From 7e4f07c45dcfbdc36df56cae4ab9109f79af1bf5 Mon Sep 17 00:00:00 2001
From: Chesars <cesarponce19544@gmail.com>
Date: Sun, 22 Feb 2026 09:37:26 -0300
Subject: [PATCH] fix(vertex): use module-level import for ModelResponseStream
 instead of type: ignore

Move ModelResponseStream to module-level import so MyPy can properly
narrow the Union type after isinstance checks, removing the need for
type: ignore suppressions.
---
 .../vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py b/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py
index 377801ea59..28e6919065 100644
--- a/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py
+++ b/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py
@@ -80,6 +80,7 @@ from litellm.types.utils import (
     ChatCompletionTokenLogprob,
     ChoiceLogprobs,
     CompletionTokensDetailsWrapper,
+    ModelResponseStream,
     PromptTokensDetailsWrapper,
     TopLogprob,
     Usage,
@@ -1931,7 +1932,7 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig):
     @staticmethod
     def _process_candidates(  # noqa: PLR0915
         _candidates: List[Candidates],
-        model_response: Union[ModelResponse, "ModelResponseStream"],
+        model_response: Union[ModelResponse, ModelResponseStream],
         standard_optional_params: dict,
         cumulative_tool_call_index: int = 0,
     ) -> Tuple[List[dict], List[dict], List, List, int]:
@@ -2093,7 +2094,7 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig):
                     chat_completion_logprobs=chat_completion_logprobs,
                     image_response=image_response,
                 )
-                model_response.choices.append(choice)  # type: ignore[arg-type]
+                model_response.choices.append(choice)
             elif isinstance(model_response, ModelResponse):
                 choice = litellm.Choices(
                     finish_reason=VertexGeminiConfig._check_finish_reason(
@@ -2104,7 +2105,7 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig):
                     logprobs=chat_completion_logprobs,
                     enhancements=None,
                 )
-                model_response.choices.append(choice)  # type: ignore[arg-type]
+                model_response.choices.append(choice)
 
         return (
             grounding_metadata,