alice-llm 2.0.0

Pure Rust LLM inference engine — GGUF v3 parser, K-quants, hybrid CPU/GPU DeltaNet+Attention, wgpu compute shaders, speculative decoding, OpenAI-compatible HTTP server.
Documentation 
1
2
3
4
5
6
7
8
9
10
11

// Residual connection: acc_buf[i] += addend[i]

@group(0) @binding(0) var<storage, read_write> acc_buf: array<f32>;
@group(0) @binding(1) var<storage, read> addend: array<f32>;

@compute @workgroup_size(256)
fn residual_add(@builtin(global_invocation_id) gid: vec3<u32>) {
    let i = gid.x;
    if (i >= arrayLength(&acc_buf)) { return; }
    acc_buf[i] = acc_buf[i] + addend[i];
}