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
Release 2026-10-11 (4): the tests run on a fresh machine
camauri2e7e8cf76039
diff --git a/tests/test_github_parse.c b/tests/test_github_parse.cindex 286bc07..8db9660 100644--- a/tests/test_github_parse.c+++ b/tests/test_github_parse.c@@ -296,6 +296,9 @@ static void seen(void){char dir[512];const char *home = getenv("HOME");+ /* ~/tmp first: a fresh machine (GitHub's runner) has none */+ snprintf(dir, sizeof dir, "%s/tmp", home ? home : ".");+ mkdir(dir, 0700);snprintf(dir, sizeof dir, "%s/tmp/janas-github-test-%d", home ? home : ".",(int)getpid());if (mkdir(dir, 0700) != 0) {diff --git a/tests/test_layout.c b/tests/test_layout.cindex 4059a0a..941764a 100644--- a/tests/test_layout.c+++ b/tests/test_layout.c@@ -100,6 +100,9 @@ int main(void)char dir[512];const char *home = getenv("HOME");+ /* ~/tmp first: a fresh machine (GitHub's runner) has none */+ snprintf(dir, sizeof dir, "%s/tmp", home ? home : ".");+ mkdir(dir, 0700);snprintf(dir, sizeof dir, "%s/tmp/janas-layout-test-%d", home ? home : ".",(int)getpid());if (mkdir(dir, 0700) != 0) {