fmt

libs/core/langchain_core/language_models/chat_models.py

@@ -38,7 +38,11 @@ from langchain_core.messages import (
     AnyMessage,
     BaseMessage,
     BaseMessageChunk,
+    FunctionMessage,
     HumanMessage,
+    SystemMessage,
+    ToolCall,
+    ToolMessage,
     convert_to_messages,
     message_chunk_to_message,
 )
@@ -131,13 +135,25 @@ class BaseChatModel(BaseLanguageModel[BaseMessage], ABC):
         """Get the output type for this runnable."""
         return AnyMessage
 
+    @property
+    def _standardize_input_messages(self) -> bool:
+        return False
+
     def _convert_input(self, input: LanguageModelInput) -> PromptValue:
-        if isinstance(input, PromptValue):
+        if isinstance(input, StringPromptValue):
             return input
         elif isinstance(input, str):
             return StringPromptValue(text=input)
+        elif isinstance(input, ChatPromptValue):
+            messages = input.messages
+            if self._standardize_input_messages:
+                messages = _standardize_messages(messages)
+            return ChatPromptValue(messages=messages)
         elif isinstance(input, Sequence):
-            return ChatPromptValue(messages=convert_to_messages(input))
+            messages = convert_to_messages(input)
+            if self._standardize_input_messages:
+                messages = _standardize_messages(messages)
+            return ChatPromptValue(messages=messages)
         else:
             raise ValueError(
                 f"Invalid input type {type(input)}. "
@@ -961,3 +977,73 @@ def _gen_info_and_msg_metadata(
         **(generation.generation_info or {}),
         **generation.message.response_metadata,
     }
+
+
+def _standardize_messages(
+    messages: Sequence[BaseMessage],
+) -> List[Union[SystemMessage, HumanMessage, AIMessage, ToolMessage]]:
+    """Convert sequence of messages to a canonical form.
+
+    - Convert function message to tool message.
+    - Convert tool_result content blocks to tool messages.
+    - Convert tool_use content blocks to AIMessage.tool_calls.
+    """
+    standardized = []
+    for msg in messages:
+        if isinstance(msg, FunctionMessage):
+            standardized.append(
+                ToolMessage(
+                    msg.content, tool_call_id=msg.name + "_" + str(uuid.uuid4())
+                )
+            )
+        elif isinstance(msg.content, list):
+            msg_blocks = []
+            for block in msg.content:
+                if block.get("type") == "tool_result":
+                    if msg_blocks:
+                        standardized.append(
+                            msg.copy(
+                                update={"content": msg_blocks, "id": str(uuid.uuid4())}
+                            )
+                        )
+                    standardized.append(
+                        ToolMessage(
+                            block.get("content", ""),
+                            tool_call_id=block.get("tool_use_id", str(uuid.uuid4())),
+                            id=str(uuid.uuid4()),
+                        )
+                    )
+                    msg_blocks = []
+                elif block.get("type") == "tool_use":
+                    if msg_blocks:
+                        standardized.append(
+                            msg.copy(
+                                update={"content": msg_blocks, "id": str(uuid.uuid4())}
+                            )
+                        )
+                    tool_call = ToolCall(
+                        name=block.get("name", ""),
+                        args=block.get("input", ""),
+                        id=block.get("id", str(uuid.uuid4())),
+                    )
+                    if standardized and isinstance(standardized[-1], AIMessage):
+                        standardized[-1].tool_calls.append(tool_call)
+                    else:
+                        standardized.append(
+                            AIMessage("", tool_calls=[tool_call], id=str(uuid.uuid4()))
+                        )
+                    msg_blocks = []
+                else:
+                    msg_blocks.append(block)
+
+            if msg_blocks and msg_blocks != msg.content:
+                standardized.append(
+                    msg.copy(update={"content": msg_blocks, "id": str(uuid.uuid4())})
+                )
+            elif msg_blocks:
+                standardized.append(msg.copy(deep=True))
+            else:
+                pass
+        else:
+            standardized.append(msg.copy(deep=True))
+    return standardized

libs/partners/openai/langchain_openai/chat_models/base.py

@@ -338,6 +338,10 @@ class ChatOpenAI(BaseChatModel):
         """Return whether this model can be serialized by Langchain."""
         return True
 
+    @property
+    def _standardize_input_messages(self) -> bool:
+        return True
+
     client: Any = Field(default=None, exclude=True)  #: :meta private:
     async_client: Any = Field(default=None, exclude=True)  #: :meta private:
     model_name: str = Field(default="gpt-3.5-turbo", alias="model")