Authors:
(1) Ben Athiwaratkun, AWS AI Labs;
(2) Sujan Kumar Gonugondla, AWS AI Labs;
(3) Sanjay Krishna Gouda, AWS AI Labs;
(4) Haifeng Qian, AWS AI Labs;
(5) Sanjay Krishna Gouda, AWS AI Labs;
(6) Hantian Ding, AWS AI Labs;
(7) Qing Sun, AWS AI Labs;
(8) Jun Wang, AWS AI Labs;
(9) Jiacheng Guo, AWS AI Labs;
(10 Liangfu Chen, AWS AI Labs;
(11) Parminder Bhatia, GE HealthCare (work done at AWS);
(12) Ramesh Nallapati, Amazon AGI (work done at AWS);
(13) Sudipta Sengupta, AWS AI Labs;
(14) Bing Xiang, Goldman Sachs (work done at AWS).
Table of Links
3.1. Notation and 3.2. Language Model Inference
3.3. Multi-Query, Multi-Head and the Generalized Multi-Query Attention
4. Context-Aware Bifurcated Attention and 4.1. Motivation
4.2. Formulation and 4.3. Memory IO Complexity
5.1. Comparing Capabilities of Multi-Head, Multi-Query, and Multi-Group Attention
5.2. Latencies of Capabilities-Equivalent Models
D. Multi-Group Attention Family
E. Context-Aware Bifurcated Attention
F. Applications: Additional Results
G. Compatibility with Speculative Decoding and Fast Decoding techniques
4. Context-Aware Bifurcated Attention
In this section, we present a novel context-aware bifurcated attention method that aims to reduce the memory IO cost during incremental decoding by efficiently handling the computation of attention for shared context across samples, as shown in Figure 2.
4.1. Motivation
4.2. Formulation
The proposed operations yield the exact same results ⟨w, V ⟩ as the original attention in Equation 1 and 2, but can significantly reduce memory I/O during incremental decoding (proof in Appendix E.1).
4.3. Memory IO Complexity
The memory IO complexity corresponding to loading KV changes from
memory IO w/o bifurcated attention = gk · bm (5)
= gk · b(mc + md)
memory IO w. bifurcated attention = gk · (mc + bmd) (6)
This paper is available on arxiv under CC BY 4.0 DEED license.
