Some data do not have ground truth in the current version
#1
by
IntJudge
- opened
Hello, thank you for your sharing of InterleavedBench. However, we have found that there are several data (aboud 350 pieces) do not have any corresponding ground truth for generated responses. These data contain only text questions:
Hi, to clarify, there are 350 instances in the context-free subset, where they do not have ground truth outputs. These instances are designed to be open-ended and free-form generation. Also, our evaluation method InterleavedEval is a reference-free metric and does not require any ground truth. Hope this helps.
I see. Thank you for your answer!
IntJudge
changed discussion status to
closed