我的专栏《大模型从入门到实战》2-1到2-3的配套源码，包含数据集

import tensorflow as tf from tensorflow import keras from tensorflow.keras import layers (x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data() #预处理数据（这些都是NumPy数组） x_train = x_train.reshape(60000, 784).astype("float32") / 255 x_test = x_test.reshape(10000, 784).astype("float32") / 255 y_train = y_train.astype("float32") y_test = y_test.astype("float32") #保留10000个样品进行验证 x_val = x_train[-10000:] y_val = y_train[-10000:] x_train = x_train[:-10000] y_train = y_train[:-10000] def get_uncompiled_model(): inputs = keras.Input(shape=(784,), name="digits") x = layers.Dense(64, activation="relu", name="dense_1")(inputs) x = layers.Dense(64, activation="relu", name="dense_2")(x) outputs = layers.Dense(10, activation="softmax", name="predictions")(x) model = keras.Model(inputs=inputs, outputs=outputs) return model def get_compiled_model(): model = get_uncompiled_model() model.compile( optimizer="rmsprop", loss="sparse_categorical_crossentropy", metrics=["sparse_categorical_accuracy"], ) return model model = get_compiled_model() # 首先，让我们创建一个训练数据集实例，接下来将使用与前面相同的MNIST数据。 train_dataset = tf.data.Dataset.from_tensor_slices((x_train, y_train)) # 洗牌并切片数据集. train_dataset = train_dataset.shuffle(buffer_size=1024).batch(64) # 现在我们得到了一个测试数据集. test_dataset = tf.data.Dataset.from_tensor_slices((x_test, y_test)) test_dataset = test_dataset.batch(64) #由于数据集已经处理批处理，所以我们不传递“batch\u size”参数。 model.fit(train_dataset, epochs=3) #还可以对数据集进行评估或预测. print("Evaluate评估:") result = model.evaluate(test_dataset) dict(zip(model.metrics_names, result))