specific method · filed under optimization
Data-Weighted Data Parallelism
A data-parallel method that keeps execution data-parallel and pulls each layer's expert weights to every rank.
Also called DWDP.
- source
- 1
- model
- 1
- labs adopt it
- 0
- strongest
- mentioned
How sources treat it
One count per evidence span, weakest treatment to strongest.
mentioned 1
Documented in
Evidence
1 span quoted from the sources, strongest treatment first.
DWDP (nvidia2026dwdp) keeps execution data-parallel and instead pulls each layer's expert weights to every rank
mentionedinference servingin Nemotron 3 UltraNVIDIA
Filed alongside
Other methods under optimization :: training parallelism.
Expert ParallelismTensor ParallelismMoonEPCommunication-Computation OverlapPipeline ParallelismAll-to-all Gradient Exchange with Local FP32 SummationCache-Based Pipeline CommunicationContext ParallelismData Replica Reduction over the Data Center NetworkDynamic Context Parallelism for Large Multimodal SamplesFully Balanced Expert-Parallel TrainingGPU Planning Kernel for Redundant Expert MigrationHybrid ZeRO Bucket Assignment for MuonKDA Context ParallelismKnapsack-Based Balanced Assignment of Dense Parameter MatricesLoad-Balanced Image ShardingModified DualPipe 1F1B Pipeline Overlap for mHCNS-FLOP-Balanced Static Parameter PartitioningPipeline Payload ExtensionsPipeline ZeRO-2 Gradient Sharding with CPU Offloadingpipeline-bubble scheduling of ViT computationRedundant-Expert Capacity ReservationSConv-Aware Tensor-Parallel ShardingSequence Parallelism for Activations