For each question, it would be beneficial if the user could provide a budget (e.g., the total number of tokens) for the question. Then, the agent can try to answer the question until the token budget is reached.
Current behavior: The condition for finishing a task is based on the number of steps (5 as default)
Expected behavior: The condition for finishing a task is based on the number of tokens used
For each question, it would be beneficial if the user could provide a budget (e.g., the total number of tokens) for the question. Then, the agent can try to answer the question until the token budget is reached.
Current behavior: The condition for finishing a task is based on the number of steps (5 as default)
Expected behavior: The condition for finishing a task is based on the number of tokens used