Cite
Notes
Only stored in your browser.
Attribution
Few-Bit Backward: Quantized Gradients of Activation Functions for Memory Footprint Reduction
arXiv 2022
Memory-Efficient Backpropagation through Large Linear Layers
from 2 papers
Daniel Bershatsky
Ivan Oseledets
Aleksandr Mikhalev
Alex Shonenkov
Alexandr Katrutsa
Daniil Merkulov
Denis Dimitrov
Georgii Novikov