From d99c70b7eae08212f288a3e609cab262eae9b600 Mon Sep 17 00:00:00 2001
From: David-Wu1119 <133224895+David-Wu1119@users.noreply.github.com>
Date: Thu, 13 Aug 2026 18:11:53 +0800
Subject: [PATCH] fix(vllm): repair the return annotation after the PEP 604
 migration

litellm_internal_staging moved this module to X | None and dropped
Optional from its typing import, while this branch added a method
annotated Optional["CostResponseTypes"]. The merge was textually clean
but left the module raising NameError on import.

Use the quoted PEP 604 form the rest of the file now uses, and record a
mutable-ok reason for the request_data dict the base class signature
fixes.
---
 litellm/llms/vllm/passthrough/transformation.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/litellm/llms/vllm/passthrough/transformation.py b/litellm/llms/vllm/passthrough/transformation.py
index 2aced877f4a..6097f70913a 100644
--- a/litellm/llms/vllm/passthrough/transformation.py
+++ b/litellm/llms/vllm/passthrough/transformation.py
@@ -39,10 +39,10 @@ class VLLMPassthroughConfig(VLLMModelInfo, BasePassthroughConfig):
         model: str,
         custom_llm_provider: str,
         httpx_response: "Response",
-        request_data: dict,
+        request_data: dict,  # mutable-ok: base class signature
         logging_obj: "LiteLLMLoggingObj",
         endpoint: str,
-    ) -> Optional["CostResponseTypes"]:
+    ) -> "CostResponseTypes | None":
         from litellm import encoding
         from litellm.llms.openai.chat.gpt_transformation import OpenAIGPTConfig
         from litellm.types.utils import EmbeddingResponse, ModelResponse, Usage