A setting that boosts or blocks specific tokens before the AI picks its next word.
It is like a bubble tea order for AI. Extra pearls, please. No cilantro, ever.
Logit Bias pushes chosen words up or down. You can use it in an API to block words or force a set term. Push too hard, and the sentence sounds weird.
Token
Logit Bias directly changes how likely each Token is chosen.
Temperature
Logit Bias targets named words, while Temperature changes overall randomness.
API
Logit Bias is often sent as an API setting to control word choice.