Does the data used for training and evaluation necessarily require ground truth? I noticed REFLECTOR_PROMPT_NO_GT in reflector.py—does this mean the framework also supports data without ground truth? However, the is_correct check in the code still relies on the ground truth target. Under what specific scenarios would REFLECTOR_PROMPT_NO_GT actually be used?
Thanks
Does the data used for training and evaluation necessarily require ground truth? I noticed REFLECTOR_PROMPT_NO_GT in reflector.py—does this mean the framework also supports data without ground truth? However, the is_correct check in the code still relies on the ground truth target. Under what specific scenarios would REFLECTOR_PROMPT_NO_GT actually be used?
Thanks