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 */
/*
* md5.h - MD5 (RFC 1321), only to check an upload against the checksum a
* client sends with it, as OpenAI's Uploads API asks: not for security.
*/
#ifndef JANAS_COMMON_MD5_H
#define JANAS_COMMON_MD5_H
#include <stddef.h>
#include <stdint.h>
struct janas_md5 {
uint32_t h[4];
uint64_t len; /* bytes hashed */
uint8_t buf[64];
size_t n; /* bytes waiting in buf */
};
void janas_md5_init(struct janas_md5 *s);
void janas_md5_update(struct janas_md5 *s, const void *data, size_t n);
/* The digest as 32 lowercase hex digits and a NUL. */
void janas_md5_hex(struct janas_md5 *s, char out[33]);
#endif