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 */
/*
* tz.h - times in a zone named as IANA names them ("Europe/Rome"), for
* Janas's services. POSIX: the C library's own, through TZ. Windows,
* whose C library knows no IANA names: the ICU that comes with Windows 10
* and 11 (icu.dll).
*/
#ifndef JANAS_SERVICES_TZ_H
#define JANAS_SERVICES_TZ_H
#include <stddef.h>
#include <time.h>
/* The time t in the zone tz (NULL or "": the machine's), into *out, and
the zone's abbreviation there and then ("CEST") into abbr when it is not
NULL. 0, or -1 (*out then in UTC). */
int janas_tz_local(const char *tz, time_t t, struct tm *out, char *abbr,
size_t abbr_len);
/* The instant of the local time *lt in the zone tz (NULL or "": the
machine's), as mktime would give it there; *lt normalized as mktime
leaves it. -1 when it cannot be told. */
time_t janas_tz_mktime(const char *tz, struct tm *lt);
/* The machine's zone as an IANA name ("Europe/Rome"). 0, or -1 when it is
not known. */
int janas_tz_machine(char *buf, size_t len);
#endif