Transformer architecture implemented in TypeScript with WebGPU.
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
nn/shader/dlds.wgsl

48 lines
1.2 KiB

struct Scalar {
size: vec4<u32>,
data: array<f32>,
}
struct MatrixBatch {
size: vec4<u32>,
data: array<f32>,
}
@group(0) @binding(0)
var<storage, read> ds: Scalar;
@group(0) @binding(1)
var<storage, read> wa: MatrixBatch;
@group(0) @binding(2)
var<storage, read> v: MatrixBatch;
@group(0) @binding(3)
var<storage, read> otdldo: MatrixBatch;
@group(0) @binding(4)
var<storage, read_write> dlds: MatrixBatch;
@compute @workgroup_size(1)
fn main(@builtin(global_invocation_id) global_id: vec3<u32>) {
dlds.size = wa.size;
var s = 0f;
var li = (global_id.z * v.size.y + global_id.x) * v.size.x;
var ri = global_id.z * otdldo.size.x * otdldo.size.y + global_id.y;
for (var i = 0u; i < v.size.x; i++) {
var cs = 0f;
var lj = global_id.z * wa.size.y * wa.size.x + global_id.y * wa.size.x;
var mj = global_id.z * v.size.x * v.size.y + i;
for (var j = 0u; j < wa.size.x; j++) {
cs += wa.data[lj] * v.data[mj];
lj++;
mj += v.size.x;
}
s += (v.data[li] - cs) * otdldo.data[ri];
li++;
ri += otdldo.size.x;
}
let resi = (global_id.z * dlds.size.y + global_id.y) * dlds.size.x + global_id.x;
dlds.data[resi] = ds.data[0] * wa.data[resi] * s;
}