summaryrefslogtreecommitdiffstats
path: root/g4f/Provider/Koala.py
diff options
context:
space:
mode:
Diffstat (limited to 'g4f/Provider/Koala.py')
-rw-r--r--g4f/Provider/Koala.py11
1 files changed, 6 insertions, 5 deletions
diff --git a/g4f/Provider/Koala.py b/g4f/Provider/Koala.py
index 0e810083..14e533df 100644
--- a/g4f/Provider/Koala.py
+++ b/g4f/Provider/Koala.py
@@ -10,7 +10,8 @@ from .helper import get_random_string, get_connector
from ..requests import raise_for_status
class Koala(AsyncGeneratorProvider, ProviderModelMixin):
- url = "https://koala.sh"
+ url = "https://koala.sh/chat"
+ api_endpoint = "https://koala.sh/api/gpt/"
working = True
supports_message_history = True
supports_gpt_4 = True
@@ -26,17 +27,17 @@ class Koala(AsyncGeneratorProvider, ProviderModelMixin):
**kwargs: Any
) -> AsyncGenerator[Dict[str, Union[str, int, float, List[Dict[str, Any]], None]], None]:
if not model:
- model = "gpt-3.5-turbo"
+ model = "gpt-4o-mini"
headers = {
"User-Agent": "Mozilla/5.0 (X11; Ubuntu; Linux x86_64; rv:122.0) Gecko/20100101 Firefox/122.0",
"Accept": "text/event-stream",
"Accept-Language": "de,en-US;q=0.7,en;q=0.3",
"Accept-Encoding": "gzip, deflate, br",
- "Referer": f"{cls.url}/chat",
+ "Referer": f"{cls.url}",
"Flag-Real-Time-Data": "false",
"Visitor-ID": get_random_string(20),
- "Origin": cls.url,
+ "Origin": "https://koala.sh",
"Alt-Used": "koala.sh",
"Sec-Fetch-Dest": "empty",
"Sec-Fetch-Mode": "cors",
@@ -67,7 +68,7 @@ class Koala(AsyncGeneratorProvider, ProviderModelMixin):
"model": model,
}
- async with session.post(f"{cls.url}/api/gpt/", json=data, proxy=proxy) as response:
+ async with session.post(f"{cls.api_endpoint}", json=data, proxy=proxy) as response:
await raise_for_status(response)
async for chunk in cls._parse_event_stream(response):
yield chunk