Recommended Size of Match and No Match Pairs for Large Dataset Before Training Phase

Hello everyone what is the recommended size of Match and No Match pairs you recommend for big dataset before moving to train phase