paper-with-me

홈 › Papers

Zero-th Order Algorithm for Softmax Attention Optimization

2023-07-17 · Yichuan Deng, Zhihang Li, Sridhar Mahadevan, Zhao Song

Large language models (LLMs) have brought about significant transformations in human society. Among the crucial computations in LLMs, the softmax unit holds great importance. Its helps the model generating a probability distribution on potential subsequent words or phrases, considering a series of input words. By utilizing this distribution, the model selects the most probable next word or phrase, based on the assigned probabilities. The softmax unit assumes a vital function in LLM training as it facilitates learning from data through the adjustment of neural network weights and biases. With the development of the size of LLMs, computing the gradient becomes expensive. However, Zero-th Order method can approximately compute the gradient with only forward passes. In this paper, we present a Zero-th Order algorithm specifically tailored for Softmax optimization. We demonstrate the convergence of our algorithm, highlighting its effectiveness in efficiently computing gradients for large-scale LLMs. By leveraging the Zeroth-Order method, our work contributes to the advancement of optimization techniques in the context of complex language models.

📄 PDF Abstract BibTeX arXiv:2307.08352

Code (0)

등록된 구현이 없습니다.

Methods 이 논문이 사용한 방법론

Softmax The Softmax output function transforms a previous layer's output into a vector of probabilities. It is commonly used for multiclass classification. Given an input vector $x$…

Similar Papers 제목 키워드 기반

ZeroS: Zero-Sum Linear Attention for Efficient Transformers

2026-02-05 · Jiecheng Lu, Xu Han, Yan Sun, Viresh Pati 외 arxiv

Linear attention methods offer Transformers $O(N)$ complexity but typically underperform standard softmax attention. We identify two fundamental limitations affecting these approaches: the restriction to convex combinati…

Fine-Grained Theoretical Analysis of Federated Zeroth-Order Optimization

2023-09-21 · NeurIPS 2023 11

Federated zeroth-order optimization (FedZO) algorithm enjoys the advantages of both zeroth-order optimization and federated learning, and has shown exceptional performance on black-box attack and softmax regression tasks…

Statistical Advantage of Softmax Attention: Insights from Single-Location Regression

2025-09-26 · O. Duranthon, P. Marion, C. Boyer, B. Loureiro 외 arxiv

Large language models rely on attention mechanisms with a softmax activation. Yet the dominance of softmax over alternatives (e.g., component-wise or linear) remains poorly understood, and many theoretical works have foc…

Training Dynamics of Softmax Self-Attention: Fast Global Convergence via Preconditioning

2026-03-02 · Gautam Goel, Mahdi Soltanolkotabi, Peter Bartlett arxiv

We study the training dynamics of gradient descent in a softmax self-attention layer trained to perform linear regression and show that a simple first-order optimization algorithm can converge to the globally optimal sel…

Zeroth-order Asynchronous Doubly Stochastic Algorithm with Variance Reduction

2016-12-05 · Bin Gu, Zhouyuan Huo, Heng Huang

Zeroth-order (derivative-free) optimization attracts a lot of attention in machine learning, because explicit gradient calculations may be computationally expensive or infeasible. To handle large scale problems both in v…