fixed sequential human messages with tool calling

libs/vertexai/langchain_google_vertexai/chat_models.py

@@ -303,7 +303,11 @@ def _convert_to_parts(message: BaseMessage) -> List[Part]:
             if system_parts is not None:
                 parts = system_parts + parts
                 system_parts = None
-            vertex_messages.append(Content(role=role, parts=parts))
+            if vertex_messages and vertex_messages[-1].role == "user":
+                prev_parts = list(vertex_messages[-1].parts)
+                vertex_messages[-1] = Content(role=role, parts=prev_parts + parts)
+            else:
+                vertex_messages.append(Content(role=role, parts=parts))
         elif isinstance(message, AIMessage):
             prev_ai_message = message
             role = "model"

libs/vertexai/tests/integration_tests/test_chat_models.py

@@ -856,3 +856,59 @@ def test_json_serializable() -> None:
     llm.prediction_client
     llm.async_prediction_client
     json.loads(llm.model_dump_json())
+
+
+@pytest.mark.release
+def test_langgraph_example() -> None:
+    llm = ChatVertexAI(
+        model_name="gemini-1.5-pro-001",
+        max_output_tokens=8192,
+        temperature=0.2,
+    )
+
+    add_declaration = {
+        "name": "add",
+        "description": "Adds a and b.",
+        "parameters": {
+            "properties": {
+                "a": {"description": "first int", "type": "integer"},
+                "b": {"description": "second int", "type": "integer"},
+            },
+            "required": ["a", "b"],
+            "type": "object",
+        },
+    }
+
+    multiply_declaration = {
+        "name": "multiply",
+        "description": "Multiply a and b.",
+        "parameters": {
+            "properties": {
+                "a": {"description": "first int", "type": "integer"},
+                "b": {"description": "second int", "type": "integer"},
+            },
+            "required": ["a", "b"],
+            "type": "object",
+        },
+    }
+
+    messages = [
+        SystemMessage(
+            content=(
+                "You are a helpful assistant tasked with performing "
+                "arithmetic on a set of inputs."
+            )
+        ),
+        HumanMessage(content="Multiply 2 and 3"),
+        HumanMessage(content="No, actually multiply 3 and 3!"),
+    ]
+    step1 = llm.invoke(
+        messages,
+        tools=[{"function_declarations": [add_declaration, multiply_declaration]}],
+    )
+    step2 = llm.invoke(
+        messages
+        + [step1, ToolMessage(content="9", tool_call_id=step1.tool_calls[0]["id"])],  # type: ignore[attr-defined]
+        tools=[{"function_declarations": [add_declaration, multiply_declaration]}],
+    )
+    assert isinstance(step2, AIMessage)