ferrum-kernels 0.8.0

Unified compute kernels (CUDA/Metal/CPU) and model runner for Ferrum inference
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
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41

//! Unified Backend trait for CUDA, Metal, and CPU compute.
//!
//! Each backend implements the same set of transformer-layer primitives
//! (GEMM, norms, RoPE, attention, activations). `layer_forward()` and
//! `ModelRunner` are generic over `Backend`, so one forward path serves
//! all hardware targets.

mod traits;
pub use traits::*;

mod types;
pub use types::*;

mod capabilities;
mod native_status;

#[cfg(any(feature = "cuda", test))]
mod reusable_execution;

mod kv_layer;
pub use kv_layer::*;

pub mod dtype;
pub use dtype::{Dtype, HostDtype};

pub mod buffer;
pub use buffer::CpuBuf;
#[cfg(feature = "cuda")]
pub use buffer::CudaBuf;

pub mod cpu;

pub mod reference;

#[cfg(feature = "metal")]
pub mod metal;

#[cfg(feature = "cuda")]
pub mod cuda;

pub mod timer;