max@k
An optimization objective referring to the reward or score of the best sample among k samples, which encourages the model to retain diverse high-reward paths.
Timeline (last 90 days)
Repeated-sampling objectives (such as pass@k and max@k) can mitigate sharpening, but the underlying mechanism has long lacked a clear explanation.
This study derives closed-form optimal policies for the pass@k and max@k objectives under KL regularization, showing that @k optimization is essentially redistributing learning pressure according to problem difficulty, shifting the optimization focus from easy problems to hard ones.
All Facts (2)
Repeated-sampling objectives (such as pass@k and max@k) can mitigate sharpening, but the underlying mechanism has long lacked a clear explanation.
50%UnverifiedThis study derives closed-form optimal policies for the pass@k and max@k objectives under KL regularization, showing that @k optimization is essentially redistributing learning pressure according to problem difficulty, shifting the optimization focus from easy problems to hard ones.
50%