Dettmers and colleagues: mixed-precision LLM.int8 inference

What changed

Combined vector-wise eight-bit quantization with higher-precision treatment of outlier feature dimensions for large Transformer inference.

Scientific scope

Combined vector-wise eight-bit quantization with higher-precision treatment of outlier feature dimensions for large Transformer inference. The mixed-precision method does not compute every value in eight bits. Model-quality preservation and memory savings are evaluated findings, not universal accuracy or speed guarantees.

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 mixed-precision method does not compute every value in eight bits. Model-quality preservation and memory savings are evaluated findings, not universal accuracy or speed guarantees.

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.

• LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale

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