Multi-Layer Transformers Gradient Can be Approximated in Almost Linear Time

AmazUtah_NLP at SemEval-2024 Task 9: A MultiChoice Question Answering System for Commonsense Defying Reasoning


View a PDF of the paper titled Multi-Layer Transformers Gradient Can be Approximated in Almost Linear Time, by Yingyu Liang and 4 other authors

View PDF
HTML (experimental)

Abstract:The computational complexity of the self-attention mechanism in popular transformer architectures poses significant challenges for training and inference, and becomes the bottleneck for long inputs. Is it possible to significantly reduce the quadratic time complexity of computing the gradients in multi-layer transformer models? This paper proves that a novel fast approximation method can calculate the gradients in almost linear time $n^{1+o(1)}$ where $n$ is the input sequence length, while it maintains a polynomially small approximation error $1 / mathrm{poly}(n)$ across the entire model. Our theory holds for general loss functions and when the multi-layer transformer model contains many practical sub-modules, such as residual connection, casual mask, and multi-head attention. By improving the efficiency of gradient computation, we hope that this work will facilitate more effective training and deployment of long-context language models based on our theoretical results.

Submission history

From: Yufa Zhou [view email]
[v1]
Fri, 23 Aug 2024 17:16:43 UTC (57 KB)
[v2]
Tue, 15 Oct 2024 04:11:19 UTC (55 KB)



Source link
lol

By stp2y

Leave a Reply

Your email address will not be published. Required fields are marked *

No widgets found. Go to Widget page and add the widget in Offcanvas Sidebar Widget Area.