Parameters that shape how deterministic or varied model generation becomes.
Temperature, top-p, top-k, and related controls change how the next token is chosen from the model's probability distribution. They affect variation but cannot repair missing knowledge, weak instructions, or a mismatched model.