GeMoE adaptively sets the number of experts per token via gating entropy, retaining 99.5% of static-routing performance while raising average sparsity by 36.5%.
arXiv preprint arXiv:2312.17238 (2023)
13 Pith papers cite this work. Polarity classification is still indexing.
13
Pith papers citing it
citation-role summary
background 2
citation-polarity summary
roles
background 2polarities
background 2representative citing papers
On bandwidth-bound edge hardware, MoE inference cost tracks total parameters rather than active ones, so sparse activation fails to deliver the expected throughput or energy gains.
WiSP achieves up to 1.95x decode throughput on low-resource MoE serving by dynamically paging reused experts and using MV-WSA to allocate VRAM between experts and KV cache, with the offline policy performing well on both prefill and decode.
PALS adds dynamic GPU power capping to LLM serving frameworks like vLLM, jointly tuning it with batch size via offline models and feedback control to improve energy efficiency up to 26.3% and cut QoS violations 4-7x on dense and MoE models.
VisMMoE exploits visual-expert affinity via token pruning to achieve up to 2.68x faster VL-MoE inference on memory-constrained hardware while keeping accuracy competitive.
Profiling shows persistent expert load imbalance and domain-specific activation patterns in large MoE models; workload-aware grouping and placement reduce all-to-all communication volume by up to 20x.
Temporally extended MoE layers using the option-critic framework with deliberation costs cut switching rates below 5% while retaining most capability on MATH, MMLU, and MMMLU.
Comprehensive profiling of expert selection in frontier MoE models reveals temporal and spatial patterns that enable 6.6x speedup on wafer-scale GPUs and 1.25x on existing systems via targeted optimizations.
Lynx exploits training-induced batch-level expert activation skews via AffinityBinning to reduce invoked experts per batch, delivering up to 1.30x throughput with under 1% accuracy loss across four model families.
CAEE reduces MoE inference latency 8-18% on 671B DeepSeek-R1 by cost-aware expert pruning and low-overhead compensation while keeping accuracy drop under 1%.
TIDE schedules I/O-aware expert offloading for MoE diffusion LLMs by solving for an optimal refresh interval that exploits temporal stability of activations, yielding up to 1.5x throughput gain losslessly.
ZipMoE delivers up to 72.77% lower inference latency and 6.76x higher throughput for on-device MoE models via lossless compression and cache-affinity scheduling with a claimed provable guarantee.
Mixtral 8x7B is a sparse MoE LLM activating 2 of 8 experts per layer that matches or exceeds Llama 2 70B and GPT-3.5 on benchmarks while using only 13B active parameters.
citing papers explorer
-
GeMoE: Gating Entropy is All You Need for Uncertainty-aware Adaptive Routing in MoE-based Large Vision-Language Models
GeMoE adaptively sets the number of experts per token via gating entropy, retaining 99.5% of static-routing performance while raising average sparsity by 36.5%.
-
Does Mixture-of-Experts Actually Help Inference on Consumer and Edge Hardware? An Empirical Study
On bandwidth-bound edge hardware, MoE inference cost tracks total parameters rather than active ones, so sparse activation fails to deliver the expected throughput or energy gains.
-
WiSP: A Working-Set View of Mixture-of-Experts Serving on Extremely Low-Resource Hardware
WiSP achieves up to 1.95x decode throughput on low-resource MoE serving by dynamically paging reused experts and using MV-WSA to allocate VRAM between experts and KV cache, with the offline policy performing well on both prefill and decode.
-
PALS: Power-Aware LLM Serving for Mixture-of-Experts Models
PALS adds dynamic GPU power capping to LLM serving frameworks like vLLM, jointly tuning it with batch size via offline models and feedback control to improve energy efficiency up to 26.3% and cut QoS violations 4-7x on dense and MoE models.
-
VisMMOE: Exploiting Visual-Expert Affinity for Efficient Visual-Language MoE Offloading
VisMMoE exploits visual-expert affinity via token pruning to achieve up to 2.68x faster VL-MoE inference on memory-constrained hardware while keeping accuracy competitive.
-
Scaling Multi-Node Mixture-of-Experts Inference Using Expert Activation Patterns
Profiling shows persistent expert load imbalance and domain-specific activation patterns in large MoE models; workload-aware grouping and placement reduce all-to-all communication volume by up to 20x.
-
Temporally Extended Mixture-of-Experts Models
Temporally extended MoE layers using the option-critic framework with deliberation costs cut switching rates below 5% while retaining most capability on MATH, MMLU, and MMMLU.
-
Patterns behind Chaos: Forecasting Data Movement for Efficient Large-Scale MoE LLM Inference
Comprehensive profiling of expert selection in frontier MoE models reveals temporal and spatial patterns that enable 6.6x speedup on wafer-scale GPUs and 1.25x on existing systems via targeted optimizations.
-
Lynx: Enabling Efficient MoE Inference through Dynamic Batch-Aware Expert Selection
Lynx exploits training-induced batch-level expert activation skews via AffinityBinning to reduce invoked experts per batch, delivering up to 1.30x throughput with under 1% accuracy loss across four model families.
-
Beyond Uniform Experts: Cost-Aware Expert Execution for Efficient Multi-Device MoE Inference
CAEE reduces MoE inference latency 8-18% on 671B DeepSeek-R1 by cost-aware expert pruning and low-overhead compensation while keeping accuracy drop under 1%.
-
TIDE: Efficient and Lossless MoE Diffusion LLM Inference with I/O-aware Expert Offload
TIDE schedules I/O-aware expert offloading for MoE diffusion LLMs by solving for an optimal refresh interval that exploits temporal stability of activations, yielding up to 1.5x throughput gain losslessly.
-
ZipMoE: Efficient On-Device MoE Serving via Lossless Compression and Cache-Affinity Scheduling
ZipMoE delivers up to 72.77% lower inference latency and 6.76x higher throughput for on-device MoE models via lossless compression and cache-affinity scheduling with a claimed provable guarantee.
-
Mixtral of Experts
Mixtral 8x7B is a sparse MoE LLM activating 2 of 8 experts per layer that matches or exceeds Llama 2 70B and GPT-3.5 on benchmarks while using only 13B active parameters.