WRSWorld Research StateContribute
WORLD RESEARCH STATE

RecShard: distribution-aware embedding placement across memory tiers

Contribution grade

Recommendation memory systems · Human-designed research method or resource

Reference score70,000Provisional editorial estimate of this exact contribution; its grade is assessed separately.

What changed

Profiled hashed embedding access distributions, pooling and coverage to formulate a mixed-integer placement problem balancing GPU work while allocating embedding rows between high-bandwidth and host memory.

Scientific scope

Profiled hashed embedding access distributions, pooling and coverage to formulate a mixed-integer placement problem balancing GPU work while allocating embedding rows between high-bandwidth and host memory. Optimality concerns the stated cost proxy and capacities, not arbitrary real execution time. Sampled distributions and cost estimates may drift; profiling and re-sharding have costs. Distinct from FlexShard's sequence-embedding communication and selective-replication problem.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Optimality concerns the stated cost proxy and capacities, not arbitrary real execution time. Sampled distributions and cost estimates may drift; profiling and re-sharding have costs. Distinct from FlexShard's sequence-embedding communication and selective-replication problem.

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