arxiv:2606.13392

MiniMax Sparse Attention

Published on Jun 11

· Submitted by

Ryanlee on Jun 12

#2 Paper of the day

MiniMax

Upvote

Authors:

Xunhao Lai ,

Weiqi Xu ,

Lunbin Zeng ,

Haohai Sun ,

Abstract

MiniMax Sparse Attention enables efficient processing of ultra-long contexts in large language models through blockwise sparsity and optimized GPU execution, achieving significant speedups while maintaining performance.

Generated by Qwen/Qwen2.5-Coder-32B-Instruct

Ultra-long-context capability is becoming indispensable for frontier LLMs: agentic workflows, repository-scale code reasoning, and persistent memory all require the model to jointly attend over hundreds of thousands to millions of tokens, yet the quadratic cost of softmax attention makes this untenable at deployment scale. We introduce MiniMax Sparse Attention (MSA), a blockwise sparse attention built upon Grouped Query Attention (GQA). A lightweight Index Branch scores key-value blocks and independently selects a Top-k subset for each GQA group, enabling group-specific sparse retrieval while maintaining efficient block-level execution; the Main Branch then performs exact block-sparse attention over only the selected blocks. Designed around a principle of simplicity and scalability, MSA is deliberately streamlined, making it straightforward to deploy efficiently across a broad range of GPUs. To translate sparsity into practical speedups, we co-design MSA with a GPU execution path that uses exp-free Top-k selection and KV-outer sparse attention to improve tensor-core utilization under block-granular access. On a 109B-parameter model with native multimodal training, MSA performs on par with GQA while reducing per-token attention compute by 28.4x at 1M context. Paired with our co-designed kernel, MSA achieves 14.2x prefill and 7.6x decoding wall-clock speedups on H800. Our inference kernel is available at: https://github.com/MiniMax-AI/MSA. A production-grade natively multimodal model powered by MSA has been publicly released at: https://huggingface.co/MiniMaxAI/MiniMax-M3.

View arXiv page View PDF GitHub 240 Add to collection

Community

ryanlee-dev

Paper submitter 1 day ago

our high-performance MSA kernel library is now open-source

librarian-bot

about 19 hours ago

This is an automated message from the Librarian Bot. I found the following papers similar to this paper.

The following papers were recommended by the Semantic Scholar API

Please give a thumbs up to this comment if you found it helpful!

If you want recommendations for any Paper on Hugging Face checkout this Space

You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: @librarian-bot recommend

avahal

about 19 hours ago

the two-branch design in MiniMax Sparse Attention finally makes ultra-long context feel practical. a lightweight indexer scores block KV and picks per-GQA top-k, then the main branch runs exact block-sparse attention on those blocks—that clean separation is a surprisingly effective compromise between fidelity and hardware efficiency. i’d love to see a focused ablation on how the per-group top-k budget scales with data distribution and context length; if the indexer mis-ranks the blocks, could you lose crucial history despite only attending a subset? the arXivLens breakdown had a nice walkthrough (https://arxivlens.com/PaperView/Details/minimax-sparse-attention-799-e528c64a) that helped me parse the method details, especially the exp-free TopK and KV-outer pattern that map to tensor-core utilization