=headers - ) as response: - if response.ok: - data = await response.json() - if "accessToken" in data: - return data["accessToken"] + if RequestConfig.access_request_id is not None: + body = await page.send(get_response_body(RequestConfig.access_request_id)) + if isinstance(body, tuple) and body: + body = body[0] + if body: + match = re.search(r'"accessToken":"(.*?)"', body) + if match: + RequestConfig.access_token = match.group(1) + except KeyError: + pass + for c in await page.send(nodriver.cdp.network.get_cookies([cls.url])): + RequestConfig.cookies[c.name] = c.value + RequestConfig.user_agent = await page.evaluate("window.navigator.userAgent") + await page.select("#prompt-textarea", 240) + while True: + if RequestConfig.proof_token: + break + await asyncio.sleep(1) + await page.close() + cls._create_request_args(RequestConfig.cookies, RequestConfig.headers, user_agent=RequestConfig.user_agent) + cls._set_api_key(RequestConfig.access_token) @staticmethod def get_default_headers() -> dict: @@ -715,7 +658,8 @@ this.fetch = async (url, options) => { def _set_api_key(cls, api_key: str): cls._api_key = api_key cls._expires = int(time.time()) + 60 * 60 * 4 - cls._headers["authorization"] = f"Bearer {api_key}" + if api_key: + cls._headers["authorization"] = f"Bearer {api_key}" @classmethod def _update_cookie_header(cls): |