When running a tool call (tools.py), it should accept None as the argument. (Issue with end_turn tool)

[thshorrock]

Description

I am using Qwen2.5 as my llm. When calling the end_turn tool in multiagent tasks, the "end_turn" tool is called with None for the arguments, rather than an empty dictionary as is done with gpt.

This causes this error:

| ERROR | Task run 'Tool call: end_turn' - Task run failed with exception: TypeError('controlflow.orchestration.turn_strategies.get_end_turn_tool..end_turn() argument after ** must be a mapping, not NoneType') - Retries are exhausted Traceback (most recent call last): controlflow

The issue is this part of the file tools/tools.py (around line 60)

    @prefect_task(task_run_name="Tool call: {self.name}")
    def run(self, input: dict):
        
        result = self.fn(**input)

This issue is resolved if a check for None is added:

    @prefect_task(task_run_name="Tool call: {self.name}")
    def run(self, input: dict):
    
        # Set input to an empty dictionary if it is None
        input = input or {} 
        
        result = self.fn(**input)

Example Code

import controlflow as cf

model = ChatOpenAI(temperature=0.0, model="Qwen/Qwen2.5-7B-Instruct",  base_url="my-vllm-server/v1")

optimist = cf.Agent(
    name="Optimist",
    instructions="Always find the best in every situation.", 
    model=model
)

pessimist = cf.Agent(
    name="Pessimist",
    instructions="Always find the worst in every situation.",
    model=model
)

cf.run(
    f"Debate world peace",
    instructions=(
        "Mark the task successful once both agents have have spoken"
    ),
    agents=[optimist, pessimist],
)

Version Information

control flow version: 0.11.2
python 3.11.0

Additional Context

No response