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 */
/*
* conf.h - a service's settings file (conf.c): lines "name = value", a
* line starting with # a comment. A line naming no setting the service
* knows is not left aside in silence: it is told, with the name nearest
* to it, so that a key written "aviationstak_key" does not leave the
* service keyless without a word.
*/
#ifndef JANAS_SERVICES_CONF_H
#define JANAS_SERVICES_CONF_H
#include <stddef.h>
/*
* The value of name in the file path, into val (cap bytes); names[0..n)
* are all the settings the file may hold. 1 with the value, 0 when the
* file or the line is missing. The first line naming none of them is told
* in why ("flights.conf, line 2: aviationstak_key is not a setting (did
* you mean aviationstack_key?)"), which is "" otherwise.
*/
int janas_conf_get(const char *path, const char *const *names, size_t n,
const char *name, char *val, size_t cap, char *why,
size_t why_len);
#endif