WRSWorld Research StateContribute
WORLD RESEARCH STATE

GLM-5.3-Flash hybrid sparse–linear attention with IndexPool

Contribution grade

Efficient multimodal foundation-model systems · Human-designed research method or resource

Reference score50,000Provisional editorial estimate of this exact contribution; its grade is assessed separately.

What changed

Combines linear and indexed sparse attention in a new multimodal base, pooling four indexer keys into one with IndexPool to reduce long-context indexing cost.

Scientific scope

Combines linear and indexed sparse attention in a new multimodal base, pooling four indexer keys into one with IndexPool to reduce long-context indexing cost. System-level integration, not invention of linear/sparse attention or mHC. Public configuration has45 layers with34 linear and11 sparse layers; does not establish the complete training algorithm. Team-only attribution: predecessor GLM-5 report contributors cannot be copied as Flash inventors. Claimed cost reductions normalize attention per head/layer, not all workload latency. No benchmark superiority or full production-stack reproduction certified.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. System-level integration, not invention of linear/sparse attention or mHC. Public configuration has45 layers with34 linear and11 sparse layers; does not establish the complete training algorithm. Team-only attribution: predecessor GLM-5 report contributors cannot be copied as Flash inventors. Claimed cost reductions normalize attention per head/layer, not all workload latency. No benchmark superiority or full production-stack reproduction certified.

Contributors

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