From 08e308348b2825f4dfe309158c25a1d55ac45271 Mon Sep 17 00:00:00 2001 From: Heiner Lohaus Date: Mon, 20 Nov 2023 14:00:40 +0100 Subject: Add webdriver module --- g4f/Provider/needs_auth/OpenaiChat.py | 131 +++++++++++++++------------------- 1 file changed, 56 insertions(+), 75 deletions(-) (limited to 'g4f/Provider/needs_auth/OpenaiChat.py') diff --git a/g4f/Provider/needs_auth/OpenaiChat.py b/g4f/Provider/needs_auth/OpenaiChat.py index 9fd90812..8c9dd1e0 100644 --- a/g4f/Provider/needs_auth/OpenaiChat.py +++ b/g4f/Provider/needs_auth/OpenaiChat.py @@ -6,7 +6,8 @@ from asyncstdlib.itertools import tee from async_property import async_cached_property from ..base_provider import AsyncGeneratorProvider -from ..helper import get_browser, get_event_loop +from ..helper import get_event_loop +from ..webdriver import get_browser from ...typing import AsyncResult, Messages from ...requests import StreamSession @@ -38,7 +39,10 @@ class OpenaiChat(AsyncGeneratorProvider): **kwargs ) -> Response: if prompt: - messages.append({"role": "user", "content": prompt}) + messages.append({ + "role": "user", + "content": prompt + }) generator = cls.create_async_generator( model, messages, @@ -49,12 +53,9 @@ class OpenaiChat(AsyncGeneratorProvider): response_fields=True, **kwargs ) - fields: ResponseFields = await anext(generator) - if "access_token" not in kwargs: - kwargs["access_token"] = cls._access_token return Response( generator, - fields, + await anext(generator), action, messages, kwargs @@ -87,7 +88,6 @@ class OpenaiChat(AsyncGeneratorProvider): headers = { "Accept": "text/event-stream", "Authorization": f"Bearer {access_token}", - "Cookie": 'intercom-device-id-dgkjq2bp=0f047573-a750-46c8-be62-6d54b56e7bf0; ajs_user_id=user-iv3vxisaoNodwWpxmNpMfekH; ajs_anonymous_id=fd91be0b-0251-4222-ac1e-84b1071e9ec1; __Host-next-auth.csrf-token=d2b5f67d56f7dd6a0a42ae4becf2d1a6577b820a5edc88ab2018a59b9b506886%7Ce5c33eecc460988a137cbc72d90ee18f1b4e2f672104f368046df58e364376ac; _cfuvid=gt_mA.q6rue1.7d2.AR0KHpbVBS98i_ppfi.amj2._o-1700353424353-0-604800000; cf_clearance=GkHCfPSFU.NXGcHROoe4FantnqmnNcluhTNHz13Tk.M-1700353425-0-1-dfe77f81.816e9bc2.714615da-0.2.1700353425; __Secure-next-auth.callback-url=https%3A%2F%2Fchat.openai.com; intercom-session-dgkjq2bp=UWdrS1hHazk5VXN1c0V5Q1F0VXdCQmsyTU9pVjJMUkNpWnFnU3dKWmtIdGwxTC9wbjZuMk5hcEc0NWZDOGdndS0tSDNiaDNmMEdIL1RHU1dFWDBwOHFJUT09--f754361b91fddcd23a13b288dcb2bf8c7f509e91; _uasid="Z0FBQUFBQmxXVnV0a3dmVno4czRhcDc2ZVcwaUpSNUdZejlDR25YSk5NYTJQQkpyNmRvOGxjTHMyTlAxWmJhaURrMVhjLXZxQXdZeVpBbU1aczA5WUpHT2dwaS1MOWc4MnhyNWFnbGRzeGdJcGFKT0ZRdnBTMVJHcGV2MGNTSnVQY193c0hqUWIycHhQRVF4dENlZ3phcDdZeHgxdVhoalhrZmtZME9NbWhMQjdVR3Vzc3FRRk0ybjJjNWMwTWtIRjdPb19lUkFtRmV2MDVqd1kwWU11QTYtQkdZenEzVHhLMGplY1hZM3FlYUt1cVZaNWFTRldleEJETzJKQjk1VTJScy1GUnMxUVZWMnVxYklxMjdockVZbkZyd1R4U1RtMnA1ZzlSeXphdmVOVk9xeEdrRkVOSjhwTVd1QzFtQjhBcWdDaE92Q1VlM2pwcjFQTXRuLVJNRVlZSGpIdlZ0aGV3PT0="; _dd_s=rum=0&expire=1700356244884; __Secure-next-auth.session-token=eyJhbGciOiJkaXIiLCJlbmMiOiJBMjU2R0NNIn0..3aK6Fbdy2_8f07bf.8eT2xgonrCnz7ySY6qXFsg3kzL6UQfXKAYaw3tyn-6_X9657zy47k9qGvmi9mF0QKozj5jau3_Ca62AQQ7FmeC6Y2F1urtzqrXqwTTsQ2LuzFPIQkx6KKb2DXc8zW2-oyEzJ_EY5yxfLB2RlRkSh3M7bYNZh4_ltEcfkj38s_kIPGMxv34udtPWGWET99MCjkdwQWXylJag4s0fETA0orsBAKnGCyqAUNJbb_D7BYtGSV-MQ925kZMG6Di_QmfO0HQWURDYjmdRNcuy1PT_xJ1DJko8sjL42i4j3RhkNDkhqCIqyYImz2eHFWHW7rYKxTkrBhlCPMS5hRdcCswD7JYPcSBiwnVRYgyOocFGXoFvQgIZ2FX9NiZ3SMEVM1VwIGSE-qH0H2nMa8_iBvsOgOWJgKjVAvzzyzZvRVDUUHzJrikSFPNONVDU3h-04c1kVL4qIu9DfeTPN7n8AvNmYwMbro0L9-IUAeXNo4-pwF0Kt-AtTsamqWvMqnK4O_YOyLnDDlvkmnOvDC2d5uinwlQIxr6APO6qFfGLlHiLZemKoekxEE1Fx70dl-Ouhk1VIzbF3OC6XNNxeBm9BUYUiHdL0wj2H9rHgX4cz6ZmS_3VTgpD6UJh-evu5KJ2gIvjYmVbyzEN0aPNDxfvBaOm-Ezpy4bUJ2bUrOwNn-0knWkDiTvjYmNhCyefPCtCF6rpKNay8PCw_yh79C4SdEP6Q4V7LI0Tvdi5uz7kLCiBC4AT9L0ao1WDX03mkUOpjvzHDvPLmj8chW3lTVm_kA0eYGQY4wT0jzleWlfV0Q8rB2oYECNLWksA3F1zlGfcl4lQjprvTXRePkvAbMpoJEsZD3Ylq7-foLDLk4-M2LYAFZDs282AY04sFjAjQBxTELFCCuDgTIgTXSIskY_XCxpVXDbdLlbCJY7XVK45ybwtfqwlKRp8Mo0B131uQAFc-migHaUaoGujxJJk21bP8F0OmhNYHBo4FQqE1rQm2JH5bNM7txKeh5KXdJgVUVbRSr7OIp_OF5-Bx_v9eRBGAIDkue26E2-O8Rnrp5zQ5TnvecQLDaUzWavCLPwsZ0_gsOLBxNOmauNYZtF8IElCsQSFDdhoiMxXsYUm4ZYKEAy3GWq8HGTAvBhNkh1hvnI7y-d8-DOaZf_D_D98-olZfm-LUkeosLNpPB9rxYMqViCiW3KrXE9Yx0wlFm5ePKaVvR7Ym_EPhSOhJBKFPCvdTdMZSNPUcW0ZJBVByq0A9sxD51lYq3gaFyqh94S4s_ox182AQ3szGzHkdgLcnQmJG9OYvKxAVcd43eg6_gODAYhx02GjbMw-7JTAhyXSeCrlMteHyOXl8hai-3LilC3PmMzi7Vbu49dhF1s4LcVlUowen5ira44rQQaB26mdaOUoQfodgt66M3RTWGPXyK1Nb72AzSXsCKyaQPbzeb6cN0fdGSdG4ktwvR04eFNEkquo_3aKu2GmUKTD0XcRx9dYrfXjgY-X1DDTVs1YND2gRhdx7FFEeBVjtbj2UqmG3Rvd4IcHGe7OnYWw2MHDcol68SsR1KckXWwWREz7YTGUnDB2M1kx_H4W2mjclytnlHOnYU3RflegRPeSTbdzUZJvGKXCCz45luHkQWN_4DExE76D-9YqbFIz-RY5yL4h-Zs-i2xjm2K-4xCMM9nQIOqhLMqixIZQ2ldDAidKoYtbs5ppzbcBLyrZM96bq9DwRBY3aacqWdlRd-TfX0wv5KO4fo0sSh5FsuhuN0zcEV_NNXgqIEM_p14EcPqgbrAvCBQ8os70TRBQLXiF0EniSofGjxwF8kQvUk3C6Wfc8cTTeN-E6GxCVTn91HBwA1iSEZlRLMVb8_BcRJNqwbgnb_07jR6-eo42u88CR3KQdAWwbQRdMxsURFwZ0ujHXVGG0Ll6qCFBcHXWyDO1x1yHdHnw8_8yF26pnA2iPzrFR-8glMgIA-639sLuGAxjO1_ZuvJ9CAB41Az9S_jaZwaWy215Hk4-BRYD-MKmHtonwo3rrxhE67WJgbbu14efsw5nT6ow961pffgwXov5VA1Rg7nv1E8RvQOx7umWW6o8R4W6L8f2COsmPTXfgwIjoJKkjhUqAQ8ceG7cM0ET-38yaC0ObU8EkXfdGGgxI28qTEZWczG66_iM4hw7QEGCY5Cz2kbO6LETAiw9OsSigtBvDS7f0Ou0bZ41pdK7G3FmvdZAnjWPjObnDF4k4uWfn7mzt0fgj3FyqK20JezRDyGuAbUUhOvtZpc9sJpzxR34eXEZTouuALrHcGuNij4z6rx51FrQsaMtiup8QVrhtZbXtKLMYnWYSbkhuTeN2wY-xV1ZUsQlakIZszzGF7kuIG87KKWMpuPMvbXjz6Pp_gWJiIC6aQuk8xl5g0iBPycf_6Q-MtpuYxzNE2TpI1RyR9mHeXmteoRzrFiWp7yEC-QGNFyAJgxTqxM3CjHh1Jt6IddOsmn89rUo1dZM2Smijv_fbIv3avXLkIPX1KZjILeJCtpU0wAdsihDaRiRgDdx8fG__F8zuP0n7ziHas73cwrfg-Ujr6DhC0gTNxyd9dDA_oho9N7CQcy6EFmfNF2te7zpLony0859jtRv2t1TnpzAa1VvMK4u6mXuJ2XDo04_6GzLO3aPHinMdl1BcIAWnqAqWAu3euGFLTHOhXlfijut9N1OCifd_zWjhVtzlR39uFeCQBU5DyQArzQurdoMx8U1ETsnWgElxGSStRW-YQoPsAJ87eg9trqKspFpTVlAVN3t1GtoEAEhcwhe81SDssLmKGLc.7PqS6jRGTIfgTPlO7Ognvg; __cf_bm=VMWoAKEB45hQSwxXtnYXcurPaGZDJS4dMi6dIMFLwdw-1700355394-0-ATVsbq97iCaTaJbtYr8vtg1Zlbs3nLrJLKVBHYa2Jn7hhkGclqAy8Gbyn5ePEhDRqj93MsQmtayfYLqY5n4WiLY=; __cflb=0H28vVfF4aAyg2hkHFH9CkdHRXPsfCUf6VpYf2kz3RX' } async with StreamSession( proxies={"https": proxy}, @@ -95,24 +95,22 @@ class OpenaiChat(AsyncGeneratorProvider): headers=headers, timeout=timeout ) as session: - data = { - "action": action, - "arkose_token": await get_arkose_token(proxy, timeout), - "conversation_id": conversation_id, - "parent_message_id": parent_id, - "model": models[model], - "history_and_training_disabled": history_disabled and not auto_continue, - } - if action != "continue": - data["messages"] = [{ - "id": str(uuid.uuid4()), - "author": {"role": "user"}, - "content": {"content_type": "text", "parts": [messages[-1]["content"]]}, - }] - first = True end_turn = EndTurn() - while first or auto_continue and not end_turn.is_end: - first = False + while not end_turn.is_end: + data = { + "action": action, + "arkose_token": await get_arkose_token(proxy, timeout), + "conversation_id": conversation_id, + "parent_message_id": parent_id, + "model": models[model], + "history_and_training_disabled": history_disabled and not auto_continue, + } + if action != "continue": + data["messages"] = [{ + "id": str(uuid.uuid4()), + "author": {"role": "user"}, + "content": {"content_type": "text", "parts": [messages[-1]["content"]]}, + }] async with session.post(f"{cls.url}/backend-api/conversation", json=data) as response: try: response.raise_for_status() @@ -120,43 +118,38 @@ class OpenaiChat(AsyncGeneratorProvider): raise RuntimeError(f"Error {response.status_code}: {await response.text()}") last_message = 0 async for line in response.iter_lines(): - if line.startswith(b"data: "): - line = line[6:] - if line == b"[DONE]": - break - try: - line = json.loads(line) - except: - continue - if "message" not in line: - continue - if "error" in line and line["error"]: - raise RuntimeError(line["error"]) - if "message_type" not in line["message"]["metadata"]: - continue - if line["message"]["author"]["role"] != "assistant": - continue - if line["message"]["metadata"]["message_type"] in ("next", "continue", "variant"): - conversation_id = line["conversation_id"] - parent_id = line["message"]["id"] - if response_fields: - response_fields = False - yield ResponseFields(conversation_id, parent_id, end_turn) - new_message = line["message"]["content"]["parts"][0] - yield new_message[last_message:] - last_message = len(new_message) - if "finish_details" in line["message"]["metadata"]: - if line["message"]["metadata"]["finish_details"]["type"] == "max_tokens": - end_turn.end() - - data = { - "action": "continue", - "arkose_token": await get_arkose_token(proxy, timeout), - "conversation_id": conversation_id, - "parent_message_id": parent_id, - "model": models[model], - "history_and_training_disabled": False, - } + if not line.startswith(b"data: "): + continue + line = line[6:] + if line == b"[DONE]": + break + try: + line = json.loads(line) + except: + continue + if "message" not in line: + continue + if "error" in line and line["error"]: + raise RuntimeError(line["error"]) + if "message_type" not in line["message"]["metadata"]: + continue + if line["message"]["author"]["role"] != "assistant": + continue + if line["message"]["metadata"]["message_type"] in ("next", "continue", "variant"): + conversation_id = line["conversation_id"] + parent_id = line["message"]["id"] + if response_fields: + response_fields = False + yield ResponseFields(conversation_id, parent_id, end_turn) + new_message = line["message"]["content"]["parts"][0] + yield new_message[last_message:] + last_message = len(new_message) + if "finish_details" in line["message"]["metadata"]: + if line["message"]["metadata"]["finish_details"]["type"] == "stop": + end_turn.end() + if not auto_continue: + break + action = "continue" await asyncio.sleep(5) @classmethod @@ -167,7 +160,7 @@ class OpenaiChat(AsyncGeneratorProvider): from selenium.webdriver.support.ui import WebDriverWait from selenium.webdriver.support import expected_conditions as EC - driver = get_browser("~/openai", proxy=proxy) + driver = get_browser(proxy=proxy) except ImportError: return try: @@ -193,18 +186,6 @@ class OpenaiChat(AsyncGeneratorProvider): raise RuntimeError("Read access token failed") return cls._access_token - @classmethod - @property - def params(cls): - params = [ - ("model", "str"), - ("messages", "list[dict[str, str]]"), - ("stream", "bool"), - ("proxy", "str"), - ("access_token", "str"), - ] - param = ", ".join([": ".join(p) for p in params]) - return f"g4f.provider.{cls.__name__} supports: ({param})" async def get_arkose_token(proxy: str = None, timeout: int = None) -> str: config = { @@ -293,7 +274,7 @@ class Response(): async def variant(self, **kwargs) -> Response: if self.action != "next": - raise RuntimeError("Can't create variant with continue or variant request.") + raise RuntimeError("Can't create variant from continue or variant request.") return await OpenaiChat.create( **self._options, messages=self._messages, -- cgit v1.2.3