Agent-Tasks

There are two approaches to encorporating the agent tasks into your agent: 1) using the agent-tasks package, or 2) using the huggingface dataset.

We recommend using the agent-tasks package as it up to date and includes user-contributed tasks.

Using the agent-tasks package

We have created a package that contains all the tasks in the ML Research Benchmark. You can install the package using the following command:

pip install agent-tasks

Usage

The library exposes a single function, get_task

get_task: - path: path to copy the task to - benchmark: name of the benchmark - task: name of the task

This function will copy the task to the specified path and return a dictionary with the task name and prompt.

{
    "name": str, - name of the task
    "prompt": str, - prompt for the task
}

To use the tasks in your agent, you can import the tasks from the package as follows:

from agent_tasks import get_task

# Example usage
result = get_task("./", "full_benchmark", "baby_lm")
print(result['prompt'])