Trait Linear 

pub trait Linear<B>: Send + Sync
where
    B: Backend,
{
    // Required methods
    fn in_features(&self) -> usize;
    fn out_features(&self) -> usize;
    fn forward(
        &self,
        ctx: &mut <B as Backend>::Context,
        input: &<B as Backend>::Buffer,
        out: &mut <B as Backend>::Buffer,
        m: usize,
    );
}

A weight-bearing linear projection.

forward computes out[m, out_features] = input[m, in_features] @ W^T. Implementations are responsible for calling the right backend kernel (B::gemm for dense, B::gemm_quant for quantized variants).

Required Methods

fn in_features(&self) -> usize

fn out_features(&self) -> usize

fn forward( &self, ctx: &mut <B as Backend>::Context, input: &<B as Backend>::Buffer, out: &mut <B as Backend>::Buffer, m: usize, )

Append GEMM work onto ctx. Caller flushes the context when results must be materialised.

Implementors

impl<B: Backend> Linear<B> for DenseLinear<B>

impl<B: Backend> Linear<B> for GgufLinear<B>

impl<B: Backend> Linear<B> for GptqLinear<B>

impl<B: Backend> Linear<B> for QuantLinear<B>