Merge pull request #21557 from SherifWaly/sherif.waly/add-reasoning-multi-turn-hosted-vllm

Convert thinking_blocks to content blocks for hosted_vllm multi-turn

litellm/llms/hosted_vllm/chat/transformation.py

@@ -137,10 +137,29 @@ class HostedVLLMChatConfig(OpenAIGPTConfig):
         self, messages: List[AllMessageValues], model: str, is_async: bool = False
     ) -> Union[List[AllMessageValues], Coroutine[Any, Any, List[AllMessageValues]]]:
         """
-        Support translating video files from file_id or file_data to video_url
+        Support translating:
+        - video files from file_id or file_data to video_url
+        - thinking_blocks on assistant messages to content blocks
         """
         for message in messages:
-            if message["role"] == "user":
+            if message["role"] == "assistant":
+                thinking_blocks = message.pop("thinking_blocks", None)  # type: ignore
+                if thinking_blocks:
+                    new_content: list = [
+                        {"type": block["type"], "thinking": block.get("thinking", "")}
+                        if block.get("type") == "thinking"
+                        else {"type": block["type"], "data": block.get("data", "")}
+                        for block in thinking_blocks
+                    ]
+                    existing_content = message.get("content")
+                    if isinstance(existing_content, str):
+                        new_content.append(
+                            {"type": "text", "text": existing_content}
+                        )
+                    elif isinstance(existing_content, list):
+                        new_content.extend(existing_content)
+                    message["content"] = new_content  # type: ignore
+            elif message["role"] == "user":
                 message_content = message.get("content")
                 if message_content and isinstance(message_content, list):
                     replaced_content_items: List[

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

@@ -165,3 +165,95 @@ def test_hosted_vllm_supports_thinking():
         drop_params=False,
     )
     assert optional_params["reasoning_effort"] == "low"
+
+
+def test_hosted_vllm_thinking_blocks_prepended_to_assistant_content():
+    """
+    Test that thinking_blocks on assistant messages are converted to content
+    blocks prepended before the existing content.
+    """
+    config = HostedVLLMChatConfig()
+    messages = [
+        {
+            "role": "user",
+            "content": "Hello",
+        },
+        {
+            "role": "assistant",
+            "content": "Here is my answer.",
+            "thinking_blocks": [
+                {
+                    "type": "thinking",
+                    "thinking": "Let me reason about this...",
+                    "signature": "abc123",
+                }
+            ],
+        },
+        {
+            "role": "user",
+            "content": "Follow up question",
+        },
+    ]
+    transformed = config.transform_request(
+        model="hosted_vllm/llama-3.1-70b-instruct",
+        messages=messages,
+        optional_params={},
+        litellm_params={},
+        headers={},
+    )
+    assistant_msg = transformed["messages"][1]
+    assert assistant_msg["role"] == "assistant"
+    assert isinstance(assistant_msg["content"], list)
+    assert assistant_msg["content"][0] == {
+        "type": "thinking",
+        "thinking": "Let me reason about this...",
+    }
+    assert assistant_msg["content"][1] == {
+        "type": "text",
+        "text": "Here is my answer.",
+    }
+    assert "thinking_blocks" not in assistant_msg
+
+
+def test_hosted_vllm_thinking_blocks_with_list_content():
+    """
+    Test thinking_blocks prepended when assistant content is already a list.
+    """
+    config = HostedVLLMChatConfig()
+    messages = [
+        {
+            "role": "assistant",
+            "content": [{"type": "text", "text": "Response text"}],
+            "thinking_blocks": [
+                {
+                    "type": "thinking",
+                    "thinking": "Step 1 reasoning",
+                    "signature": "sig1",
+                },
+                {
+                    "type": "thinking",
+                    "thinking": "Step 2 reasoning",
+                    "signature": "sig2",
+                },
+            ],
+        },
+    ]
+    transformed = config.transform_request(
+        model="hosted_vllm/llama-3.1-70b-instruct",
+        messages=messages,
+        optional_params={},
+        litellm_params={},
+        headers={},
+    )
+    assistant_msg = transformed["messages"][0]
+    assert len(assistant_msg["content"]) == 3
+    assert assistant_msg["content"][0] == {
+        "type": "thinking",
+        "thinking": "Step 1 reasoning",
+    }
+    assert assistant_msg["content"][1] == {
+        "type": "thinking",
+        "thinking": "Step 2 reasoning",
+    }
+    assert assistant_msg["content"][2] == {"type": "text", "text": "Response text"}
+    assert "thinking_blocks" not in assistant_msg