step: 10, loss: 25,200823, accuracy: 0,75 
step: 20, loss: 17,497072, accuracy: 0,8125 
step: 30, loss: 4,7848167, accuracy: 0,96875 
step: 40, loss: 11,228649, accuracy: 0,90625 
step: 50, loss: 8,46461, accuracy: 0,90625 
step: 60, loss: 12,897742, accuracy: 0,875 
step: 70, loss: 5,8186455, accuracy: 0,90625 
step: 80, loss: 1,6474819, accuracy: 1 
step: 90, loss: 5,5969973, accuracy: 0,90625 
Test Accuracy: 0,9 
18.11.2021 00:18:13 Completed MNIST CNN (Eager)
Example: MNIST CNN (Eager) in 4,7636146s is OK!

step: 10, loss: 73,61097, accuracy: 0,125 
step: 20, loss: 74,02429, accuracy: 0,03125 
step: 30, loss: 73,439026, accuracy: 0,03125 
step: 40, loss: 73,71028, accuracy: 0,0625 
step: 50, loss: 73,987946, accuracy: 0,125 
step: 60, loss: 73,33862, accuracy: 0,15625 
step: 70, loss: 72,83645, accuracy: 0,21875 
step: 80, loss: 73,17415, accuracy: 0,21875 
step: 90, loss: 73,84353, accuracy: 0,09375 
Test Accuracy: 0,11 
18.11.2021 00:19:28 Completed MNIST CNN (Eager)
Example: MNIST CNN (Eager) in 4,4328711s is Failed!

var ((x_train, y_train), (x_test, y_test)) = keras.datasets.mnist.load_data();

var x_train_norm = (x_train.astype(TF_DataType.DtFloatRef) / 255f).reshape((-1, 28, 28, 1));
var x_test_norm = (x_test.astype(TF_DataType.DtFloatRef) / 255f).reshape((-1, 28, 28, 1));

var model = keras.Sequential(new()
{
    keras.layers.InputLayer((28, 28, 1)),
    keras.layers.Conv2D(32, 3, activation: keras.activations.Relu),
    keras.layers.MaxPooling2D(),
    keras.layers.Conv2D(64, 2, activation: keras.activations.Relu),
    keras.layers.MaxPooling2D(),
    keras.layers.Flatten(),
    keras.layers.Dense(10, keras.activations.Softmax)
});

model.compile(keras.optimizers.Adam(), keras.losses.SparseCategoricalCrossentropy(from_logits: true), new [] { "accuracy" } );

model.summary();

model.fit(x_train_norm, y_train, 128, 2);

Epoch: 001/002, Step: 0001/0469, loss: 2,302554, accuracy: 0,062500
...
Epoch: 001/002, Step: 0469/0469, loss: 1,604421, accuracy: 0,870233
Epoch: 002/002, Step: 0001/0469, loss: 1,497208, accuracy: 0,968750
...
Epoch: 002/002, Step: 0469/0469, loss: 1,498555, accuracy: 0,965783

Epoch: 001/002, Step: 0001/0469, loss: 2,302585, accuracy: 0,796875
...
Epoch: 001/002, Step: 0469/0469, loss: 1,736288, accuracy: 0,768267
Epoch: 002/002, Step: 0001/0469, loss: 1,726776, accuracy: 0,734375
...
Epoch: 002/002, Step: 0469/0469, loss: 1,692857, accuracy: 0,768283