Currently, cpu(optimiser) won’t move it. For instance, the state still consists of variables in GPU.

Getting what you want here might require a bit of extra effort. Flux’s current optimizers use IdDicts to map weights to optimizer state and when you move parameters to and from the gpu you create new copies. Result is that the IdDict will not recoginize them as the same weighs and instead you have …

How to move optimiser from gpu to cpu?

Specific Domains Machine Learning

ToucheSir December 12, 2021, 6:29pm 11

So it turns out there is an easier way to go about this, see Deepcopy Flux Model - #9 by ToucheSir.

Topic		Replies	Views
Flux.jl: Save model and optimizer from gpu Machine Learning question	1	426	November 9, 2022
Training with Flux.jl on the GPU causes ArgumentError: cannot take the CPU address of a CuArray GPU question , gpu , flux , machine-learning , neural-network	4	1071	May 28, 2022
Flux.jl: training fails at GPU but works on CPU Machine Learning gpu , flux	1	625	September 19, 2019
Flux: Hard to use train! and DataLoader for minibatched NamedTuple dataset with GPU Machine Learning flux	2	1430	September 24, 2020
Flux on GPU too slow Machine Learning gpu , cuda , flux	5	1070	September 22, 2022

How to move optimiser from gpu to cpu?

Related topics