AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

MACKO: Sparse matrix-vector multiplication for low sparsity

Vladimír Macko, Vladimír Boža

ICML 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: Layer-wise pruning of LLMs · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use MACKO sparse format instead of dense cuBLAS at 30% sparsityLlama2-7b generation · Llama2-7b Wanda, RTX 4090tokens/s: 66.49 -> 73.88Improved
use MACKO instead of dense at 50% sparsityLlama2-7b generation · Llama2-7b Wanda, RTX 4090tokens/s: 66.53 -> 98.60Improved
use MACKO instead of dense at 90% sparsityLlama2-7b generation · Llama2-7b Wanda, RTX 4090tokens/s: 66.48 -> 255.01Improved
use MACKO instead of dense at 20% sparsityLlama2-7b generation · Llama2-7b Wanda, RTX 4090tokens/s and size: 66.55 -> 66.36 toks/sec; 13.59 -> 13.77 GBHurt
use MACKO model size at 90% sparsity vs denseLlama2-7b · Llama2-7b Wanda, RTX 4090model size (GB): 13.59 -> 2.67Improved
raise sparsity from 50% to 70% with MACKO SpMVMACKO-SpMV 12288x12288 · RTX 4090speedup over cuBLAS: 1.3x -> 1.96xImproved
raise sparsity to 95% with MACKO SpMVMACKO-SpMV 12288x12288 · RTX 4090speedup over cuBLAS: 5.62x at 95% sparsity, but outperformed by SputnikMixed
not pruning embedding layers (Wanda) limits end-to-end gain at 90% sparsityLlama2-7b with MACKO · Llama2-7b 90% sparsityspeedup and memory reduction: 2% of original size -> 10% of 90%-sparse modelHurt
use b_delta=1 index width (ablation)MACKO format · storagememory vs dense: improves over dense from 7% sparsityMixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy