diff --git a/src/initialize.jl b/src/initialize.jl
index 8cd919c..16da6ff 100644
--- a/src/initialize.jl
+++ b/src/initialize.jl
@@ -7,7 +7,7 @@ using Random: Random
 
 using LuxCore: LuxCore
 
-using ..Utils: is_extension_loaded, unwrap_val
+using ..Utils: is_extension_loaded
 
 get_pretrained_weights_path(name::Symbol) = get_pretrained_weights_path(string(name))
 function get_pretrained_weights_path(name::String)
diff --git a/test/vision_tests.jl b/test/vision_tests.jl
index f5d1595..9945050 100644
--- a/test/vision_tests.jl
+++ b/test/vision_tests.jl
@@ -1,9 +1,9 @@
 @testitem "AlexNet" setup=[SharedTestSetup] tags=[:vision] begin
     for (mode, aType, dev, ongpu) in MODES
         @testset "pretrained: $(pretrained)" for pretrained in [true, false]
-            model, ps, st = Vision.AlexNet(; pretrained)
-            ps = ps |> dev
-            st = Lux.testmode(st) |> dev
+            model = Vision.AlexNet(; pretrained)
+            ps, st = Lux.setup(Random.default_rng(), model) |> dev
+            st = Lux.testmode(st)
             img = randn(Float32, 224, 224, 3, 2) |> aType
 
             @jet model(img, ps, st)
@@ -16,9 +16,9 @@ end
 
 @testitem "ConvMixer" setup=[SharedTestSetup] tags=[:vision] begin
     for (mode, aType, dev, ongpu) in MODES, name in [:small, :base, :large]
-        model, ps, st = Vision.ConvMixer(name; pretrained=false)
-        ps = ps |> dev
-        st = Lux.testmode(st) |> dev
+        model = Vision.ConvMixer(name; pretrained=false)
+        ps, st = Lux.setup(Random.default_rng(), model) |> dev
+        st = Lux.testmode(st)
         img = randn(Float32, 256, 256, 3, 2) |> aType
 
         @jet model(img, ps, st)
@@ -30,9 +30,9 @@ end
 
 @testitem "GoogLeNet" setup=[SharedTestSetup] tags=[:vision] begin
     for (mode, aType, dev, ongpu) in MODES
-        model, ps, st = Vision.GoogLeNet(; pretrained=false)
-        ps = ps |> dev
-        st = Lux.testmode(st) |> dev
+        model = Vision.GoogLeNet(; pretrained=false)
+        ps, st = Lux.setup(Random.default_rng(), model) |> dev
+        st = Lux.testmode(st)
         img = randn(Float32, 224, 224, 3, 2) |> aType
 
         @jet model(img, ps, st)
@@ -44,9 +44,9 @@ end
 
 @testitem "MobileNet" setup=[SharedTestSetup] tags=[:vision] begin
     for (mode, aType, dev, ongpu) in MODES, name in [:v1, :v2, :v3_small, :v3_large]
-        model, ps, st = Vision.MobileNet(name; pretrained=false)
-        ps = ps |> dev
-        st = Lux.testmode(st) |> dev
+        model = Vision.MobileNet(name; pretrained=false)
+        ps, st = Lux.setup(Random.default_rng(), model) |> dev
+        st = Lux.testmode(st)
         img = randn(Float32, 224, 224, 3, 2) |> aType
 
         @jet model(img, ps, st)
@@ -58,9 +58,9 @@ end
 
 @testitem "ResNet" setup=[SharedTestSetup] tags=[:vision] begin
     for (mode, aType, dev, ongpu) in MODES, depth in [18, 34, 50, 101, 152]
-        model, ps, st = Vision.ResNet(depth; pretrained=false)
-        ps = ps |> dev
-        st = Lux.testmode(st) |> dev
+        model = Vision.ResNet(depth; pretrained=false)
+        ps, st = Lux.setup(Random.default_rng(), model) |> dev
+        st = Lux.testmode(st)
         img = randn(Float32, 224, 224, 3, 2) |> aType
 
         @jet model(img, ps, st)
@@ -72,9 +72,9 @@ end
 
 @testitem "ResNeXt" setup=[SharedTestSetup] tags=[:vision] begin
     for (mode, aType, dev, ongpu) in MODES, depth in [50, 101, 152]
-        model, ps, st = Vision.ResNeXt(depth; pretrained=false)
-        ps = ps |> dev
-        st = Lux.testmode(st) |> dev
+        model = Vision.ResNeXt(depth; pretrained=false)
+        ps, st = Lux.setup(Random.default_rng(), model) |> dev
+        st = Lux.testmode(st)
         img = randn(Float32, 224, 224, 3, 2) |> aType
 
         @jet model(img, ps, st)
@@ -90,9 +90,9 @@ end
                 false, true],
             batchnorm in [false, true]
 
-            model, ps, st = Vision.VGG(depth; batchnorm, pretrained)
-            ps = ps |> dev
-            st = Lux.testmode(st) |> dev
+            model = Vision.VGG(depth; batchnorm, pretrained)
+            ps, st = Lux.setup(Random.default_rng(), model) |> dev
+            st = Lux.testmode(st)
             img = randn(Float32, 224, 224, 3, 2) |> aType
 
             @jet model(img, ps, st)
@@ -106,9 +106,9 @@ end
 @testitem "VisionTransformer" setup=[SharedTestSetup] tags=[:vision] begin
     for (mode, aType, dev, ongpu) in MODES, name in [:tiny, :small, :base]
         # :large, :huge, :giant, :gigantic --> too large for CI
-        model, ps, st = Vision.VisionTransformer(name; pretrained=false)
-        ps = ps |> dev
-        st = Lux.testmode(st) |> dev
+        model = Vision.VisionTransformer(name; pretrained=false)
+        ps, st = Lux.setup(Random.default_rng(), model) |> dev
+        st = Lux.testmode(st)
         img = randn(Float32, 256, 256, 3, 2) |> aType
 
         @jet model(img, ps, st)