From 0d25ab4096d7d464d2002ef994fd5ba168725b94 Mon Sep 17 00:00:00 2001
From: Maximilian Rink <maximilian.rink@telekom.de>
Date: Mon, 12 Oct 2026 20:20:39 +0200
Subject: [PATCH] fix: keep type-check delta stable after cache update

---
 .../llms/openrouter/image_edit/transformation.py   | 12 ++++++------
 litellm/llms/sagemaker/common_utils.py             |  2 +-
 .../context_caching/vertex_ai_context_caching.py   | 14 +++++++-------
 3 files changed, 14 insertions(+), 14 deletions(-)

diff --git a/litellm/llms/openrouter/image_edit/transformation.py b/litellm/llms/openrouter/image_edit/transformation.py
index a73a2fe4192..393feefa75a 100644
--- a/litellm/llms/openrouter/image_edit/transformation.py
+++ b/litellm/llms/openrouter/image_edit/transformation.py
@@ -211,7 +211,7 @@ class OpenRouterImageEditConfig(BaseImageEditConfig):
                 if img is None:
                     continue
                 mime_type = ImageEditRequestUtils.get_image_content_type(img)
-                image_bytes = self._read_image_bytes(img)
+                image_bytes = self._read_image_bytes(cast(FileTypes, img))
                 b64_data = base64.b64encode(image_bytes).decode("utf-8")
                 content_parts.append(
                     {
@@ -372,13 +372,13 @@ class OpenRouterImageEditConfig(BaseImageEditConfig):
                 input_image_tokens = prompt_tokens_details.get("image_tokens", 0)
 
             model_response.usage = ImageUsage(
-                input_tokens=prompt_tokens,
+                input_tokens=int(prompt_tokens),
                 input_tokens_details=ImageUsageInputTokensDetails(
-                    image_tokens=input_image_tokens,
-                    text_tokens=prompt_tokens - input_image_tokens,
+                    image_tokens=int(input_image_tokens),
+                    text_tokens=int(prompt_tokens - input_image_tokens),
                 ),
-                output_tokens=image_tokens,
-                total_tokens=total_tokens,
+                output_tokens=int(image_tokens),
+                total_tokens=int(total_tokens),
             )
 
             cost: Final = usage_data.get("cost")
diff --git a/litellm/llms/sagemaker/common_utils.py b/litellm/llms/sagemaker/common_utils.py
index 14f0e3ead62..66531518888 100644
--- a/litellm/llms/sagemaker/common_utils.py
+++ b/litellm/llms/sagemaker/common_utils.py
@@ -254,7 +254,7 @@ class AWSEventStreamDecoder:
             )
 
         return GChunk(
-            text=_text,
+            text=_text or "",
             index=_index,
             is_finished=is_finished,
             finish_reason=finish_reason,
diff --git a/litellm/llms/vertex_ai/context_caching/vertex_ai_context_caching.py b/litellm/llms/vertex_ai/context_caching/vertex_ai_context_caching.py
index f8939152326..3208a3e5180 100644
--- a/litellm/llms/vertex_ai/context_caching/vertex_ai_context_caching.py
+++ b/litellm/llms/vertex_ai/context_caching/vertex_ai_context_caching.py
@@ -1,4 +1,4 @@
-from collections.abc import Sequence
+from collections.abc import Mapping, Sequence
 from typing import (
     Final,
     Literal,
@@ -417,7 +417,7 @@ class ContextCachingEndpoints(VertexBase):
             "Content-Type": "application/json",
         }
         if isinstance(token, dict):
-            headers.update(token)
+            headers.update(cast(Mapping[str, str], token))
         elif token is not None:
             headers["Authorization"] = f"Bearer {token}"
         if extra_headers is not None:
@@ -471,7 +471,7 @@ class ContextCachingEndpoints(VertexBase):
         cached_content_request_body: Final = transform_openai_messages_to_gemini_context_caching(
             model=model,
             messages=cached_messages,
-            cache_key=generated_cache_key,
+            cache_key=generated_cache_key or "",
             custom_llm_provider=custom_llm_provider,
             vertex_project=vertex_project,
             vertex_location=vertex_location,
@@ -496,7 +496,7 @@ class ContextCachingEndpoints(VertexBase):
             response: Final = client.post(
                 url=url,
                 headers=headers,
-                json=cached_content_request_body,
+                json=cast(dict[str, object], cached_content_request_body),
             )
             response.raise_for_status()
         except httpx.HTTPStatusError as err:
@@ -597,7 +597,7 @@ class ContextCachingEndpoints(VertexBase):
             "Content-Type": "application/json",
         }
         if isinstance(token, dict):
-            headers.update(token)
+            headers.update(cast(Mapping[str, str], token))
         elif token is not None:
             headers["Authorization"] = f"Bearer {token}"
         if extra_headers is not None:
@@ -648,7 +648,7 @@ class ContextCachingEndpoints(VertexBase):
             cached_content_request_body: Final = transform_openai_messages_to_gemini_context_caching(
                 model=model,
                 messages=cached_messages,
-                cache_key=generated_cache_key,
+                cache_key=generated_cache_key or "",
                 custom_llm_provider=custom_llm_provider,
                 vertex_project=vertex_project,
                 vertex_location=vertex_location,
@@ -673,7 +673,7 @@ class ContextCachingEndpoints(VertexBase):
                 response: Final = await client.post(
                     url=url,
                     headers=headers,
-                    json=cached_content_request_body,
+                    json=cast(dict[str, object], cached_content_request_body),
                 )
                 response.raise_for_status()
             except httpx.HTTPStatusError as err: