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 */
/* sys/mman.h for Windows (src/common/win/): only the advice, which Windows
has no use for; memory is mapped by code of its own there. */
#ifndef JANAS_WIN_SYS_MMAN_H
#define JANAS_WIN_SYS_MMAN_H
#include <stddef.h>
#define MADV_NORMAL 0
#define MADV_WILLNEED 3
#define MADV_DONTNEED 4
#define MADV_HUGEPAGE 14
static inline int madvise(void *addr, size_t len, int advice)
{
(void)addr;
(void)len;
(void)advice;
return 0;
}
#endif