🛰️ Daily AI Frontier
‹ back to 2026-08-17

Every Expert Counts: ExactMoE for Memory-Efficient W4A16 Inference

Research Efficiency & Systems

Ranking

Overall 79
Content 95
Popularity 41

Observed public metrics from 1 member.

Representative image for Every Expert Counts: ExactMoE for Memory-Efficient W4A16 Inference

Merged summary

TL;DR - ExactMoE enables memory-efficient W4A16 inference for sparse MoE models by keeping quantized experts in host memory and caching routed experts on the GPU. It cuts peak GPU memory by 87.04% while preserving 81.85% of BF16 throughput and 99.23% of baseline accuracy.

  • Quantizes only expert weights to group-128 INT4; all non-expert components remain BF16.
  • Preserves every expert and the original top-k routing, without pruning, substitution, or CPU execution.
  • On one NVIDIA L4, a 16-slot cache uses 1.836 GiB versus 14.168 GiB for BF16.
  • A fully resident 64-slot setup reaches 31.923 tokens/s, exceeding BF16’s 21.662 tokens/s.

Sources (1)

Every Expert Counts: ExactMoE for Memory-Efficient W4A16 Inference

arXiv cs.LG Amjad Saab 2026-08-15 arXiv:2608.15383
Public signals Semantic Scholar citations 0 · Semantic Scholar influential citations 0
Providers: Hugging Face · N/A OpenAlex · N/A Publisher · N/A Semantic Scholar · Citations 0 · Influential citations 0 X · N/A Fetched 2026-09-14 14:22:26.152807 UTC

TL;DR - ExactMoE enables memory-efficient W4A16 inference for sparse MoE models by keeping quantized experts in host memory and caching routed experts on the GPU. It cuts peak GPU memory by 87.04% while preserving 81.85% of BF16 throughput and 99.23% of baseline accuracy.

  • Quantizes only expert weights to group-128 INT4; all non-expert components remain BF16.
  • Preserves every expert and the original top-k routing, without pruning, substitution, or CPU execution.
  • On one NVIDIA L4, a 16-slot cache uses 1.836 GiB versus 14.168 GiB for BF16.
  • A fully resident 64-slot setup reaches 31.923 tokens/s, exceeding BF16’s 21.662 tokens/s.
item →