ToolQA
ToolQA copied to clipboard
problems of tabletools.py
i am running the benchmark of coffee-hard, and i noticed a problem of the tabletool.py.
questions like this kind: "How much did the coffee price change from 2017-09-11 to 2018-04-03?" When the agent filtered the former date's price, self.data will only contain the exact data of that date. So when the agent filtered the later date, the environment will only return fail information.
i don't think this meet human's intuition to operate.