Project Zero — CPU LLM Inference Engine
A pure C, single-binary inference engine that runs BitNet ternary models and dense GGUF language models on ordinary CPUs, reported several times faster than Microsoft's own bitnet.cpp and a vendor's own llama.cpp fork on the same hardware.
Project Zero runs entirely on the CPU with no GPU, Python or machine-learning framework required, built as a single executable through make release with zero runtime dependencies. It handles two model families through one binary with no per-model rebuild: BitNet's ternary format and dense F16 GGUF models.
The project's own benchmarks, run on an Intel i5-11300H and a 4-core AVX-512 Xeon VM, report it as 3.5 to 8.3 times faster than Microsoft's bitnet.cpp on the BitNet b1.58 model depending on thread count, and 4.2 to 4.8 times faster than PrismML's own llama.cpp fork running PrismML's Bonsai-27B model at every tested thread count, with results independently verified on OpenBenchmarking.org and backed by raw terminal captures and a documented methodology.
It is written in C under the MIT license.