diff --git a/examples/ConvMixer/main.jl b/examples/ConvMixer/main.jl
index ac36b6f570..29b85246af 100644
--- a/examples/ConvMixer/main.jl
+++ b/examples/ConvMixer/main.jl
@@ -30,10 +30,10 @@ function get_dataloaders(batchsize; kwargs...)
     test_transform = ImageToTensor() |> Normalize(cifar10_mean, cifar10_std)
 
     trainset = TensorDataset(CIFAR10(:train), train_transform)
-    trainloader = DataLoader(trainset; batchsize, shuffle=true, parallel=true, kwargs...)
+    trainloader = DataLoader(trainset; batchsize, shuffle=true, kwargs...)
 
     testset = TensorDataset(CIFAR10(:test), test_transform)
-    testloader = DataLoader(testset; batchsize, shuffle=false, parallel=true, kwargs...)
+    testloader = DataLoader(testset; batchsize, shuffle=false, kwargs...)
 
     return trainloader, testloader
 end
@@ -74,8 +74,8 @@ function accuracy(model, ps, st, dataloader)
 end
 
 Comonicon.@main function main(; batchsize::Int=512, hidden_dim::Int=256, depth::Int=8,
-        patch_size::Int=2, kernel_size::Int=5, weight_decay::Float64=1e-5,
-        clip_norm::Bool=false, seed::Int=42, epochs::Int=25, lr_max::Float64=0.01,
+        patch_size::Int=2, kernel_size::Int=5, weight_decay::Float64=1e-3,
+        clip_norm::Bool=false, seed::Int=42, epochs::Int=25, lr_max::Float64=0.001,
         backend::String="reactant")
     rng = StableRNG(seed)
 
@@ -118,7 +118,7 @@ Comonicon.@main function main(; batchsize::Int=512, hidden_dim::Int=256, depth::
         model_compiled = model
     end
 
-    loss = CrossEntropyLoss(; logits=Val(true))
+    loss_fn = CrossEntropyLoss(; logits=Val(true))
 
     @printf "[Info] Training model\n"
     for epoch in 1:epochs
@@ -127,8 +127,8 @@ Comonicon.@main function main(; batchsize::Int=512, hidden_dim::Int=256, depth::
         for (i, (x, y)) in enumerate(trainloader)
             lr = lr_schedule((epoch - 1) + (i + 1) / length(trainloader))
             train_state = Optimisers.adjust!(train_state, lr)
-            (_, _, _, train_state) = Training.single_train_step!(
-                adtype, loss, (x, y), train_state
+            (_, loss, _, train_state) = Training.single_train_step!(
+                adtype, loss_fn, (x, y), train_state
             )
         end
         ttime = time() - stime