improve the way the workgroup_sizes are structured in the shaders

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "intricate"
-version = "0.1.3"
+version = "0.1.4"
 edition = "2021"
 license = "MIT"
 keywords = ["neural-networks", "machine-learning", "backpropagation"]

src/gpu/shaders/apply_gradients_to_dense_weights.wgsl

@@ -41,7 +41,7 @@ fn compute_sample_weight_gradient(sample_index: u32, input_index: u32, output_in
 }
 
 @compute
-@workgroup_size(255)
+@workgroup_size(16, 16)
 fn main(@builtin(global_invocation_id) global_id: vec3<u32>) {
     var input_index: u32 = global_id.x;
     var output_index: u32 = global_id.y;
@@ -58,4 +58,4 @@ fn main(@builtin(global_invocation_id) global_id: vec3<u32>) {
     var old_weight: f32 = flattened_layer_weights[flattened_weight_index];
 
     flattened_layer_weights[flattened_weight_index] = old_weight + weight_gradient;
-}
+}

src/gpu/shaders/calculate_dense_input_to_error_derivatives.wgsl

@@ -36,7 +36,7 @@ fn compute_input_to_error_derivative(sample_index: u32, input_index: u32) -> f32
 }
 
 @compute
-@workgroup_size(255)
+@workgroup_size(16, 16)
 fn main(@builtin(global_invocation_id) global_id: vec3<u32>) {
     var sample_index: u32 = global_id.x;
     var input_index: u32 = global_id.y;

src/gpu/shaders/propagate_through_weights_and_biases.wgsl

@@ -40,7 +40,7 @@ fn calculate_output_sample_for_all_inputs(sample_index: u32, output_index: u32)
 }
 
 @compute
-@workgroup_size(255)
+@workgroup_size(16, 16)
 fn main(@builtin(global_invocation_id) global_id: vec3<u32>) {
     var sample_index = global_id.x;
     var output_index = global_id.y;

src/tests/gpu_xor.rs

@@ -39,7 +39,7 @@ async fn should_decerase_error_test() {
             &training_output_samples, 
             &TrainingOptionsF32 {
                 loss_algorithm: Box::new(MeanSquared),
-                learning_rate: 0.3,
+                learning_rate: 0.1,
                 should_print_information: false,
                 instantiate_gpu: true,
                 epochs: 0,