4-bit shampoo for memory-efficient network training
Second-order optimizers, maintaining a matrix termed a preconditioner, are superior to first-order optimizers in both theory and practice. The states forming the preconditioner and its inverse root restrict the maximum size of models trained by second-order optimizers. To address this, compressing 3...
Saved in:
Main Authors: | WANG, Sike, ZHOU, Pan, LI, Jia, HUANG, Hua |
---|---|
格式: | text |
語言: | English |
出版: |
Institutional Knowledge at Singapore Management University
2024
|
主題: | |
在線閱讀: | https://ink.library.smu.edu.sg/sis_research/9731 https://ink.library.smu.edu.sg/context/sis_research/article/10731/viewcontent/4_bit.pdf |
標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
相似書籍
-
ScaleLong: Towards more stable training of diffusion model via scaling network long skip connection
由: HUANG, Zhongzhan, et al.
出版: (2023) -
Sequential recommendation with user memory networks
由: CHEN, Xu, et al.
出版: (2018) -
Quantization-aware interval bound propagation for training certifiably robust quantized neural networks
由: LECHNER, Mathias, et al.
出版: (2023) -
Win: Weight-decay-integrated Nesterov acceleration for faster network training
由: ZHOU, Pan, et al.
出版: (2024) -
MemLock: Memory usage guided fuzzing
由: WEN, Cheng, et al.
出版: (2020)