init commit (#16200)

Author: anthonyivn2

litellm/litellm_core_utils/prompt_templates/common_utils.py

@@ -94,6 +94,15 @@ def handle_messages_with_content_list_to_str_conversion(
     return messages
 
 
+def strip_name_from_message(message: AllMessageValues, allowed_name_roles: List[str] = ["user"]) -> AllMessageValues:
+    """
+    Removes 'name' from message
+    """
+    msg_copy = message.copy()
+    if msg_copy.get("role") not in allowed_name_roles:
+        msg_copy.pop("name", None)  # type: ignore
+    return msg_copy
+
 def strip_name_from_messages(
     messages: List[AllMessageValues], allowed_name_roles: List[str] = ["user"]
 ) -> List[AllMessageValues]:

litellm/llms/databricks/chat/transformation.py

@@ -26,7 +26,7 @@
     _should_convert_tool_call_to_json_mode,
 )
 from litellm.litellm_core_utils.prompt_templates.common_utils import (
-    strip_name_from_messages,
+    strip_name_from_message
 )
 from litellm.llms.base_llm.base_model_iterator import BaseModelResponseIterator
 from litellm.types.llms.anthropic import AllAnthropicToolsValues
@@ -332,8 +332,11 @@ def _transform_messages(
                 _message = message.model_dump(exclude_none=True)
             else:
                 _message = message
+            _message = strip_name_from_message(_message, allowed_name_roles=["user"])
+            # Move message-level cache_control into a content block when content is a string.
+            if "cache_control" in _message and isinstance(_message.get("content"), str):
+                _message = self._move_cache_control_into_string_content_block(_message)
             new_messages.append(_message)
-        new_messages = strip_name_from_messages(new_messages)
 
         if is_async:
             return super()._transform_messages(
@@ -344,6 +347,32 @@ def _transform_messages(
                 messages=new_messages, model=model, is_async=cast(Literal[False], False)
             )
 
+    def _move_cache_control_into_string_content_block(self, message: AllMessageValues) -> AllMessageValues:
+        """
+        Moves message-level cache_control into a content block when content is a string.
+        
+        Transforms:
+            {"role": "user", "content": "text", "cache_control": {...}}
+        Into:
+            {"role": "user", "content": [{"type": "text", "text": "text", "cache_control": {...}}]}
+        
+        This is required for Anthropic's prompt caching API when cache_control is specified
+        at the message level but content is a simple string (not already an array of content blocks).
+        """
+        content = message.get("content")
+        # Create new message with cache_control moved into content block
+        transformed_message = cast(dict[str, Any], message.copy())
+        cache_control = transformed_message.pop("cache_control")
+        transformed_message["content"] = [
+            {
+                "type": "text",
+                "text": content,
+                "cache_control": cache_control,
+            }
+        ]
+        return cast(AllMessageValues, transformed_message)
+        
+
     @staticmethod
     def extract_content_str(
         content: Optional[AllDatabricksContentValues],