Replace hub.Module with TF1 Hub format when referring to the TF1 Hub …

…file format.

PiperOrigin-RevId: 320047485

README.md

@@ -36,7 +36,7 @@ as well as other associated code and documentation.
   * The asset types of [tfhub.dev](https://tfhub.dev/)
       * [SavedModels for TensorFlow 2](docs/tf2_saved_model.md)
         and the [Reusable SavedModel interface](docs/reusable_saved_models.md).
-      * Deprecated: [hub.Modules for TensorFlow 1](docs/tf1_hub_module.md) and
+      * Deprecated: [Models in TF1 Hub format](docs/tf1_hub_module.md) and
         their [Common Signatures](docs/common_signatures/index.md) collection.
   * Using the library
       * [Installation](docs/installation.md)

docs/_book.yaml

@@ -32,7 +32,7 @@ upper_tabs:
       - heading: Asset types
       - title: SavedModels for TensorFlow 2
         path: /hub/tf2_saved_model
-      - title: "Deprecated: hub.Modules for TensorFlow 1"
+      - title: "Deprecated: TF1 Hub format"
         path: /hub/tf1_hub_module
         status: deprecated
       # Contributing to TF-Hub - either models or code

docs/common_issues.md

@@ -12,10 +12,10 @@ embed = hub.load('https://tfhub.dev/google/nnlm-en-dim128/1')
 embed(['my text', 'batch'])
 ```
 
-This error frequently arises when loading TF1 hub.Modules with the `hub.load()`
-API in TF2. Adding the correct signature should fix this problem. See the
-[TF-Hub migration guide for TF2](migration_tf2.md) for more details on moving to
-TF2 and the use of TF1 hub.Modules in TF2.
+This error frequently arises when loading models in TF1 Hub format with the
+`hub.load()` API in TF2. Adding the correct signature should fix this problem.
+See the [TF-Hub migration guide for TF2](migration_tf2.md) for more details on
+moving to TF2 and the use of models in TF1 Hub format in TF2.
 
 ```python
 

docs/migration_tf2.md

@@ -5,19 +5,19 @@ migrating your TensorFlow code from TensorFlow 1 to TensorFlow 2.
 It complements TensorFlow's general
 [migration guide](https://www.tensorflow.org/guide/migrate).
 
-For TF2, TF Hub has switched away from the
-[custom hub.Module format](tf1_hub_module.md) and its `hub.Module` API
-to the native [SavedModel format of TF2](tf2_saved_model.md)
-and its associated API of `hub.load()` and `hub.KerasLayer`.
+For TF2, TF Hub has switched away from the [TF1 Hub format](tf1_hub_module.md)
+and its `hub.Module` API to the native
+[SavedModel format of TF2](tf2_saved_model.md) and its associated API of
+`hub.load()` and `hub.KerasLayer`.
 
-The `hub.Module` API remains available in the `tensorflow_hub` library
-for use in TF1 and in the TF1 compatibility mode of TF2.
-It can only load assets in the hub.Module format.
+The `hub.Module` API remains available in the `tensorflow_hub` library for use
+in TF1 and in the TF1 compatibility mode of TF2. It can only load models in the
+TF1 Hub format.
 
 The new API of `hub.load()` (and `hub.KerasLayer`, which wraps it for Keras)
-works for TensorFlow 1.15 (in eager and graph mode) and in TensorFlow 2.
-This new API can load the new TF2 SavedModel assets, and, with
-the restrictions laid out below, for the legacy hub.Module assets.
+works for TensorFlow 1.15 (in eager and graph mode) and in TensorFlow 2. This
+new API can load the new TF2 SavedModel assets, and, with the restrictions laid
+out below, for the legacy models in TF1 Hub format.
 
 In general, it is recommended to use new API wherever possible.
 
@@ -48,8 +48,9 @@ Many tutorials show these APIs in action. See in particular
   * [Text classification example notebook](https://github.com/tensorflow/hub/blob/master/examples/colab/tf2_text_classification.ipynb)
   * [Image classification example notebook](https://github.com/tensorflow/hub/blob/master/examples/colab/tf2_image_retraining.ipynb)
 
-If the hub.Module you use has a newer version that comes in the TF2 SavedModel
-format, we recommend to switch the API and the module version at the same time.
+If the model in TF1 Hub format you use has a newer version that comes in the TF2
+SavedModel format, we recommend to switch the API and the module version at the
+same time.
 
 ### Using the new API in Estimator training
 
@@ -72,29 +73,28 @@ estimator = tf.estimator.Estimator(..., config=run_config)
 Starting with TF2.2, this option is no longer experimental, and
 the `.experimental` piece can be dropped.
 
+## Loading legacy models in TF1 Hub format
 
-## Loading legacy hub.Modules
-
-It can happen that a new TF2 SavedModel is not yet available for your
-use-case and you need to load an legacy hub.Module. Starting in `tensorflow_hub`
-release 0.7, you can use legacy hub.Modules together with `hub.KerasLayer` as
-shown below:
+It can happen that a new TF2 SavedModel is not yet available for your use-case
+and you need to load an legacy model in TF1 Hub format. Starting in
+`tensorflow_hub` release 0.7, you can use legacy model in TF1 Hub format
+together with `hub.KerasLayer` as shown below:
 
 ```python
 m = hub.KerasLayer(handle)
 tensor_out = m(tensor_in)
 ```
 
 Additionally `KerasLayer` exposes the ability to specify `tags`, `signature`,
-`output_key` and `signature_outputs_as_dict` for more specific usages of
-legacy hub.Modules and legacy SavedModels.
-
-Note: `trainable=True` is NOT supported when loading old hub.Modules.
+`output_key` and `signature_outputs_as_dict` for more specific usages of legacy
+models in TF1 Hub format and legacy SavedModels.
 
+Note: `trainable=True` is NOT supported when loading legacy models in TF1 Hub
+format.
 
 ## Using lower level APIs
 
-Old hub.Modules can be loaded via `tf.saved_model.load`. Instead of
+Legacy TF1 Hub format models can be loaded via `tf.saved_model.load`. Instead of
 
 ```python
 # DEPRECATED: TensorFlow 1
@@ -115,13 +115,13 @@ keyed by signature names. Calling such a function computes all its outputs,
 even if unused. (This is different from the lazy evaluation of TF1's
 graph mode.)
 
-## Retraining legacy hub.Modules
+## Retraining legacy models in TF1 Hub format
 
-Retraining legacy hub.Modules with the new APIs is not supported. This is due to
-them depending on `tf.saved_model.load` converting a `flat graph view` into
-an `object view` and dropping important details. Such as: trainable variables
-are imported as such, but update ops (for batch normalization etc.),
+Retraining legacy models in TF1 Hub format with the new APIs is not supported.
+This is due to them depending on `tf.saved_model.load` converting a `flat graph
+view` into an `object view` and dropping important details. Such as: trainable
+variables are imported as such, but update ops (for batch normalization etc.),
 regularization losses and cond/while contexts for differentiation are dropped.
 
-If you need to retrain legacy hub.Modules you will need to keep using the
-1.x APIs.
+If you need to retrain legacy models in TF1 Hub format you will need to keep
+using the 1.x APIs.

docs/tf1_hub_module.md

@@ -1,27 +1,24 @@
-# hub.Modules in TensorFlow 1
+# TF1 Hub format
 
-At its launch in 2018, TensorFlow Hub offered a single type of asset:
-hub.Modules for import into TensorFlow 1 programs.
+At its launch in 2018, TensorFlow Hub offered a single type of asset: TF1 Hub
+format for import into TensorFlow 1 programs.
 
-This page explains how to use hub.Modules in TensorFlow 1
-(or the TF1 compatibility mode of TF2) with the `hub.Module` class
-and associated APIs. (The typical use is to build a `tf.Graph`,
-possibly inside a TF1 `Estimator`, by combining one or more `hub.Modules`
-with `tf.compat.layers` or `tf.layers`).
+This page explains how to use TF1 Hub format in TF1 (or the TF1 compatibility
+mode of TF2) with the `hub.Module` class and associated APIs. (The typical use
+is to build a `tf.Graph`, possibly inside a TF1 `Estimator`, by combining one or
+more models in TF1 Hub format with `tf.compat.layers` or `tf.layers`).
 
 Users of TensorFlow 2 (outside TF1 compatibility mode) must use
-[the new API with `hub.load()` or `hub.KerasLayer`](tf2_saved_model.md).
-The new API loads the new TF2 SavedModel asset type, but also has
-limited [support for loading hub.Modules into TF2](migration_tf2.md).
+[the new API with `hub.load()` or `hub.KerasLayer`](tf2_saved_model.md). The new
+API loads the new TF2 SavedModel asset type, but also has limited
+[support for loading TF1 Hub format into TF2](migration_tf2.md).
 
+## Using a model in TF1 Hub format
 
-## Using a Module
+### Instantiating a model in TF1 Hub format
 
-### Instantiating a Module
-
-A hub.Module is imported into a TensorFlow program by
-creating a `hub.Module` object from a string with its URL or filesystem path,
-such as:
+A model in TF1 Hub format is imported into a TensorFlow program by creating a
+`hub.Module` object from a string with its URL or filesystem path, such as:
 
 ```python
 m = hub.Module("path/to/a/module_dir")
@@ -93,18 +90,16 @@ To this end, we maintain a collection of recommended
 
 ### Compatibility note
 
-The hub.Module format is geared towards TensorFlow 1.
-It is only partially supported by TF Hub in TensorFlow 2.
-Please do consider publishing in the new
+The TF1 Hub format is geared towards TensorFlow 1. It is only partially
+supported by TF Hub in TensorFlow 2. Please do consider publishing in the new
 [TF2 SavedModel](tf2_saved_model.md) format instead.
 
-The hub.Module format is similar to the SavedModel format of TensorFlow 1
-on a syntactic level (same file names and protocol messages) but semantically
+The TF1 Hub format is similar to the SavedModel format of TensorFlow 1 on a
+syntactic level (same file names and protocol messages) but semantically
 different to allow for module reuse, composition and re-training (e.g.,
-different storage of resource initializers, different tagging conventions
-for metagraphs). The easiest way to tell them apart on disk is the
-presence or absence of the `tfhub_module.pb` file.
-
+different storage of resource initializers, different tagging conventions for
+metagraphs). The easiest way to tell them apart on disk is the presence or
+absence of the `tfhub_module.pb` file.
 
 ### General approach
 

docs/tf2_saved_model.md

@@ -1,17 +1,17 @@
 # SavedModels from TF Hub in TensorFlow 2
 
-The [SavedModel format of
-TensorFlow 2](https://www.tensorflow.org/guide/saved_model)
-is the recommended way to share pre-trained models and model pieces
-on TensorFlow Hub. It replaces the older [hub.Module format for
-TensorFlow 1](tf1_hub_module.md) and comes with a new set of APIs.
-
-This page explains how to reuse TF2 SavedModels in a TensorFlow 2
-program with the low-level `hub.load()` API and its `hub.KerasLayer`
-wrapper. (Typically, `hub.KerasLayer` is combined with other `tf.keras.layers`
-to build a Keras model or the `model_fn` of a TF2 Estimator.)
-These APIs can also load the older hub.Modules for TF1, within limits,
-see the [migration guide](migration_tf2.md).
+The
+[SavedModel format of TensorFlow 2](https://www.tensorflow.org/guide/saved_model)
+is the recommended way to share pre-trained models and model pieces on
+TensorFlow Hub. It replaces the older [TF1 Hub format](tf1_hub_module.md) and
+comes with a new set of APIs.
+
+This page explains how to reuse TF2 SavedModels in a TensorFlow 2 program with
+the low-level `hub.load()` API and its `hub.KerasLayer` wrapper. (Typically,
+`hub.KerasLayer` is combined with other `tf.keras.layers` to build a Keras model
+or the `model_fn` of a TF2 Estimator.) These APIs can also load the legacy
+models in TF1 Hub format, within limits, see the
+[migration guide](migration_tf2.md).
 
 Users of TensorFlow 1 can update to TF 1.15 and then use the same APIs.
 Older versions of TF1 do not work.

examples/README.md

@@ -38,8 +38,8 @@ embeddings for real-time similarity matching and retrieval.
 
 #### [`colab/semantic_approximate_nearest_neighbors`](colab/semantic_approximate_nearest_neighbors.ipynb)
 
-This tutorial illustrates how to generate embeddings from a
-[TF1 legacy hub.Module](https://www.tensorflow.org/hub/tf1_hub_module) given
+This tutorial illustrates how to generate embeddings from a model in the legacy
+[TF1 Hub format](https://www.tensorflow.org/hub/tf1_hub_module) given
 input data and build an approximate nearest neighbours (ANN) index using the
 extracted embeddings for real-time similarity matching and retrieval.
 

examples/colab/tf2_image_retraining.ipynb

@@ -133,7 +133,7 @@
       "source": [
         "## Select the TF2 SavedModel module to use\n",
         "\n",
-        "For starters, use https://tfhub.dev/google/imagenet/mobilenet_v2_100_224/feature_vector/4. The same URL can be used in code to identify the SavedModel and in your browser to show its documentation. (Note that `hub.Modules` for TF 1.x won't work here.)"
+        "For starters, use https://tfhub.dev/google/imagenet/mobilenet_v2_100_224/feature_vector/4. The same URL can be used in code to identify the SavedModel and in your browser to show its documentation. (Note that models in TF1 Hub format won't work here.)"
       ]
     },
     {

tensorflow_hub/e2e_test.py

@@ -162,7 +162,7 @@ def test_load(self):
     if not hasattr(tf_v1.saved_model, "load_v2"):
       try:
         hub.load("@my/tf2_module/2")
-        self.fail("Failure expected. hub.module() not support in TF 1.x")
+        self.fail("Failure expected. hub.load() not supported in TF 1.x")
       except NotImplementedError:
         pass
     elif tf_v1.executing_eagerly():

tensorflow_hub/estimator.py

@@ -37,7 +37,7 @@
 def register_module_for_export(module, export_name):
   """Register a Module to be exported under `export_name`.
 
-  DEPRECATION NOTE: This belongs to the hub.Module API and file format for TF1.
+  DEPRECATION NOTE: This belongs to the hub.Module API and TF1 Hub format.
 
   This function registers `module` to be exported by `LatestModuleExporter`
   under a subdirectory named `export_name`.
@@ -64,7 +64,7 @@ def register_module_for_export(module, export_name):
 class LatestModuleExporter(tf_v1.estimator.Exporter):
   """Regularly exports registered modules into timestamped directories.
 
-  DEPRECATION NOTE: This belongs to the hub.Module API and file format for TF1.
+  DEPRECATION NOTE: This belongs to the hub.Module API and TF1 Hub format.
 
   Modules can be registered to be exported by this class by calling
   `register_module_for_export` when constructing the graph. The

tensorflow_hub/image_util.py

@@ -22,8 +22,8 @@
 from tensorflow_hub import native_module
 
 
-# hub.Modules for images can provide further information for the utilities
-# in this file by attaching an ImageModuleInfo message under this key.
+# Models in TF1 Hub format for images can provide further information for the
+# utilities in this file by attaching an ImageModuleInfo message under this key.
 IMAGE_MODULE_INFO_KEY = "image_module_info"
 
 
@@ -34,7 +34,7 @@
 def attach_image_module_info(image_module_info):
   """Attaches an ImageModuleInfo message from within a module_fn.
 
-  DEPRECATION NOTE: This belongs to the hub.Module API and file format for TF1.
+  DEPRECATION NOTE: This belongs to the hub.Module API and TF1 Hub format.
 
   Args:
     image_module_info: an ImageModuleInfo message.
@@ -45,7 +45,7 @@ def attach_image_module_info(image_module_info):
 def get_image_module_info(module_or_spec, required=False):
   """Returns the module's attached ImageModuleInfo message, or None if missing.
 
-  DEPRECATION NOTE: This belongs to the hub.Module API and file format for TF1.
+  DEPRECATION NOTE: This belongs to the hub.Module API and TF1 Hub format
 
   Args:
     module_or_spec: a hub.Module or module_spec object.