Activation Sparsity with Weight Approximation for Faster LLM Decoding on Offloaded Weights

arXiv:2610.02598 · cs.LG, cs.PF · Submitted 2026-10-01 · Read on arXiv

cs.LG, cs.PF

Submitted: 2026-10-01

Updated: 2026-10-01

Code: https://github.com/ggml-org/llama

Terminology

• Activation Sparsity — 1× in this paper · explained in 1 episode
• Sparsity — 1× in this paper · explained in 4 episodes

Sources

• Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge
• Training Verifiers to Solve Math Word Problems
• The Llama 3 Herd of Models
• Scaling Up On-Device LLMs via Active-Weight Swapping Between DRAM and Flash
• GRINQH: Graded Input-based Quantization Hierarchy for Efficient LLM Generation
• GLU Variants Improve Transformer
• Turbo Sparse: Achieving LLM SOTA Performance with Minimal Activated Parameters
• Enabling Dynamic Sparsity in Quantized LLM Inference
• PowerInfer-2: Fast Large Language Model Inference on a Smartphone

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks