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
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55

#version 450

// LayerNorm: out[i] = (x[i] - mean) / sqrt(var + eps) * weight[i] + bias[i]
// Single-workgroup approach for vectors up to ~16K elements.
layout(local_size_x = 256) in;

layout(set = 0, binding = 0) readonly buffer Input { float x[]; };
layout(set = 0, binding = 1) readonly buffer Weight { float w[]; };
layout(set = 0, binding = 2) readonly buffer Bias { float b[]; };
layout(set = 0, binding = 3) writeonly buffer Output { float result[]; };

layout(push_constant) uniform Params {
    int n;
    float eps;
};

shared float sdata[256];

void main() {
    uint tid = gl_LocalInvocationID.x;

    // Pass 1: compute sum for mean
    float local_sum = 0.0;
    for (uint i = tid; i < n; i += 256) {
        local_sum += x[i];
    }
    sdata[tid] = local_sum;
    barrier();
    for (uint s = 128; s > 0; s >>= 1) {
        if (tid < s) sdata[tid] += sdata[tid + s];
        barrier();
    }
    float mean = sdata[0] / float(n);
    barrier();

    // Pass 2: compute variance
    float local_var = 0.0;
    for (uint i = tid; i < n; i += 256) {
        float d = x[i] - mean;
        local_var += d * d;
    }
    sdata[tid] = local_var;
    barrier();
    for (uint s = 128; s > 0; s >>= 1) {
        if (tid < s) sdata[tid] += sdata[tid + s];
        barrier();
    }
    float inv_std = inversesqrt(sdata[0] / float(n) + eps);

    // Pass 3: normalize and scale
    for (uint i = tid; i < n; i += 256) {
        float normed = (x[i] - mean) * inv_std;
        result[i] = normed * w[i] + b[i];
    }
}