AWhat changed
Used a faster draft model and parallel target-model scoring with corrected acceptance sampling to produce multiple tokens per target call while preserving its sampling distribution.
Scientific scope
Used a faster draft model and parallel target-model scoring with corrected acceptance sampling to produce multiple tokens per target call while preserving its sampling distribution. Two independently developed papers share the core sampler and are counted once. The nine authors belong to two distinct bylines. Gains depend on acceptance, hardware and workloads; floating-point numerics still matter.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Two independently developed papers share the core sampler and are counted once. The nine authors belong to two distinct bylines. Gains depend on acceptance, hardware and workloads; floating-point numerics still matter.
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.
- Fast Inference from Transformers via Speculative Decoding
- Accelerating Large Language Model Decoding with Speculative Sampling
- Google Research: Fast Inference from Transformers via Speculative Execution
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