Getting FinishReason.MALFORMED_FUNCTION_CALL when function calling arugments contain large amount of text content

Gemini API
1

In the python sdk when function calling is returned by gemini and it contains large amount of text it fails and returns malformed function call.
For example i have a write_file tool attached that takes filename and file content, when i ask it to create a large report of something it uses the tool to write it and second argument contains that large text and it just fails with automatic function calling on and off, first I was confused but this is the only thing that seems to be consistent.

I dont know if there is a limit on how large the function calling response should be, but maybe that is the reason? Seems like a parsing faliure or gemini just stops midway.

3

Hi @FlinCode,

Welcome to forum!!

Oh interesting, to help me better understand the issue and replicate it on my end, could you possibly provide a simplified sample of the prompt you’re using, along with a representative example of the kind of “large report” content that’s causing the problem? This would allow me to test the scenario and see if I can reproduce the error.

Cheers!

4

It would also have access to a search tool and get website content tool and i would ask it to look up specific topic like Reinforcement learning and tell it to write a learn large documenation/paper/article on it and thats the usual scenario where it would fail, to add it would usually have access to around 30 more tools including those. But I would expect gemini to not hallucinate and write the improper function calling structure that could cause the malformed function call.
I am still not 100% sure if the large content is the core issue here but most of the time this is what caused that error for me.

6

Hi @FlinCode,

import os
import datetime
from google import genai
from google.genai import types
from google.colab import userdata

STORAGE_DIR = "gemini_outputs"  # Directory to store outputs

# Create the storage directory if it doesn't exist
if not os.path.exists(STORAGE_DIR):
    os.makedirs(STORAGE_DIR)

def save_content_to_file(file_name: str, file_content: str):
    """Givem file name and store the content in the file.

    Args:
        text_to_store: file name.
        prompt: Content to be stored in the file.

    Returns:
        None
    """

    with open(file_name, "w") as f:
        f.write(f"Response:\n{file_content}")  # Store the response

client = genai.Client(api_key=userdata.get('GOOGLE_API_KEY'))
model_id = "gemini-2.0-flash"

save_content_to_file_tool = types.Tool(function_declarations=[
    types.FunctionDeclaration(
        name="save_content_to_file",
        description="The content generated in response to the prompt is stored to the file name provided.",
        parameters=types.Schema(
            properties={
                'file_name': types.Schema(type='STRING'),
                'file_content': types.Schema(type='STRING'),
            },
            type='OBJECT',
        ),
    )
])

google_search_tool = types.Tool(
    google_search = types.GoogleSearch()
)

response = client.models.generate_content(
    model=model_id,
    contents="What is reinforcement learning? store the content into reinforcement_learning.txt",
    config=types.GenerateContentConfig(
        tools=[save_content_to_file_tool, google_search_tool],
        tool_config={

        }
    )
)

It seems like sometimes I am as well getting malformed function call. Let me check again and dig some more about it. Sorry in the delayed response and thanks for reporting this.

1 Like