Attentions Under the Microscope: A Comparative Study of Resource Utilization for Variants of Self-Attention
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915380620951552 |
|---|---|
| author | Tian, Zhengyu Kumar, Anantha Padmanaban Krishna Krishnakumar, Hemant Rawassizadeh, Reza |
| author_facet | Tian, Zhengyu Kumar, Anantha Padmanaban Krishna Krishnakumar, Hemant Rawassizadeh, Reza |
| contents | As large language models (LLMs) and visual language models (VLMs) grow in scale and application, attention mechanisms have become a central computational bottleneck due to their high memory and time complexity. While many efficient attention variants have been proposed, there remains a lack of rigorous evaluation on their actual energy usage and hardware resource demands during training. In this work, we benchmark eight attention mechanisms in training GPT-2 architecture, measuring key metrics including training time, GPU memory usage, FLOPS, CPU usage, and power consumption. Our results reveal that attention mechanisms with optimized kernel implementations, including Flash Attention, Locality-Sensitive Hashing (LSH) Attention, and Multi-Head Latent Attention (MLA), achieve the best energy efficiency. We further show that lower GPU power alone does not guarantee reduced energy use, as training time plays an equally important role. Our study highlights the importance of energy-aware benchmarking in attention design and provides a practical insight for selecting resource-efficient mechanisms. All our codes are available at GitHub. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2507_07247 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Attentions Under the Microscope: A Comparative Study of Resource Utilization for Variants of Self-Attention Tian, Zhengyu Kumar, Anantha Padmanaban Krishna Krishnakumar, Hemant Rawassizadeh, Reza Machine Learning Artificial Intelligence Neural and Evolutionary Computing As large language models (LLMs) and visual language models (VLMs) grow in scale and application, attention mechanisms have become a central computational bottleneck due to their high memory and time complexity. While many efficient attention variants have been proposed, there remains a lack of rigorous evaluation on their actual energy usage and hardware resource demands during training. In this work, we benchmark eight attention mechanisms in training GPT-2 architecture, measuring key metrics including training time, GPU memory usage, FLOPS, CPU usage, and power consumption. Our results reveal that attention mechanisms with optimized kernel implementations, including Flash Attention, Locality-Sensitive Hashing (LSH) Attention, and Multi-Head Latent Attention (MLA), achieve the best energy efficiency. We further show that lower GPU power alone does not guarantee reduced energy use, as training time plays an equally important role. Our study highlights the importance of energy-aware benchmarking in attention design and provides a practical insight for selecting resource-efficient mechanisms. All our codes are available at GitHub. |
| title | Attentions Under the Microscope: A Comparative Study of Resource Utilization for Variants of Self-Attention |
| topic | Machine Learning Artificial Intelligence Neural and Evolutionary Computing |
| url | https://arxiv.org/abs/2507.07247 |