BitNet — 1-Bit LLM Operations

Set up and run Microsoft's BitNet (1.58-bit ternary LLMs) for efficient CPU inference. Models use weights of {-1, 0, +1} — no GPU required.

Quick Start

# Full setup in 5 commands
./scripts/install-bitnet.sh
./scripts/download-model.sh microsoft/BitNet-b1.58-2B-4T-gguf
./scripts/build-bitnet.sh
./scripts/run-inference.sh -p "You are a helpful assistant" -cnv

Or manually: