GLOSSARY TERM
What is Temperature Scaling?
A technique adjusting the softmax logits to control model confidence or generative randomness.
Temperature scaling divides the logits by a constant before applying softmax. A high temperature flattens the distribution for diverse outputs, while a low temperature sharpens it, making the model highly deterministic and confident.
Calibrate Model Confidence
Adjust and monitor post-training calibrations effortlessly within MLOps.