EEP makes wide expert-parallel MoE serving survive single-rank failures with an 11s recovery pause, 8s reintegration pause, and throughput restored to 95% of pre-fault level within 52s while staying within 4.4% of a fixed-membership baseline in steady state.
Lazarus: Resilient and elastic training of mixture-of-experts models with adaptive expert placement
5 Pith papers cite this work. Polarity classification is still indexing.
5
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.DC 5verdicts
UNVERDICTED 5roles
background 1polarities
background 1representative citing papers
RLBoost harvests preemptible GPUs for RL rollout via a hybrid architecture with adaptive offload, pull-based transfer, and token-level migration, delivering 1.51x-1.97x throughput and 28-49% better cost efficiency than on-demand-only setups.
GRACE-MoE integrates expert grouping, dynamic replication, and locality-aware routing with hierarchical sparse communication to reduce end-to-end latency in distributed SMoE inference.
Production-scale empirical study of a 63-node 504-GPU cluster reports multi-signal failure detection needs, low checkpoint bandwidth utilization, heavy-tailed node exclusions, and 2.7x higher success for auto-retry chains.
Prism optimizes expert placement and uses runtime migration for distributed MoE inference on heterogeneous edge GPUs, achieving up to 30.6% lower latency than baselines.
citing papers explorer
-
Surviving Partial Rank Failures in Wide Expert-Parallel MoE Inference
EEP makes wide expert-parallel MoE serving survive single-rank failures with an 11s recovery pause, 8s reintegration pause, and throughput restored to 95% of pre-fault level within 52s while staying within 4.4% of a fixed-membership baseline in steady state.
-
RLBoost: Harvesting Preemptible Resources for Cost-Efficient Reinforcement Learning on LLMs
RLBoost harvests preemptible GPUs for RL rollout via a hybrid architecture with adaptive offload, pull-based transfer, and token-level migration, delivering 1.51x-1.97x throughput and 28-49% better cost efficiency than on-demand-only setups.
-
GRACE-MoE: Grouping and Replication with Locality-Aware Routing for Efficient Distributed MoE Inference
GRACE-MoE integrates expert grouping, dynamic replication, and locality-aware routing with hierarchical sparse communication to reduce end-to-end latency in distributed SMoE inference.
-
From Detection to Recovery: Operational Analysis on LLM Pre-training with 504 GPUs
Production-scale empirical study of a 63-node 504-GPU cluster reports multi-signal failure detection needs, low checkpoint bandwidth utilization, heavy-tailed node exclusions, and 2.7x higher success for auto-retry chains.
-
Accelerating Edge Inference for Distributed MoE Models with Latency-Optimized Expert Placement
Prism optimizes expert placement and uses runtime migration for distributed MoE inference on heterogeneous edge GPUs, achieving up to 30.6% lower latency than baselines.