From 5c16856c3f7d358f109045fcaf7de835b216c772 Mon Sep 17 00:00:00 2001 From: hs_junxiang Date: Thu, 16 Nov 2023 17:13:35 +0800 Subject: fix: GPTalk can only use 5 times (#1255) --- g4f/Provider/GPTalk.py | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) (limited to 'g4f/Provider') diff --git a/g4f/Provider/GPTalk.py b/g4f/Provider/GPTalk.py index b5881e5d..5749ff2e 100644 --- a/g4f/Provider/GPTalk.py +++ b/g4f/Provider/GPTalk.py @@ -13,6 +13,7 @@ class GPTalk(AsyncGeneratorProvider): working = True supports_gpt_35_turbo = True _auth = None + used_times = 0 @classmethod async def create_async_generator( @@ -44,7 +45,7 @@ class GPTalk(AsyncGeneratorProvider): 'x-auth-timestamp': f"{timestamp}", } async with ClientSession(headers=headers) as session: - if not cls._auth or cls._auth["expires_at"] < timestamp: + if not cls._auth or cls._auth["expires_at"] < timestamp or cls.used_times == 5: data = { "fingerprint": secrets.token_hex(16).zfill(32), "platform": "fingerprint" @@ -52,6 +53,7 @@ class GPTalk(AsyncGeneratorProvider): async with session.post(f"{cls.url}/api/chatgpt/user/login", json=data, proxy=proxy) as response: response.raise_for_status() cls._auth = (await response.json())["data"] + cls.used_times = 0 data = { "content": format_prompt(messages), "accept": "stream", @@ -72,6 +74,7 @@ class GPTalk(AsyncGeneratorProvider): async with session.post(f"{cls.url}/api/chatgpt/chatapi/text", json=data, headers=headers, proxy=proxy) as response: response.raise_for_status() token = (await response.json())["data"]["token"] + cls.used_times += 1 last_message = "" async with session.get(f"{cls.url}/api/chatgpt/chatapi/stream", params={"token": token}, proxy=proxy) as response: response.raise_for_status() -- cgit v1.2.3