Abstract
Efficient training of large-scale graph neural networks (GNNs) has been studied with a specific focus on reducing their memory consumption. Work by Liu et al. (2022) proposed extreme activation compression (EXACT) which demonstrated drastic reduction in memory consumption by performing quantization of the intermediate activation maps down to using INT2 precision. They showed little to no reduction in performance while achieving large reductions in GPU memory consumption. In this work, we present an improvement to the EXACT strategy by using block-wise quantization of the intermediate activations. We experimentally analyze different block sizes and show further reduction in memory consumption (> 15%), and runtime speedup per epoch (≈ 5%) even when performing extreme extents of quantization with similar performance trade-offs as with the original EXACT. Further, we present a correction to the assumptions on the distribution of intermediate activation maps in EXACT (assumed to be uniform) and show improved variance estimations of the quantization and dequantization steps.
Original language | English |
---|---|
Title of host publication | 2024 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2024 - Proceedings |
Number of pages | 5 |
Publisher | IEEE |
Publication date | 2024 |
Pages | 7430-7434 |
ISBN (Electronic) | 9798350344851 |
DOIs | |
Publication status | Published - 2024 |
Event | 49th IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2024 - Seoul, Korea, Republic of Duration: 14 Apr 2024 → 19 Apr 2024 |
Conference
Conference | 49th IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2024 |
---|---|
Country/Territory | Korea, Republic of |
City | Seoul |
Period | 14/04/2024 → 19/04/2024 |
Sponsor | The Institute of Electrical and Electronics Engineers Signal Processing Society |
Bibliographical note
Publisher Copyright:© 2024 IEEE.
Keywords
- activation compression
- deep learning
- efficient machine learning
- graph neural networks
- quantization