fix(vertex): align stream checks and share code-assist singleton

litellm/llms/gemini/fallback_handler.py

@@ -2,9 +2,9 @@ from typing import Any, Awaitable, Callable, Dict
 
 from litellm._logging import verbose_logger
 from litellm.llms.gemini.common_utils import should_fallback_to_google_code_assist
-from litellm.llms.google_code_assist.chat import GoogleCodeAssistChat
+from litellm.llms.google_code_assist.chat import get_google_code_assist_chat
 
-_google_code_assist_chat = GoogleCodeAssistChat()
+_google_code_assist_chat = get_google_code_assist_chat()
 
 
 async def run_gemini_acompletion_with_code_assist_fallback(

litellm/llms/google_code_assist/chat.py

@@ -246,3 +246,13 @@ class GoogleCodeAssistChat:
         if isinstance(e, GoogleCodeAssistError):
             return e
         return GoogleCodeAssistError(status_code=500, message=str(e))
+
+
+_shared_google_code_assist_chat: Optional[GoogleCodeAssistChat] = None
+
+
+def get_google_code_assist_chat() -> GoogleCodeAssistChat:
+    global _shared_google_code_assist_chat
+    if _shared_google_code_assist_chat is None:
+        _shared_google_code_assist_chat = GoogleCodeAssistChat()
+    return _shared_google_code_assist_chat

litellm/llms/vertex_ai/common_utils.py

@@ -349,7 +349,6 @@ def _get_gemini_url(
         "v1alpha" if VertexGeminiConfig._is_gemini_3_or_newer(model) else "v1beta"
     )
 
-    endpoint = "generateContent"
     if mode == "chat":
         endpoint = "generateContent"
         if stream is True:
@@ -374,7 +373,7 @@ def _get_gemini_url(
     params = []
     if gemini_api_key and not gemini_oauth_token:
         params.append(f"key={gemini_api_key}")
-    if mode == "chat" and stream:
+    if mode == "chat" and stream is True:
         params.append("alt=sse")
 
     if params:

litellm/main.py

@@ -101,7 +101,7 @@ from litellm.llms.cohere.common_utils import CohereModelInfo
 from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler, HTTPHandler
 from litellm.llms.openai.chat.gpt_5_transformation import OpenAIGPT5Config
 from litellm.llms.openai_like.json_loader import JSONProviderRegistry
-from litellm.llms.google_code_assist.chat import GoogleCodeAssistChat
+from litellm.llms.google_code_assist.chat import get_google_code_assist_chat
 from litellm.llms.gemini.fallback_handler import (
     run_gemini_acompletion_with_code_assist_fallback,
     run_gemini_completion_with_code_assist_fallback,
@@ -161,7 +161,7 @@ from litellm.utils import (
     validate_openai_optional_params,
 )
 
-_google_code_assist_chat = GoogleCodeAssistChat()
+_google_code_assist_chat = get_google_code_assist_chat()
 
 from ._logging import verbose_logger
 from .caching.caching import disable_cache, enable_cache, update_cache

tests/test_litellm/llms/google_code_assist/test_google_code_assist.py

@@ -2,11 +2,17 @@ import pytest
 from unittest.mock import AsyncMock, MagicMock, patch
 import httpx
 import json
-from litellm.llms.google_code_assist.chat import GoogleCodeAssistChat
+from litellm.llms.google_code_assist.chat import (
+    GoogleCodeAssistChat,
+    get_google_code_assist_chat,
+)
 from litellm.types.utils import ModelResponse
 
 
 class TestGoogleCodeAssist:
+    def test_get_google_code_assist_chat_returns_singleton(self):
+        assert get_google_code_assist_chat() is get_google_code_assist_chat()
+
     @patch("litellm.llms.google_code_assist.chat._get_httpx_client")
     @patch("litellm.llms.gemini.common_utils.get_gemini_oauth_token")
     def test_completion_basic(self, mock_get_token, mock_get_client):

tests/test_litellm/llms/vertex_ai/test_vertex_ai_common_utils.py

@@ -581,6 +581,19 @@ def test_get_gemini_url_stream_query_param_only_for_chat_mode():
     assert "alt=sse" not in embedding_url
 
 
+def test_get_gemini_url_requires_literal_true_for_streaming_endpoint_and_alt_sse():
+    url, endpoint = _get_gemini_url(
+        mode="chat",
+        model="gemini-1.5-flash",
+        stream=1,  # truthy non-bool should not be treated as streaming=True
+        gemini_api_key="test-key",
+        gemini_oauth_token=None,
+    )
+
+    assert endpoint == "generateContent"
+    assert "alt=sse" not in url
+
+
 @pytest.mark.parametrize(
     "model_cost_entry, vertex_region, expected_region",
     [