Skip to main content

Crate llama_sampling

Crate llama_sampling 

Source
Expand description

§llama-sampling

Sampling and decoding strategies for llama.rs.

Supports:

  • Greedy (argmax)
  • Temperature scaling
  • Top-k filtering
  • Top-p (nucleus) filtering
  • Repetition penalty
  • Deterministic seeded RNG for reproducible generation

Structs§

Sampler
Sampling configuration and strategy.
SeededRng
Deterministic RNG for reproducible sampling.

Enums§

SamplingError
Sampling error type.

Type Aliases§

SamplingResult