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 */
/*
* inflate.h - DEFLATE data (RFC 1951) inside its zlib wrapping (RFC 1950)
* made whole again (inflate.c), as PNG keeps its pixels.
*/
#ifndef JANAS_COMMON_INFLATE_H
#define JANAS_COMMON_INFLATE_H
#include <stddef.h>
#include <stdint.h>
/* The n bytes of zlib data at p into out, exactly cap bytes expected:
0, or -1 (data broken, or not cap bytes) with the reason in why. */
int janas_zlib_inflate(const uint8_t *p, size_t n, uint8_t *out, size_t cap,
char *why, size_t wn);
#endif