paint-brush
Recurrent Models: Enhancing Latency and Throughput Efficiencyby@gating

Recurrent Models: Enhancing Latency and Throughput Efficiency

by Gating Technology
Gating Technology HackerNoon profile picture

Gating Technology

@gating

We cover HOW gating is used in technology adoption. #Meritocracy...

January 14th, 2025
Read on Terminal Reader
Read this story in a terminal
Print this story
Read this story w/o Javascript
Read this story w/o Javascript
tldt arrow

Too Long; Didn't Read

This research shows recurrent models reduce cache size, improving latency and throughput over Transformers for long sequences.
featured image - Recurrent Models: Enhancing Latency and Throughput Efficiency
1x
Read by Dr. One voice-avatar

Listen to this story

Gating Technology HackerNoon profile picture
Gating Technology

Gating Technology

@gating

We cover HOW gating is used in technology adoption. #Meritocracy #Gating #TechnologyResearch

About @gating
LEARN MORE ABOUT @GATING'S
EXPERTISE AND PLACE ON THE INTERNET.
0-item

STORY’S CREDIBILITY

Academic Research Paper

Academic Research Paper

Part of HackerNoon's growing list of open-source research papers, promoting free access to academic material.

Authors:

(1) Soham De, Google DeepMind and with Equal contributions;

(2) Samuel L. Smith, Google DeepMind and with Equal contributions;

(3) Anushan Fernando, Google DeepMind and with Equal contributions;

(4) Aleksandar Botev, Google DeepMind and with Equal contributions;

(5) George Cristian-Muraru, Google DeepMind and with Equal contributions;

(6) Albert Gu, Work done while at Google DeepMind;

(7) Ruba Haroun, Google DeepMind;

(8) Leonard Berrada, Google DeepMind;

(9) Yutian Chen, Google DeepMind;

(10) Srivatsan Srinivasan, Google DeepMind;

(11) Guillaume Desjardins, Google DeepMind;

(12) Arnaud Doucet, Google DeepMind;

(13) David Budden, Google DeepMind;

(14) Yee Whye Teh, Google DeepMind;

(15) David Budden, Google DeepMind;

(16) Razvan Pascanu, Google DeepMind;

(17) Nando De Freitas, Google DeepMind;

(18) Caglar Gulcehre, Google DeepMind.

1 Introduction

2 Model Architecture

3 Recurrent Models Scale as Efficiently as Transformers

3.1. Scaling curves

3.2. Evaluation on downstream tasks

4 Training Recurrent Models Efficiently on Device and 4.1. Model parallelism for large scale training

4.2. Efficient linear recurrences on device

4.3. Training speed on longer sequences

5. Inference Speed

5.1. A simple model of the decode step

5.2. Results

6. Long Context Modeling and 6.1. Improving next token prediction with longer contexts

6.2. Copy and retrieval capabilities

7. Related Works

8. Conclusion, Acknowledgements, and References


A. RG-LRU Recurrence Gate

B. Complex-Gated Linear Recurrent Unit (CG-LRU)

C. Model Scale Hyper-Parameters

D. Efficient Linear Recurrences on Device

E. The Local Attention Window Size of Griffin

F. Inference Speeds

G. Improving Next Token Prediction with Longer Contexts: Additional Results

H. Additional Details of the Copy and Retrieval Tasks

5.1. A simple model of the decode step

image


Here, cache size refers to either the size of the KV cache at batch size 1 (for Transformers), or to the size of the recurrent state at batch size 1 (for RNNs).


Cache sizes The difference in cache size relative to model parameters has important implications for sampling efficiency. In recurrent and local attention blocks, parameter loading is the primary bottleneck, (because the cache size is substantially smaller). In contrast, global attention’s KV cache scales with the sequence length 𝑇 and can be comparable to, or even exceed, the size of the model parameters. This introduces considerable overhead when the sequence length 𝑇 is large enough (as shown in F.4). Consequently, an equally sized recurrent model can exhibit substantially lower latency than a Transformer when 𝑇 is large. Note however that as the model size grows the sequence length at which we see latency benefits (where the KV cache size is comparable to parameter size) also increases. It is important to note that, as well as improving latency, having a small recurrent state can also increase the largest batch size that fits in memory on a single device, leading to higher throughput.


Figure 4 | Latency of different 1B parameter models for a range of sequence lengths for (a) sampling from an empty prefill and (b) sampling from a prefill of 4k tokens.

Figure 4 | Latency of different 1B parameter models for a range of sequence lengths for (a) sampling from an empty prefill and (b) sampling from a prefill of 4k tokens.


This paper is available on arxiv under CC BY 4.0 DEED license.


L O A D I N G
. . . comments & more!

About Author

Gating Technology HackerNoon profile picture
Gating Technology@gating
We cover HOW gating is used in technology adoption. #Meritocracy #Gating #TechnologyResearch

TOPICS

THIS ARTICLE WAS FEATURED IN...

Permanent on Arweave
Read on Terminal Reader
Read this story in a terminal
 Terminal
Read this story w/o Javascript
Read this story w/o Javascript
 Lite
Hackernoon
Bsky