Not able to use BatchNorm with track_stats=false on GPU. julia> using Flux julia> model = Chain(Dense(10,5),BatchNorm(5,relu; track_stats=false)) Chain(Dense(10, 5), BatchNorm(5, relu)) julia> modelgpu = gpu(model) Chain(Dense(10, 5), BatchNorm(5, relu)) julia> x = randn(Float32, 10, 100); juli…

BatchNorm: only track_stats=true supported on gpu

ymtoo May 31, 2021, 3:50am 3

Thanks for pointing out the functions of BatchNorm on GPU.

Topic		Replies	Views
Issue with Flux BatchNorm Machine Learning question , flux , machine-learning	2	558	June 18, 2023
Modifying Flux source code for GPU Specific Domains	4	220	March 27, 2023
Inconsitent gradients of BachNorm on GPU in testmode GPU bug , cuda , flux , zygote , cudnn	0	232	October 19, 2023
BatchNorm gradient is inconsistent with PyTorch in training mode Machine Learning question , flux	2	874	December 18, 2020
Compilation error in Zygote, Flux and CUDA interaction Machine Learning question , cuda , error , flux , zygote	3	413	January 6, 2023