LLM Digest

Story

arxiv_cs_ai · Sep 4, 2026 · paper

Source brief

ACE: Adaptive Calibration-Free Expert Skipping for MoE-based LLMs

arxiv.orgSep 4, 2026
original source linked

In brief

Mixture-of-Experts (MoE) architectures provide an efficient paradigm for scaling large language models (LLMs), yet fixed top-k routing activates the same number of expert slots for every token, causing substantial red...

Feed lens

eval

Read the original at arxiv.org →Open in live feed Read that day’s brief

ACE: Adaptive Calibration-Free Expert Skipping for MoE-based LLMs

Earlier in this thread 4 items

PRIME: Mitigating Subgroup Optimization Competition in Shared CTR Top Networks with Plug-in Residual Input-Conditioned Mixture of Expert

Batch-wise Adaptive Pruning: Periodic Neuron Activation-Aware Weight Pruning for Language Reasoning Model

Adaptive Inference Batching using Policy Gradients

Adaptive Utility driven Resource Orchestration for Resilient AI (AURORA-AI)