#![allow(clippy::doc_markdown)]

//! EveryCycle inference runtime.
//!
//! Loads quantized models, places layers across heterogeneous compute (GPU,
//! CPU, NVMe overflow), drives the forward pass, and streams tokens to the
//! scheduler.
//!
//! See `docs/architecture.md` in the workspace root for the broader sketch.

pub mod executor;

pub use executor::{
    ActivationHandle, BoundDevice, Dtype, Executor, ExecutorError, PortableActivation,
    SubGraphPlan, TensorShape, Vendor,
};