Tay and colleagues: Synthesizer token-mixing alternatives

What changed

Compared learned and random synthetic alignment matrices with conventional query-key dot-product attention on language tasks.

Scientific scope

Compared learned and random synthetic alignment matrices with conventional query-key dot-product attention on language tasks. Competitive benchmark behavior does not make query-key attention unnecessary for every task, nor make every Synthesizer variant linear complexity.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Competitive benchmark behavior does not make query-key attention unnecessary for every task, nor make every Synthesizer variant linear complexity.

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.

• Synthesizer: Rethinking Self-Attention in Transformer Models
• Tay and colleagues: Synthesizer token-mixing alternatives — author or proceedings record

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