Saved in:
Bibliographic Details
Main Author: Zhao, Jialin
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2510.01718
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908573141827584
author Zhao, Jialin
author_facet Zhao, Jialin
contents Attention is a core operation in large language models (LLMs) and vision-language models (VLMs). We present BD Attention (BDA), the first lossless algorithmic reformulation of attention. BDA is enabled by a simple matrix identity from Basis Decomposition (BD), which restructures multi-head projections into a compact form while preserving exact outputs. Unlike I/O-aware system optimizations such as FlashAttention, BDA provides a mathematically guaranteed acceleration that is architecture-agnostic. On DeepSeek-V2-Lite (16B, FP16), BDA requires only 4s of offline preparation with no retraining required and, on modern GPUs, achieves 32% faster key/value projections and 25% smaller weights, while increasing end-to-end perplexity (PPL) by just 0.02% (FP16) or 0.0004% (FP32), a negligible effect on model performance. These results position BDA as the first theoretically exact method for lossless attention acceleration that is complementary to existing engineering-level optimizations. Our code is available at https://github.com/abcbdf/basis-decomposition-official.
format Preprint
id arxiv_https___arxiv_org_abs_2510_01718
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Accelerating Attention with Basis Decomposition
Zhao, Jialin
Machine Learning
Attention is a core operation in large language models (LLMs) and vision-language models (VLMs). We present BD Attention (BDA), the first lossless algorithmic reformulation of attention. BDA is enabled by a simple matrix identity from Basis Decomposition (BD), which restructures multi-head projections into a compact form while preserving exact outputs. Unlike I/O-aware system optimizations such as FlashAttention, BDA provides a mathematically guaranteed acceleration that is architecture-agnostic. On DeepSeek-V2-Lite (16B, FP16), BDA requires only 4s of offline preparation with no retraining required and, on modern GPUs, achieves 32% faster key/value projections and 25% smaller weights, while increasing end-to-end perplexity (PPL) by just 0.02% (FP16) or 0.0004% (FP32), a negligible effect on model performance. These results position BDA as the first theoretically exact method for lossless attention acceleration that is complementary to existing engineering-level optimizations. Our code is available at https://github.com/abcbdf/basis-decomposition-official.
title Accelerating Attention with Basis Decomposition
topic Machine Learning
url https://arxiv.org/abs/2510.01718