summaryrefslogtreecommitdiffstats
path: root/g4f/Provider/Vercel.py
diff options
context:
space:
mode:
Diffstat (limited to '')
-rw-r--r--g4f/Provider/Vercel.py14
1 files changed, 9 insertions, 5 deletions
diff --git a/g4f/Provider/Vercel.py b/g4f/Provider/Vercel.py
index 2d20ca6a..2d856664 100644
--- a/g4f/Provider/Vercel.py
+++ b/g4f/Provider/Vercel.py
@@ -2,7 +2,7 @@ from __future__ import annotations
import json, base64, requests, execjs, random, uuid
-from ..typing import Any, TypedDict, CreateResult
+from ..typing import Messages, TypedDict, CreateResult
from .base_provider import BaseProvider
from abc import abstractmethod
@@ -17,8 +17,9 @@ class Vercel(BaseProvider):
@abstractmethod
def create_completion(
model: str,
- messages: list[dict[str, str]],
+ messages: Messages,
stream: bool,
+ proxy: str = None,
**kwargs
) -> CreateResult:
if not model:
@@ -52,15 +53,18 @@ class Vercel(BaseProvider):
'model' : model_info[model]['id'],
'messages' : messages,
'playgroundId': str(uuid.uuid4()),
- 'chatIndex' : 0} | model_info[model]['default_params']
+ 'chatIndex' : 0,
+ **model_info[model]['default_params'],
+ **kwargs
+ }
max_retries = kwargs.get('max_retries', 20)
for i in range(max_retries):
response = requests.post('https://sdk.vercel.ai/api/generate',
- headers=headers, json=json_data, stream=True)
+ headers=headers, json=json_data, stream=True, proxies={"https": proxy})
try:
response.raise_for_status()
- except:
+ except Exception:
continue
for token in response.iter_content(chunk_size=None):
yield token.decode()