fix(hosted_vllm): stop rewriting the caller's video file part into video_url in place

hosted_vllm converted a video file part into video_url by assigning into the caller's own content list, so a Router fallback to another provider received a part it does not read and the video was dropped without an error. Build a new content list for the vLLM request instead

litellm/llms/hosted_vllm/chat/transformation.py

@@ -21,6 +21,7 @@ from litellm.types.llms.openai import (
     ChatCompletionToolCallFunctionChunk,
     ChatCompletionVideoObject,
     ChatCompletionVideoUrlObject,
+    OpenAIMessageContentListBlock,
 )
 
 from ....utils import remove_additional_properties, remove_strict_from_schema
@@ -142,6 +143,12 @@ class HostedVLLMChatConfig(OpenAIGPTConfig):
             return ChatCompletionVideoObject(type="video_url", video_url=ChatCompletionVideoUrlObject(url=file_data))
         raise ValueError("file_id or file_data is required")
 
+    def _video_file_as_video_url(self, content_item: OpenAIMessageContentListBlock) -> OpenAIMessageContentListBlock:
+        if content_item.get("type") != "file":
+            return content_item
+        file_item: Final = cast(ChatCompletionFileObject, content_item)
+        return self._convert_file_to_video_url(file_item) if self._is_video_file(file_item) else content_item
+
     @overload
     def _transform_messages(
         self, messages: list[AllMessageValues], model: str, is_async: Literal[True]
@@ -224,14 +231,7 @@ class HostedVLLMChatConfig(OpenAIGPTConfig):
             elif message["role"] == "user":
                 message_content = message.get("content")
                 if message_content and isinstance(message_content, list):
-                    replaced_content_items: list[tuple[int, ChatCompletionFileObject]] = []
-                    for idx, content_item in enumerate(message_content):
-                        if content_item.get("type") == "file":
-                            content_item = cast(ChatCompletionFileObject, content_item)
-                            if self._is_video_file(content_item):
-                                replaced_content_items.append((idx, content_item))
-                    for idx, content_item in replaced_content_items:
-                        message_content[idx] = self._convert_file_to_video_url(content_item)
+                    message["content"] = [self._video_file_as_video_url(item) for item in message_content]
 
         if is_async:
             return super()._transform_messages(messages, model, is_async=cast(Literal[True], True))

tests/unit/llms/hosted_vllm/chat/test_hosted_vllm_chat_transformation.py

@@ -1,9 +1,11 @@
 import json
 from unittest.mock import MagicMock, patch
 
+import httpx
 import pytest
+import respx
 
-
+import litellm
 from litellm.constants import (
     DEFAULT_REASONING_EFFORT_HIGH_THINKING_BUDGET,
     DEFAULT_REASONING_EFFORT_LOW_THINKING_BUDGET,
@@ -43,6 +45,34 @@ def test_hosted_vllm_chat_transformation_file_url():
     ]
 
 
+def test_hosted_vllm_video_file_conversion_leaves_caller_messages_untouched():
+    video_file_part = {"type": "file", "file": {"file_id": "https://example.com/video.mp4", "format": "video/mp4"}}
+    messages = [{"role": "user", "content": [{"type": "text", "text": "Describe this video"}, video_file_part]}]
+    with respx.mock() as upstream:
+        route = upstream.post("http://vllm.test/v1/chat/completions").mock(
+            return_value=httpx.Response(
+                200,
+                json={
+                    "id": "chatcmpl-1",
+                    "object": "chat.completion",
+                    "created": 1,
+                    "model": "qwen-vl",
+                    "choices": [
+                        {"index": 0, "message": {"role": "assistant", "content": "A cat."}, "finish_reason": "stop"}
+                    ],
+                    "usage": {"prompt_tokens": 1, "completion_tokens": 1, "total_tokens": 2},
+                },
+            )
+        )
+        litellm.completion(model="hosted_vllm/qwen-vl", messages=messages, api_base="http://vllm.test/v1")
+
+    assert json.loads(route.calls.last.request.content)["messages"][0]["content"][1] == {
+        "type": "video_url",
+        "video_url": {"url": "https://example.com/video.mp4"},
+    }
+    assert messages[0]["content"][1] is video_file_part
+
+
 def test_hosted_vllm_supports_reasoning_effort():
     config = HostedVLLMChatConfig()
     supported_params = config.get_supported_openai_params(model="hosted_vllm/gpt-oss-120b")