Sampling

Scales logits before sampling; higher increases randomness/diversity, lower increases determinism.

Samples from the k highest-probability tokens to limit unlikely outputs.

Samples from the smallest set of tokens whose probabilities sum to p, adapting set size by context.

Raw model outputs before converting to probabilities; manipulated during decoding and calibration.

Reducing numeric precision of weights/activations to speed inference and reduce memory with acceptable accuracy loss.

Removing weights or neurons to shrink models and improve efficiency; can be structured or unstructured.

Training a smaller “student” model to mimic a larger “teacher,” often improving efficiency while retaining performance.

Search algorithm for generation that keeps top-k partial sequences; can improve likelihood but reduce diversity.