llama-gguf 0.14.0

A high-performance Rust implementation of llama.cpp - LLM inference engine with full GGUF support
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

#version 450

// Matrix-vector multiply: out[i] = sum_j A[i,j] * x[j]
// A: [m, n] row-major, x: [n], out: [m]
// One thread per output row.

layout(local_size_x = 256) in;

layout(set = 0, binding = 0) readonly buffer InputA { float a[]; };
layout(set = 0, binding = 1) readonly buffer InputX { float x[]; };
layout(set = 0, binding = 2) writeonly buffer Output { float result[]; };

layout(push_constant) uniform Params {
    int m;  // rows of A / output dimension
    int n;  // cols of A / input dimension
};

void main() {
    uint i = gl_GlobalInvocationID.x;
    if (i >= uint(m)) return;

    float acc = 0.0;
    uint row_base = i * uint(n);
    for (int j = 0; j < n; j++) {
        acc += a[row_base + uint(j)] * x[j];
    }
    result[i] = acc;
}