diff --git a/README.md b/README.md
index aa41938..0fabb96 100644
--- a/README.md
+++ b/README.md
@@ -31,7 +31,7 @@ y = layer(x, p, st)
 We compare the performance of different implementation of KAN with an MLP that has the same number of parameters (see `examples/eg1.jl`).
 ```julia
 using Lux, KolmogorovArnold
-using CUDA, LuxDeviceUtils
+using CUDA
 
 CUDA.allowscalar(false)
 device = Lux.gpu_device()