BWhat changed
Used row access probabilities and marginal memory/communication costs to combine replicated and row-sharded embeddings; a third strategy shards within a node while replicating across nodes to exploit network tiers.
Scientific scope
Used row access probabilities and marginal memory/communication costs to combine replicated and row-sharded embeddings; a third strategy shards within a node while replicating across nodes to exploit network tiers. Memory-neutrality and cost trade-offs are relative to the specified row-wise baseline and analytical distribution/topology model. Linearity of expected accessed rows does not require row independence. Nearly sixfold improvement concerns training communication latency, not universal whole-training throughput. Production datasets and selected 32-GPU topology constrain generalization.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Memory-neutrality and cost trade-offs are relative to the specified row-wise baseline and analytical distribution/topology model. Linearity of expected accessed rows does not require row independence. Nearly sixfold improvement concerns training communication latency, not universal whole-training throughput. Production datasets and selected 32-GPU topology constrain generalization.
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