Minor changes to improve terminology consistency (DTensor colab notebooks)

PiperOrigin-RevId: 446005959

Author: tensorflower-gardener

site/en/tutorials/distribute/dtensor_keras_tutorial.ipynb

@@ -13,6 +13,7 @@
       "cell_type": "code",
       "execution_count": null,
       "metadata": {
+        "cellView": "form",
         "id": "tuOe1ymfHZPu"
       },
       "outputs": [],
@@ -36,7 +37,7 @@
         "id": "MT-LkFOl2axM"
       },
       "source": [
-        "# DTensor Integration with Keras"
+        "# Using DTensors with Keras"
       ]
     },
     {
@@ -739,32 +740,17 @@
         "\n",
         "print(model.layers[2].kernel.layout)"
       ]
-    },
-    {
-      "cell_type": "code",
-      "execution_count": null,
-      "metadata": {
-        "id": "00dPVoSlRLFA"
-      },
-      "outputs": [],
-      "source": [
-        ""
-      ]
     }
   ],
   "metadata": {
     "colab": {
       "collapsed_sections": [],
       "name": "dtensor_keras_tutorial.ipynb",
-      "provenance": [],
       "toc_visible": true
     },
     "kernelspec": {
       "display_name": "Python 3",
       "name": "python3"
-    },
-    "language_info": {
-      "name": "python"
     }
   },
   "nbformat": 4,

site/en/tutorials/distribute/dtensor_ml_tutorial.ipynb

@@ -13,6 +13,7 @@
       "cell_type": "code",
       "execution_count": null,
       "metadata": {
+        "cellView": "form",
         "id": "tuOe1ymfHZPu"
       },
       "outputs": [],
@@ -36,7 +37,7 @@
         "id": "MfBg1C5NB3X0"
       },
       "source": [
-        "# DTensor Maching Learning Tutorial\n"
+        "# Distributed Training with DTensors\n"
       ]
     },
     {
@@ -75,7 +76,7 @@
         " \n",
         " - Data Parallel training, where the training samples are sharded (partitioned) to devices.\n",
         " - Model Parallel training, where the model variables are sharded to devices. \n",
-        " - Spatial Parallel training, where the features of input data are  sharded to devices.\n",
+        " - Spatial Parallel training, where the features of input data are  sharded to devices. (Also known as [Spatial Partitioning](https://cloud.google.com/blog/products/ai-machine-learning/train-ml-models-on-large-images-and-3d-volumes-with-spatial-partitioning-on-cloud-tpus))\n",
         "\n",
         "The training portion of this tutorial is inspired [A Kaggle guide on Sentiment Analysis](https://www.kaggle.com/code/anasofiauzsoy/yelp-review-sentiment-analysis-tensorflow-tfds/notebook) notebook. To learn about the complete training and evaluation workflow (without DTensor), refer to that notebook. \n",
         "\n",
@@ -237,8 +238,7 @@
         "    'y': dataset_y,\n",
         "})\n",
         "\n",
-        "dataset.take(1).get_single_element()\n",
-        "\n"
+        "dataset.take(1).get_single_element()\n"
       ]
     },
     {
@@ -297,7 +297,6 @@
         "id": "PMCt-Gj3b3Jy"
       },
       "source": [
-        "\n",
         "### Dense Layer\n",
         "\n",
         "The following custom Dense layer defines 2 layer variables: $W_{ij}$ is the variable for weights, and $b_i$ is the variable for the biases.\n",
@@ -809,8 +808,7 @@
         "- The 2 devices within a single model replica receive replicated training data.\n",
         "\n",
         "\n",
-        "<img src=\"https://www.tensorflow.org/tutorials/distribute/images/dtensor_model_para.png\" alt=\"Model parallel mesh\" class=\"no-filter\">\n",
-        "\n"
+        "<img src=\"https://www.tensorflow.org/tutorials/distribute/images/dtensor_model_para.png\" alt=\"Model parallel mesh\" class=\"no-filter\">\n"
       ]
     },
     {
@@ -905,7 +903,7 @@
         "id": "u-bK6IZ9GCS9"
       },
       "source": [
-        "When training data of very high dimensionality (e.g. a very large image or a video), it may be desirable to shard along the feature dimension. This is called Spatial Parallel training.\n",
+        "When training data of very high dimensionality (e.g. a very large image or a video), it may be desirable to shard along the feature dimension. This is called [Spatial Partitioning](https://cloud.google.com/blog/products/ai-machine-learning/train-ml-models-on-large-images-and-3d-volumes-with-spatial-partitioning-on-cloud-tpus), which was first introduced into TensorFlow for training models with large 3-d input samples.\n",
         "\n",
         "<img src=\"https://www.tensorflow.org/tutorials/distribute/images/dtensor_spatial_para.png\" alt=\"Spatial parallel mesh\" class=\"no-filter\">\n",
         "\n",
@@ -1067,32 +1065,17 @@
         "Composing a model with `tf.Module` from scratch is a lot of work, and reusing existing building blocks such as layers and helper functions can drastically speed up model development.\n",
         "As of TensorFlow 2.9, all Keras Layers under `tf.keras.layers` accepts DTensor layouts as their arguments, and can be used to build DTensor models. You can even directly reuse a Keras model with DTensor without modifying the model implementation. Refer to the [DTensor Keras Integration Tutorial](link) (TODO: add link) for information on using DTensor Keras. "
       ]
-    },
-    {
-      "cell_type": "code",
-      "execution_count": null,
-      "metadata": {
-        "id": "A-YWPfJyHPcX"
-      },
-      "outputs": [],
-      "source": [
-        ""
-      ]
     }
   ],
   "metadata": {
     "colab": {
       "collapsed_sections": [],
       "name": "dtensor_ml_tutorial.ipynb",
-      "provenance": [],
       "toc_visible": true
     },
     "kernelspec": {
       "display_name": "Python 3",
       "name": "python3"
-    },
-    "language_info": {
-      "name": "python"
     }
   },
   "nbformat": 4,