Crate alice_llm 

ALICE-LLM: Pure Rust LLM inference engine.

Re-exports

pub use matrix::dot_flat;

pub use matrix::Matrix;

pub use crate::attention::*;

pub use crate::batch_inference::*;

pub use crate::kv_cache::*;

pub use crate::linalg::*;

pub use crate::quantization::*;

pub use crate::rope::*;

pub use crate::sampling::*;

pub use crate::tokenizer::*;

Modules

attention

attention.

batch_inference

batch inference.

deepseek_streaming

DeepSeek-V3 / Kimi K3-family routed-expert streaming from disk (Phase 4a, Issue #34; extended for Kimi K3 at Phase X.4.e).

gguf

GGUF v3 format parser with fused quantized matvec.

iq1_s

kv_cache

kv cache.

linalg

linalg.

llama3

Multi-architecture LLM inference engine with GGUF quantized weights.

matrix

Flat cache-aligned matrix (SoA row-major layout).

prelude

Prelude re-exports.

quantization

quantization.

rope

rope.

sampling

sampling.

sparse_attention

sparse attention (KV-outer block-sparse).

speculative_dspark

DSpark speculative decoding primitives.

tokenizer

tokenizer.

training

Ternary Quantization-Aware Training (QAT).