# Efficient LLM Inference using Dynamic Input Pruning and Cache-Aware Masking

arXiv: https://arxiv.org/abs/2412.01380 · HTML: https://mlsanity.com/paper/2412.01380/

## Key facts

- Cited by 11 papers in the map (data through 14 Oct 2026).
- Placed in Serving systems and request scheduling, one of 12 directions in Efficient LLM inference and serving.
- Builds on 31 of its 48 references that are in the map.
- First posted on arXiv 2 Dec 2024.

## Placed in

- Efficient LLM inference and serving / [Serving systems and request scheduling](https://mlsanity.com/inference/serving-systems-and-request-scheduling/)

## Builds on (most cited first, 30 of 31)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [OPT: Open Pre-trained Transformer Language Models](https://mlsanity.com/paper/2205.01068/) (2022)
- [Qwen Technical Report](https://mlsanity.com/paper/2309.16609/) (2023)
- [Mistral 7B](https://mlsanity.com/paper/2310.06825/) (2023)
- [A Survey of Large Language Models](https://mlsanity.com/paper/2303.18223/) (2023)
- [GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers](https://mlsanity.com/paper/2210.17323/) (2022)
- [Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone](https://mlsanity.com/paper/2404.14219/) (2024)
- [MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark](https://mlsanity.com/paper/2406.01574/) (2024)
- [SparseGPT: Massive Language Models Can Be Accurately Pruned in One-Shot](https://mlsanity.com/paper/2301.00774/) (2023)
- [Sigmoid-Weighted Linear Units for Neural Network Function Approximation in Reinforcement Learning](https://mlsanity.com/paper/1702.03118/) (2017)
- [A Simple and Effective Pruning Approach for Large Language Models](https://mlsanity.com/paper/2306.11695/) (2023)
- [Language Modeling with Gated Convolutional Networks](https://mlsanity.com/paper/1612.08083/) (2016)
- [LLM-Pruner: On the Structural Pruning of Large Language Models](https://mlsanity.com/paper/2305.11627/) (2023)
- [A Comprehensive Overview of Large Language Models](https://mlsanity.com/paper/2307.06435/) (2023)
- [A White Paper on Neural Network Quantization](https://mlsanity.com/paper/2106.08295/) (2021)
- [PowerInfer: Fast Large Language Model Serving with a Consumer-grade GPU](https://mlsanity.com/paper/2312.12456/) (2023)
- [Competitive caching with machine learned advice](https://mlsanity.com/paper/1802.05399/) (2018)
- [LLM in a flash: Efficient Large Language Model Inference with Limited Memory](https://mlsanity.com/paper/2312.11514/) (2023)
- [BiLLM: Pushing the Limit of Post-Training Quantization for LLMs](https://mlsanity.com/paper/2402.04291/) (2024)
- [PowerInfer-2: Fast Large Language Model Inference on a Smartphone](https://mlsanity.com/paper/2406.06282/) (2024)
- [ReLU Strikes Back: Exploiting Activation Sparsity in Large Language Models](https://mlsanity.com/paper/2310.04564/) (2023)
- [AI and Memory Wall](https://mlsanity.com/paper/2403.14123/) (2024)
- [GPTVQ: The Blessing of Dimensionality for LLM Quantization](https://mlsanity.com/paper/2402.15319/) (2024)
- [Pruning vs Quantization: Which is Better?](https://mlsanity.com/paper/2307.02973/) (2023)
- [ProSparse: Introducing and Enhancing Intrinsic Activation Sparsity within Large Language Models](https://mlsanity.com/paper/2402.13516/) (2024)
- [CATS: Contextually-Aware Thresholding for Sparsity in Large Language Models](https://mlsanity.com/paper/2404.08763/) (2024)
- [Turbo Sparse: Achieving LLM SOTA Performance with Minimal Activated Parameters](https://mlsanity.com/paper/2406.05955/) (2024)
- [OATS: Outlier-Aware Pruning Through Sparse and Low Rank Decomposition](https://mlsanity.com/paper/2409.13652/) (2024)

## Built on by (most cited first, 11 of 11)

- [DocPruner: A Storage-Efficient Framework for Multi-Vector Visual Document Retrieval via Adaptive Patch-Level Embedding Pruning](https://mlsanity.com/paper/2509.23883/) (2025)
- [AutoNeural: Co-Designing Vision-Language Models for NPU Inference](https://mlsanity.com/paper/2512.02924/) (2025)
- [Memory-Augmented Transformers: A Systematic Review from Neuroscience Principles to Enhanced Model Architectures](https://mlsanity.com/paper/2508.10824/) (2025)
- [HELIOS: Adaptive Model And Early-Exit Selection for Efficient LLM Inference Serving](https://mlsanity.com/paper/2504.10724/) (2025)
- [Universal Properties of Activation Sparsity in Modern Large Language Models](https://mlsanity.com/paper/2509.00454/) (2025)
- [Dynamic Tool Dependency Retrieval for Lightweight Function Calling](https://mlsanity.com/paper/2512.17052/) (2025)
- [RAP: Runtime Adaptive Pruning for LLM Inference](https://mlsanity.com/paper/2505.17138/) (2025)
- [When Cloud Agents Meet Device Agents: Lessons from Hybrid Multi-Agent Systems](https://mlsanity.com/paper/2605.30102/) (2026)
- [SparSEEty: Extracting Tokens from Sparsity-Exploiting LLM Serving Systems via Deterministic Side Channels](https://mlsanity.com/paper/2608.02995/) (2026)
- [LeanStream: A Speculate-and-Refine Streaming Framework for Efficient on-Device LLM Inference](https://mlsanity.com/paper/2609.03079/) (2026)
- [Activation Sparsity with Weight Approximation for Faster LLM Decoding on Offloaded Weights](https://mlsanity.com/paper/2610.02598/) (2026)

## Abstract (arXiv)

While mobile devices provide ever more compute power, improvements in DRAM bandwidth are much slower. This is unfortunate for large language model (LLM) token generation, which is heavily memory-bound. Previous work has proposed to leverage natural dynamic activation sparsity in ReLU-activated LLMs to reduce effective DRAM bandwidth per token. However, more recent LLMs use SwiGLU instead of ReLU, which results in little inherent sparsity. While SwiGLU activations can be pruned based on magnitude, the resulting sparsity patterns are difficult to predict, rendering previous approaches ineffective. To circumvent this issue, our work introduces Dynamic Input Pruning (DIP): a predictor-free dynamic sparsification approach, which preserves accuracy with minimal fine-tuning. DIP can further use lightweight LoRA adapters to regain some performance lost during sparsification. Lastly, we describe a novel cache-aware masking strategy, which considers the cache state and activation magnitude to further increase cache hit rate, improving LLM token rate on mobile devices. DIP outperforms other methods in terms of accuracy, memory and throughput trade-offs across simulated hardware settings. On Phi-3-Medium, DIP achieves a 46% reduction in memory and 40% increase in throughput with < 0.1 loss in perplexity when compared to streaming the dense model from Flash. The open source code for HW simulator, methods, and experiments in this paper is available at https://github.com/Qualcomm-AI-research/dynamic-sparsity .