Editorial library
Each guide names its editor, review date, and primary sources. We publish procedures and evidence—not unrepeatable performance claims.
A source-backed guide to Microsoft’s BitNet embedding releases: choose a vector size, follow the official CPU setup, and validate retrieval before benchmarking.
A source-backed introduction to ternary weights, BitNet b1.58 research, and the difference between a model-training method and an inference runtime.
Follow the official bitnet.cpp build path with explicit version checks, a reproducible command log, and safe troubleshooting boundaries.
A reproducible benchmark protocol for bitnet.cpp that separates prompt processing from generation and publishes the context needed to interpret a result.
A practical framework for separating a paper’s result, a released model, a runtime benchmark, and a claim that still needs evidence.