A draft exhibit label is dull at a low randomness setting and wildly off-brand at a high one. What kind of controls are those?
Select an answer to reveal the explanation.
Short Explanation
Same label, two moods: dull at a low randomness setting, off-brand at a high one. Those are inference sampling knobs like temperature, not the training loss. Tree depth and a power cap do not pick the next token.
Full Explanation
Temperature and related sampling controls change how diverse the next-token choices are at inference. They are not the training loss, not tree depth, and not a power cap.