Crate llama_sampling 

llama-sampling

Sampling and decoding strategies for llama.rs. Implements greedy, temperature, top-k, top-p, and repetition penalty sampling with deterministic seeded RNG for reproducible test runs.

Structs

Sampler

Stateful sampler using deterministic RNG.

SamplingConfig

Configuration for token sampling.

Enums

SamplingError

Error type for sampling operations.

SamplingStrategy

Sampling strategy.

Functions

apply_repetition_penalty

Apply repetition penalty to logits for tokens present in history.

greedy_sample

Greedy decoding (argmax).