Peng and colleagues: RWKV parallel training and recurrent sequence inference

What changed

Used receptance-weighted key-value computation to combine parallelizable sequence training with a recurrent decoding state.

Scientific scope

Used receptance-weighted key-value computation to combine parallelizable sequence training with a recurrent decoding state. Fixed per-token state cost assumes fixed model dimensions; total sequence work still grows with length. Benchmarked parity is not a universal replacement theorem for Transformers.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Fixed per-token state cost assumes fixed model dimensions; total sequence work still grows with length. Benchmarked parity is not a universal replacement theorem for Transformers.

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.

• RWKV: Reinventing RNNs for the Transformer Era

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