Top‑k Sampling

Sampling restricted to the k most‑likely tokens at each step, trading off creativity and risk of nonsense for control and stability.

Top‑p (Nucleus Sampling)

Sampling from the smallest set of tokens whose cumulative probability exceeds p (e.g., 0.9), balancing diversity with coherence in generation.

Temperature (Sampling)

Decoding parameter that sharpens or flattens the probability distribution over next tokens; lower values are more deterministic, higher values more diverse.

Tokenization (AI)

Process of splitting text into tokens using schemes like BPE or WordPiece; affects vocabulary, context use, and cost.

Token (LLM Token)

Smallest unit of text a model processes, roughly a word piece; context limits are measured in tokens and billed per token in many APIs.

Transformer

Neural network architecture built on self‑attention and feed‑forward layers that models long‑range dependencies efficiently; foundation of modern LLMs.

Zigbee (Mesh Protocol)

Low-power wireless mesh networking standard used in IoT. Relevant to oracle hardware, sensor networks, and trusted data collection at the edge.

Zsh (Z Shell)

Unix shell with advanced completion and scripting features, widely used by developers. Pairs with tools like Oh My Zsh for productivity on blockchain ops boxes.

Zarr (Chunked Array Format)

Format for storing large N-dimensional arrays in chunks across filesystems or object storage; handy for ML datasets and on-chain analytics features at scale.

zkey (SNARK Proving Key File)

File format used by tools like snarkjs containing proving and verification material derived from a trusted setup ceremony for zk-SNARK circuits. Handle with care.

ZK‑Friendly Hash Function

Hash primitives designed for efficient use inside ZK circuits, e.g., Poseidon or Rescue, minimizing arithmetic constraints for Merkle trees and commitments.