When optimizing agent workflows for high-frequency operations, what specific metric must be monitored to identify the exact point where model distillation begins to degrade task consistency?

The specific metric that must be monitored is Kullback-Leibler divergence, commonly referred to as KL divergence. Model distillation is the process of training a smaller, faster student model to mimic the output behavior of a larger, more complex teacher model. Task consistency refers to the student model's ability to produce the same high-quality, reliable results as the teacher across repetitive high-frequency....