AWhat changed
Extended the binary-weight design with absolute-mean ternary quantization into negative, zero and positive values, paired with an adapted Transformer training recipe and eight-bit activations.
Scientific scope
Extended the binary-weight design with absolute-mean ternary quantization into negative, zero and positive values, paired with an adapted Transformer training recipe and eight-bit activations. The 1.58-bit description concerns ternary weight information, not the precision of the complete training system or every stored tensor. Matching full-precision quality is an empirical tested regime, not all model sizes/tasks; original BitNet and earlier ternary networks remain prior inputs.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. The 1.58-bit description concerns ternary weight information, not the precision of the complete training system or every stored tensor. Matching full-precision quality is an empirical tested regime, not all model sizes/tasks; original BitNet and earlier ternary networks remain prior inputs.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
- The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits
- Ma et al.: BitNet b1.58 ternary-weight language-model recipe — official project source
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction