Skip to main content

Crate ik_llama_cpp_2

Crate ik_llama_cpp_2 

Source
Expand description

Safe Rust bindings for ik_llama.cpp (ikawrakow’s SOTA-quant fork).

Mirrors the API and codestyle of the llama-cpp-2 crate (utilityai/llama-cpp-rs), adapted for ik_llama.cpp’s divergent C API: legacy llama_sample_* sampling, model-pointer tokenizer, ik-specific model/context params, and the MTP / NextN speculative path.

v1 covers the general generation path (load → tokenize → decode → sample) plus the MTP scaffolding.

Re-exports§

pub use context::params::LlamaContextParams;
pub use context::params::LlamaContextType;
pub use context::LlamaContext;
pub use grammar::json_schema_to_grammar;
pub use grammar::JsonSchemaError;
pub use grammar::DryInitError;
pub use grammar::DryParams;
pub use grammar::GrammarInitError;
pub use grammar::LlamaDrySampler;
pub use grammar::LlamaGrammar;
pub use llama_backend::LlamaBackend;
pub use llama_batch::LlamaBatch;
pub use model::params::LlamaModelParams;
pub use model::AddBos;
pub use model::LlamaModel;
pub use sampling::LlamaSampler;
pub use speculative::MtpOpType;
pub use speculative::MtpSpeculativeParams;
pub use speculative::MtpSpeculative;
pub use speculative::MtpStep;
pub use token::data::LlamaTokenData;
pub use token::data_array::LlamaTokenDataArray;
pub use token::LlamaToken;

Modules§

context
Safe wrapper around llama_context (LlamaContext).
gguf
Safe wrapper around gguf_context for reading GGUF file metadata.
grammar
Stateful GBNF grammar + DRY samplers.
llama_backend
Process-wide ik_llama.cpp backend initialization.
llama_batch
Token batch for decode (LlamaBatch) over ik’s llama_batch.
model
Safe wrapper around llama_model (LlamaModel).
quantize
Model quantization write-path over ik’s llama_model_quantize.
sampling
Sampling — a chain-object LlamaSampler matching llama-cpp-2’s API, emulated over ik_llama.cpp’s legacy llama_sample_* array functions.
speculative
Multi-Token Prediction (MTP / NextN) support.
timing
Safe wrapper around llama_timings.
token
Token newtype.

Enums§

LlamaError
Errors returned by the safe wrapper.