Tools outputs truncated during execution?

[krish240574]

Bug description

I'm calling a tool, inside DataInterpreter - foo() - whch returns "data". The output of the tool always gets truncated, to approx. 1918 characters. This kills my execution, since I lose a lot of context.

Bug solved method

Environment information

• LLM type and model name:
• System version:
• Python version:
• 3.10
• MetaGPT version or branch: git clone https://github.com/geekan/MetaGPT

• packages version:
• installation method:

Screenshots or logs

[seehi]

Is it reproducible, could you provide more details, like the code that was executed. Additionally, does this issue describe the same problem as this: https://github.com/geekan/MetaGPT/issues/1162

[krish240574]

No, #1162 is an independent bug, somewhat related to this, but needs attention or a workaround.

Code I executed :

from metagpt.roles.di.data_interpreter import DataInterpreter from metagpt.tools.tool_registry import register_tool

@register_tool() def magic_function(arg1: str, arg2: int, arg3: str) -> dict: """ The magic function that does something.

	Args:
			arg1 (str): ...
			arg2 (int): ...

	Returns:
			dict: ...
	"""
	foo()
	return {"arg1": arg1 * 3, "arg2": arg2 * 5, "arg3" : arg3}

@register_tool() def foo(): """ Just return dummy output for testing """ return """<4000 character string - could be anything>"""

async def main(): di = DataInterpreter(tools=["magic_function", "foo"]) await di.run("""Call foo first and store its return value as "output". Then call the magic function with arg1 'A' and arg2 2 and arg3 output. Tell me the result.""")

if name == "main": import asyncio asyncio.run(main())

[seehi]

The reason is that parse_outputs has a parameter keep_len with a default value of 2000, which should be configurable later on. You can try to modify it manually to see the effect.

[krish240574]

Brilliant ! I changed the keep_len to 10000 and my outputs are fine now. Maybe make it a configurable parameter? Thank you for your prompt response.