Merge pull request #1485 from rstudio/retether-3-8-0

Retether Keras 3.8.0

Author: t-kalinowski

.tether/man/Layer.txt

@@ -244,7 +244,7 @@ class Layer(keras.src.backend.tensorflow.layer.TFLayer, keras.src.ops.operation.
  |    autocast=True,
  |    regularizer=None,
  |    constraint=None,
- |    aggregation='mean',
+ |    aggregation='none',
  |    name=None
  |  )
  |      Add a weight variable to the layer.
@@ -271,10 +271,11 @@ class Layer(keras.src.backend.tensorflow.layer.TFLayer, keras.src.ops.operation.
  |          constraint: Contrainst object to call on the variable after any
  |              optimizer update, or string name of a built-in constraint.
  |              Defaults to `None`.
- |          aggregation: String, one of `'mean'`, `'sum'`,
- |              `'only_first_replica'`. Annotates the variable with the type
- |              of multi-replica aggregation to be used for this variable
- |              when writing custom data parallel training loops.
+ |          aggregation: Optional string, one of `None`, `"none"`, `"mean"`,
+ |              `"sum"` or `"only_first_replica"`. Annotates the variable with
+ |              the type of multi-replica aggregation to be used for this
+ |              variable when writing custom data parallel training loops.
+ |              Defaults to `"none"`.
  |          name: String name of the variable. Useful for debugging purposes.
  |
  |  build(self, input_shape)

.tether/man/activation_sparse_plus.txt

@@ -0,0 +1,14 @@
+__signature__
+keras.activations.sparse_plus(x)
+__doc__
+SparsePlus activation function.
+
+SparsePlus is defined as:
+
+`sparse_plus(x) = 0` for `x <= -1`.
+`sparse_plus(x) = (1/4) * (x + 1)^2` for `-1 < x < 1`.
+`sparse_plus(x) = x` for `x >= 1`.
+
+Args:
+    x: Input tensor.
+

.tether/man/activation_sparsemax.txt

@@ -0,0 +1,22 @@
+__signature__
+keras.activations.sparsemax(x, axis=-1)
+__doc__
+Sparsemax activation function.
+
+For each batch `i`, and class `j`,
+sparsemax activation function is defined as:
+
+`sparsemax(x)[i, j] = max(x[i, j] - τ(x[i, :]), 0).`
+
+Args:
+    x: Input tensor.
+    axis: `int`, axis along which the sparsemax operation is applied.
+
+Returns:
+    A tensor, output of sparsemax transformation. Has the same type and
+    shape as `x`.
+
+Reference:
+
+- [Martins et.al., 2016](https://arxiv.org/abs/1602.02068)
+

.tether/man/activation_threshold.txt

@@ -0,0 +1,19 @@
+__signature__
+keras.activations.threshold(
+  x,
+  threshold,
+  default_value
+)
+__doc__
+Threshold activation function.
+
+It is defined as:
+
+`threshold(x) = x` if `x > threshold`,
+`threshold(x) = default_value` otherwise.
+
+Args:
+    x: Input tensor.
+    threshold: The value that decides when to retain or replace x.
+    default_value: Value to assign when `x <= threshold`.
+

.tether/man/callback_backup_and_restore.txt

@@ -33,6 +33,7 @@ class BackupAndRestore(keras.src.callbacks.callback.Callback)
  |  >>> callback = keras.callbacks.BackupAndRestore(backup_dir="/tmp/backup")
  |  >>> model = keras.models.Sequential([keras.layers.Dense(10)])
  |  >>> model.compile(keras.optimizers.SGD(), loss='mse')
+ |  >>> model.build(input_shape=(None, 20))
  |  >>> try:
  |  ...   model.fit(np.arange(100).reshape(5, 20), np.zeros(5), epochs=10,
  |  ...             batch_size=1, callbacks=[callback, InterruptingCallback()],

.tether/man/callback_model_checkpoint.txt

@@ -64,12 +64,13 @@ class ModelCheckpoint(keras.src.callbacks.callback.Callback)
  |          which will be filled the value of `epoch` and keys in `logs`
  |          (passed in `on_epoch_end`).
  |          The `filepath` name needs to end with `".weights.h5"` when
- |          `save_weights_only=True` or should end with `".keras"` when
- |          checkpoint saving the whole model (default).
+ |          `save_weights_only=True` or should end with `".keras"` or `".h5"`
+ |          when checkpoint saving the whole model (default).
  |          For example:
- |          if `filepath` is `"{epoch:02d}-{val_loss:.2f}.keras"`, then the
- |          model checkpoints will be saved with the epoch number and the
- |          validation loss in the filename. The directory of the filepath
+ |          if `filepath` is `"{epoch:02d}-{val_loss:.2f}.keras"` or
+ |          "{epoch:02d}-{val_loss:.2f}.weights.h5"`, then the model
+ |          checkpoints will be saved with the epoch number and the validation
+ |          loss in the filename. The directory of the filepath
  |          should not be reused by any other callbacks to avoid conflicts.
  |      monitor: The metric name to monitor. Typically the metrics are set by
  |          the `Model.compile` method. Note:
@@ -183,3 +184,4 @@ class ModelCheckpoint(keras.src.callbacks.callback.Callback)
  |          batch: Integer, index of batch within the current epoch.
  |          logs: Dict. Aggregated metric results up until this batch.
  |
+

.tether/man/export_savedmodel.keras.src.models.model.Model.txt

@@ -3,40 +3,65 @@ keras.Model.export(
   self,
   filepath,
   format='tf_saved_model',
-  verbose=True
+  verbose=True,
+  input_signature=None,
+  **kwargs
 )
 __doc__
-Create a TF SavedModel artifact for inference.
+Export the model as an artifact for inference.
 
-**Note:** This can currently only be used with
-the TensorFlow or JAX backends.
-
-This method lets you export a model to a lightweight SavedModel artifact
-that contains the model's forward pass only (its `call()` method)
-and can be served via e.g. TF-Serving. The forward pass is registered
-under the name `serve()` (see example below).
+Args:
+    filepath: `str` or `pathlib.Path` object. The path to save the
+        artifact.
+    format: `str`. The export format. Supported values:
+        `"tf_saved_model"` and `"onnx"`.  Defaults to
+        `"tf_saved_model"`.
+    verbose: `bool`. Whether to print a message during export. Defaults
+        to `True`.
+    input_signature: Optional. Specifies the shape and dtype of the
+        model inputs. Can be a structure of `keras.InputSpec`,
+        `tf.TensorSpec`, `backend.KerasTensor`, or backend tensor. If
+        not provided, it will be automatically computed. Defaults to
+        `None`.
+    **kwargs: Additional keyword arguments:
+        - Specific to the JAX backend and `format="tf_saved_model"`:
+            - `is_static`: Optional `bool`. Indicates whether `fn` is
+                static. Set to `False` if `fn` involves state updates
+                (e.g., RNG seeds and counters).
+            - `jax2tf_kwargs`: Optional `dict`. Arguments for
+                `jax2tf.convert`. See the documentation for
+                [`jax2tf.convert`](
+                    https://github.com/google/jax/blob/main/jax/experimental/jax2tf/README.md).
+                If `native_serialization` and `polymorphic_shapes` are
+                not provided, they will be automatically computed.
 
-The original code of the model (including any custom layers you may
-have used) is *no longer* necessary to reload the artifact -- it is
-entirely standalone.
+**Note:** This feature is currently supported only with TensorFlow, JAX
+and Torch backends.
 
-Args:
-    filepath: `str` or `pathlib.Path` object. Path where to save
-        the artifact.
-    verbose: whether to print all the variables of the exported model.
+Examples:
 
-Example:
+Here's how to export a TensorFlow SavedModel for inference.
 
 ```python
-# Create the artifact
-model.export("path/to/location")
+# Export the model as a TensorFlow SavedModel artifact
+model.export("path/to/location", format="tf_saved_model")
 
-# Later, in a different process/environment...
+# Load the artifact in a different process/environment
 reloaded_artifact = tf.saved_model.load("path/to/location")
 predictions = reloaded_artifact.serve(input_data)
 ```
 
-If you would like to customize your serving endpoints, you can
-use the lower-level `keras.export.ExportArchive` class. The
-`export()` method relies on `ExportArchive` internally.
+Here's how to export an ONNX for inference.
+
+```python
+# Export the model as a ONNX artifact
+model.export("path/to/location", format="onnx")
+
+# Load the artifact in a different process/environment
+ort_session = onnxruntime.InferenceSession("path/to/location")
+ort_inputs = {
+    k.name: v for k, v in zip(ort_session.get_inputs(), input_data)
+}
+predictions = ort_session.run(None, ort_inputs)
+```
 

.tether/man/keras.activations.txt

@@ -30,8 +30,15 @@ soft_shrink(x, threshold=0.5)
 softmax(x, axis=-1)
 softplus(x)
 softsign(x)
+sparse_plus(x)
+sparsemax(x, axis=-1)
 squareplus(x, b=4)
 swish(x)
 tanh(x)
 tanh_shrink(x)
+threshold(
+  x,
+  threshold,
+  default_value
+)
 

.tether/man/keras.layers.txt

@@ -542,6 +542,12 @@ Embedding(
   lora_rank=None,
   **kwargs
 )
+Equalization(
+  value_range=(0, 255),
+  bins=256,
+  data_format=None,
+  **kwargs
+)
 Flatten(data_format=None, **kwargs)
 FlaxLayer(
   module,
@@ -913,6 +919,12 @@ MelSpectrogram(
 )
 minimum(inputs, **kwargs)
 Minimum(**kwargs)
+MixUp(
+  alpha=0.2,
+  data_format=None,
+  seed=None,
+  **kwargs
+)
 MultiHeadAttention(
   num_heads,
   key_dim,
@@ -950,14 +962,41 @@ PReLU(
   shared_axes=None,
   **kwargs
 )
+RandAugment(
+  value_range=(0, 255),
+  num_ops=2,
+  factor=0.5,
+  interpolation='bilinear',
+  seed=None,
+  data_format=None,
+  **kwargs
+)
 RandomBrightness(
   factor,
   value_range=(0, 255),
   seed=None,
   **kwargs
 )
+RandomColorDegeneration(
+  factor,
+  value_range=(0, 255),
+  data_format=None,
+  seed=None,
+  **kwargs
+)
+RandomColorJitter(
+  value_range=(0, 255),
+  brightness_factor=None,
+  contrast_factor=None,
+  saturation_factor=None,
+  hue_factor=None,
+  seed=None,
+  data_format=None,
+  **kwargs
+)
 RandomContrast(
   factor,
+  value_range=(0, 255),
   seed=None,
   **kwargs
 )
@@ -975,6 +1014,26 @@ RandomFlip(
   data_format=None,
   **kwargs
 )
+RandomGrayscale(
+  factor=0.5,
+  data_format=None,
+  seed=None,
+  **kwargs
+)
+RandomHue(
+  factor,
+  value_range=(0, 255),
+  data_format=None,
+  seed=None,
+  **kwargs
+)
+RandomPosterization(
+  factor,
+  value_range=(0, 255),
+  data_format=None,
+  seed=None,
+  **kwargs
+)
 RandomRotation(
   factor,
   fill_mode='reflect',
@@ -984,6 +1043,30 @@ RandomRotation(
   data_format=None,
   **kwargs
 )
+RandomSaturation(
+  factor,
+  value_range=(0, 255),
+  data_format=None,
+  seed=None,
+  **kwargs
+)
+RandomSharpness(
+  factor,
+  value_range=(0, 255),
+  data_format=None,
+  seed=None,
+  **kwargs
+)
+RandomShear(
+  x_factor=0.0,
+  y_factor=0.0,
+  interpolation='bilinear',
+  fill_mode='reflect',
+  fill_value=0.0,
+  data_format=None,
+  seed=None,
+  **kwargs
+)
 RandomTranslation(
   height_factor,
   width_factor,

.tether/man/keras.losses.txt

@@ -210,13 +210,15 @@ tversky(
   y_true,
   y_pred,
   alpha=0.5,
-  beta=0.5
+  beta=0.5,
+  axis=None
 )
 Tversky(
   alpha=0.5,
   beta=0.5,
   reduction='sum_over_batch_size',
   name='tversky',
+  axis=None,
   dtype=None
 )