AWhat changed
Combined compressed-block context, selected fine-grained blocks and local attention in a trainable hierarchical sparse design, co-designed with GPU-friendly kernels for forward, backward and decoding workloads.
Scientific scope
Combined compressed-block context, selected fine-grained blocks and local attention in a trainable hierarchical sparse design, co-designed with GPU-friendly kernels for forward, backward and decoding workloads. Approximate attention architecture, not distribution-identical full attention. Speed/quality results depend on hardware, sequence length and sparsity settings. Distinct from later DSA lightning indexing; do not attribute DSA implementation to this earlier three-branch method.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Approximate attention architecture, not distribution-identical full attention. Speed/quality results depend on hardware, sequence length and sparsity settings. Distinct from later DSA lightning indexing; do not attribute DSA implementation to this earlier three-branch method.
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.
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