Keras 3.0发布：全面拥抱 PyTorch！

Keras 3.0 介绍

https://keras.io/keras_3/

Keras 3.0 升级是对 Keras 的全面重写，引入了一系列令人振奋的新特性，为深度学习领域带来了全新的可能性。

多框架支持

Keras 3.0 的最大亮点之一是支持多框架。Keras 3 实现了完整的 Keras API，并使其可用于 TensorFlow、JAX 和 PyTorch ------ 包括一百多个层、数十种度量标准、损失函数、优化器和回调函数，以及 Keras 的训练和评估循环，以及 Keras 的保存和序列化基础设施。所有您熟悉和喜爱的 API 都在这里。

大规模模型训练和部署

新版本的 Keras 为大规模模型训练和部署提供了全新的能力。借助优化的算法和性能改进，现在您可以处理更大规模、更复杂的深度学习模型，而无需担心性能问题。

使用任何来源的数据管道。

Keras 3 的 fit()/evaluate()/predict()例程兼容 tf.data.Dataset 对象、PyTorch 的 DataLoader 对象、NumPy 数组和 Pandas 数据框，无论您使用的是哪个后端。您可以在 PyTorch 的 DataLoader 上训练 Keras 3 + TensorFlow 模型，或者在 tf.data.Dataset 上训练 Keras 3 + PyTorch 模型。

案例1：搭配Pytorch训练

https://keras.io/guides/custom_train_step_in_torch/

• 导入环境

  import os

  This guide can only be run with the torch backend.

  os.environ["KERAS_BACKEND"] = "torch"

  import torch
  import keras
  from keras import layers
  import numpy as np

• 定义模型

在 train_step() 方法的主体中，实现了一个常规的训练更新，类似于您已经熟悉的内容。重要的是，我们通过 self.compute_loss() 计算损失，它包装了传递给 compile() 的损失函数。

class CustomModel(keras.Model):
    def train_step(self, data):
        # Unpack the data. Its structure depends on your model and
        # on what you pass to `fit()`.
        x, y = data

        # Call torch.nn.Module.zero_grad() to clear the leftover gradients
        # for the weights from the previous train step.
        self.zero_grad()

        # Compute loss
        y_pred = self(x, training=True)  # Forward pass
        loss = self.compute_loss(y=y, y_pred=y_pred)

        # Call torch.Tensor.backward() on the loss to compute gradients
        # for the weights.
        loss.backward()

        trainable_weights = [v for v in self.trainable_weights]
        gradients = [v.value.grad for v in trainable_weights]

        # Update weights
        with torch.no_grad():
            self.optimizer.apply(gradients, trainable_weights)

        # Update metrics (includes the metric that tracks the loss)
        for metric in self.metrics:
            if metric.name == "loss":
                metric.update_state(loss)
            else:
                metric.update_state(y, y_pred)

        # Return a dict mapping metric names to current value
        # Note that it will include the loss (tracked in self.metrics).
        return {m.name: m.result() for m in self.metrics}

• 训练模型

  Construct and compile an instance of CustomModel

  inputs = keras.Input(shape=(32,))
  outputs = keras.layers.Dense(1)(inputs)
  model = CustomModel(inputs, outputs)
  model.compile(optimizer="adam", loss="mse", metrics=["mae"])

  Just use fit as usual

  x = np.random.random((1000, 32))
  y = np.random.random((1000, 1))
  model.fit(x, y, epochs=3)

案例2：自定义Pytorch流程

https://keras.io/guides/writing_a_custom_training_loop_in_torch/

• 导入环境

  import os

  This guide can only be run with the torch backend.

  os.environ["KERAS_BACKEND"] = "torch"

  import torch
  import keras
  from keras import layers
  import numpy as np

• 定义模型、加载数据集

  Let's consider a simple MNIST model

  def get_model():
  inputs = keras.Input(shape=(784,), name="digits")
  x1 = keras.layers.Dense(64, activation="relu")(inputs)
  x2 = keras.layers.Dense(64, activation="relu")(x1)
  outputs = keras.layers.Dense(10, name="predictions")(x2)
  model = keras.Model(inputs=inputs, outputs=outputs)
  return model

  Create load up the MNIST dataset and put it in a torch DataLoader

  Prepare the training dataset.

  batch_size = 32
  (x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data()
  x_train = np.reshape(x_train, (-1, 784)).astype("float32")
  x_test = np.reshape(x_test, (-1, 784)).astype("float32")
  y_train = keras.utils.to_categorical(y_train)
  y_test = keras.utils.to_categorical(y_test)

  Reserve 10,000 samples for validation.

  x_val = x_train[-10000:]
  y_val = y_train[-10000:]
  x_train = x_train[:-10000]
  y_train = y_train[:-10000]

  Create torch Datasets

  train_dataset = torch.utils.data.TensorDataset(
  torch.from_numpy(x_train), torch.from_numpy(y_train)
  )
  val_dataset = torch.utils.data.TensorDataset(
  torch.from_numpy(x_val), torch.from_numpy(y_val)
  )

  Create DataLoaders for the Datasets

  train_dataloader = torch.utils.data.DataLoader(
  train_dataset, batch_size=batch_size, shuffle=True
  )
  val_dataloader = torch.utils.data.DataLoader(
  val_dataset, batch_size=batch_size, shuffle=False
  )

• 定义优化器

  Instantiate a torch optimizer

  model = get_model()
  optimizer = torch.optim.Adam(model.parameters(), lr=1e-3)

  Instantiate a torch loss function

  loss_fn = torch.nn.CrossEntropyLoss()

• 训练模型

  epochs = 3
  for epoch in range(epochs):
  for step, (inputs, targets) in enumerate(train_dataloader):
  # Forward pass
  logits = model(inputs)
  loss = loss_fn(logits, targets)

        # Backward pass
        model.zero_grad()
        loss.backward()
  
        # Optimizer variable updates
        optimizer.step()
  
        # Log every 100 batches.
        if step % 100 == 0:
            print(
                f"Training loss (for 1 batch) at step {step}: {loss.detach().numpy():.4f}"
            )
            print(f"Seen so far: {(step + 1) * batch_size} samples")