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 2026-09-27: a .gitignore for clones (build output, models, the tools' work)
camauri51e876213fb2
diff --git a/.gitignore b/.gitignorenew file mode 100644index 0000000..613b625--- /dev/null+++ b/.gitignore@@ -0,0 +1,27 @@+# What building Janas and using its tools leave in a clone+/bin/+/lib/+*.o+*.a+*.d+*.so+*.exe+*.dll+*.dylib++# Models and weights: never in the repository+/models/++# The work of tools/janas-try.sh, and janas-bench's reports+/janas-try/+janas-bench-*.txt++# Editors and OS+*~+*.swp+.DS_Store+Thumbs.db+desktop.ini+__pycache__/+*.pyc+.venv/