37 tok/s from Qwen3-Next-80B (48 GB) on a 32 GB laptop, 24 without drafts vs 2.8 for llama.cpp; reading 62 vs 16 tok/s. Faster than llama.cpp on all 13 models compared in Q4_K_M. LLMs bigger than your RAM, pure C, CPU or GPU.
C avx2ccpu-inferenceggufinference-enginelinuxllmlocal-llmmixture-of-expertsmoenvmequantizationqwen3simdspeculative-decodingvulkan
// SPDX-License-Identifier: GPL-3.0-or-later
// Copyright (C) 2026 Maurizio Cammalleri
//
// One reply at a time, in the order asked: the model holds one
// conversation. The engine keeps its own order too; the line here makes a
// request's parameters and its reply one step, and says how many wait.
export class Line {
private tail: Promise<void> = Promise.resolve();
private count = 0;
/** requests waiting or running */
get length(): number {
return this.count;
}
async run<T>(job: () => Promise<T>): Promise<T> {
this.count++;
const before = this.tail;
let done!: () => void;
this.tail = new Promise<void>((resolve) => (done = resolve));
try {
await before;
return await job();
} finally {
this.count--;
done();
}
}
}