发布

  • v1.2 79b1c6088d

    [BFCL Chore] Ensure Correct Input Format for Eval Checker (#860)

    frostbyte_neo 发布于 2025-01-04 06:03:59 +00:00

    In some cases, the model handler’s decode_ast method returns
    successfully but produces output in an unexpected format, causing issues
    in downstream evaluations that do not perform argument format
    validation. This problem is especially common when the model does not
    output any function calls, resulting in a human-readable string instead
    of the expected structure.

    This PR refines the is_function_calling_format_output function to
    enforce that outputs must be a list of dictionaries in the following
    format before calling the checker function:

    [
      {func1: {param1: val1, param2: val2, ...}},
      {func2: {param1: val1, param2: val2, ...}},
      ...
    ]
    

    Note: This PR will not affect the leaderboard score.

    下载附件