On a real multi-node H100 cluster the authors show that for MLA, routing the ~1 KB compressed query row is cheaper than moving cache chunks and supply a topology-aware cost model accurate to ~7% on IBGDA fabrics.
Rethinking llm inference bottlenecks: Insights from latent attention and mixture-of-experts.arXiv preprint arXiv:2507.15465, 2025
6 Pith papers cite this work. Polarity classification is still indexing.
6
Pith papers citing it
representative citing papers
LLM serving should triage by five-resource analytical floors and wall ordering, not grid search; on 16×H20, TP16 is capacity-capped at ~70 while EP+DP attention reaches ~644 concurrent 8K requests.
Switchless topologies such as 3D full-mesh are 20.6-56.2% more cost-effective than scale-up networks for MoE LLM serving, with current link bandwidths over-provisioned by up to 27%.
JANUS disaggregates attention and MoE layers onto separate GPU pools with an expert-balancing scheduler and SLO-aware scaling, delivering up to 4.7x higher per-GPU throughput than prior MoE systems under token-level latency constraints.
Layered prefill replaces token-chunked prefill with layer-group interleaving in MoE models, cutting TTFT by up to 70%, end-to-end latency by 41%, and per-token energy by 22% while preserving stall-free TBT.
StableHLO serves as a viable unified representation for cross-architecture performance modeling of distributed ML workloads, preserving relative trends while exposing fidelity trade-offs.
citing papers explorer
-
Move the Query, Not the Cache: Characterizing Cross-Instance Latent Attention Redistribution Across GPU Fabrics
On a real multi-node H100 cluster the authors show that for MLA, routing the ~1 KB compressed query row is cheaper than moving cache chunks and supply a topology-aware cost model accurate to ~7% on IBGDA fabrics.
-
Think Before You Grid-Search: Floor-First Triage for LLM Serving
LLM serving should triage by five-resource analytical floors and wall ordering, not grid search; on 16×H20, TP16 is capacity-capped at ~70 while EP+DP attention reaches ~644 concurrent 8K requests.
-
Rethinking Network Topologies for Cost-Effective Mixture-of-Experts LLM Serving
Switchless topologies such as 3D full-mesh are 20.6-56.2% more cost-effective than scale-up networks for MoE LLM serving, with current link bandwidths over-provisioned by up to 27%.
-
Janus: Disaggregating Attention and Experts for Scalable MoE Inference
JANUS disaggregates attention and MoE layers onto separate GPU pools with an expert-balancing scheduler and SLO-aware scaling, delivering up to 4.7x higher per-GPU throughput than prior MoE systems under token-level latency constraints.
-
From Tokens to Layers: Redefining Stall-Free Scheduling for MoE Serving with Layered Prefill
Layered prefill replaces token-chunked prefill with layer-group interleaving in MoE models, cutting TTFT by up to 70%, end-to-end latency by 41%, and per-token energy by 22% while preserving stall-free TBT.
-
Evaluating Cross-Architecture Performance Modeling of Distributed ML Workloads Using StableHLO
StableHLO serves as a viable unified representation for cross-architecture performance modeling of distributed ML workloads, preserving relative trends while exposing fidelity trade-offs.