cortiq-engine 0.1.2

Portable, dependency-free inference runtime for the CMF model format: runs on CPU and GPU (Vulkan / Metal / DX12), with tokenizer, chat templates and dynamic per-skill weight overlay.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

//! Cortiq inference engine — sparse forward pass, attention, tokenization, sampling.

pub mod attention;
pub mod gpu;
#[cfg(target_os = "macos")]
pub mod gpu_metal;
#[cfg(feature = "gpu")]
pub mod gpu_wgpu;
pub mod inference;
pub mod kv_cache;
pub mod linear_core;
pub mod loader;
pub mod pipeline;
pub mod pool;
pub mod qtensor;
pub mod router;
pub mod runtime;
pub mod sampler;
pub mod swarm;
pub mod tokenizer;

pub use pipeline::{GenerateResult, Pipeline, TokenCallback, TokenTrace};
pub use runtime::CortiqRuntime;
pub use sampler::SamplerConfig;