ferrum-kernels 0.7.4

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

//! Concrete `Linear<B>` impls for quantized weights.
//!
//! Phase 3e moves the per-backend kernel-dispatch logic out of the
//! `BackendQuantMarlin` / `BackendQuantGguf` trait method bodies and
//! into these concrete `Linear<B>` types. Each impl owns the
//! cudarc / metal / cpu kernel call directly — no more `B::gemm_gptq`
//! indirection.
//!
//! Why these live in `ferrum-kernels` rather than `ferrum-quantization`:
//! the `forward()` body needs cudarc / metal-rs types, and pulling
//! those into ferrum-quantization would create a dep cycle (kernels
//! → quantization → kernels). ferrum-quantization stays as the
//! weight-format parser layer; backend-specific Linear impls live here.

pub mod cpu_dequant;
pub mod cpu_gguf;
pub mod cpu_marlin_stack;

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

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

#[cfg(all(target_os = "macos", feature = "metal"))]
pub mod metal_gguf;

#[cfg(all(target_os = "macos", feature = "metal"))]
pub mod metal_gguf_moe;