Add additional background for why strip_pruning and gzip are relevant in exporting compressed models.

PiperOrigin-RevId: 325046042

Author: alanchiao

tensorflow_model_optimization/g3doc/guide/pruning/pruning_with_keras.ipynb

@@ -12,7 +12,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 0,
+      "execution_count": null,
       "metadata": {
         "cellView": "form",
         "colab": {},
@@ -107,7 +107,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 0,
+      "execution_count": null,
       "metadata": {
         "colab": {},
         "colab_type": "code",
@@ -120,7 +120,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 0,
+      "execution_count": null,
       "metadata": {
         "colab": {},
         "colab_type": "code",
@@ -151,7 +151,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 0,
+      "execution_count": null,
       "metadata": {
         "colab": {},
         "colab_type": "code",
@@ -202,7 +202,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 0,
+      "execution_count": null,
       "metadata": {
         "colab": {},
         "colab_type": "code",
@@ -257,7 +257,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 0,
+      "execution_count": null,
       "metadata": {
         "colab": {},
         "colab_type": "code",
@@ -319,7 +319,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 0,
+      "execution_count": null,
       "metadata": {
         "colab": {},
         "colab_type": "code",
@@ -351,7 +351,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 0,
+      "execution_count": null,
       "metadata": {
         "colab": {},
         "colab_type": "code",
@@ -378,7 +378,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 0,
+      "execution_count": null,
       "metadata": {
         "colab": {},
         "colab_type": "code",
@@ -420,12 +420,16 @@
         "Both `tfmot.sparsity.keras.strip_pruning` and applying a standard compression algorithm (e.g. via gzip) are necessary to see the compression\n",
         "benefits of pruning.\n",
         "\n",
+        "*   `strip_pruning` is necessary since it removes every tf.Variable that pruning only needs during training, which would otherwise add to model size during inference\n",
+        "*   Applying a standard compression algorithm is necessary since the serialized weight matrices are the same size as they were before pruning. However, pruning makes most of the weights zeros, which is\n",
+        "added redundancy that algorithms can utilize to further compress the model.\n",
+        "\n",
         "First, create a compressible model for TensorFlow."
       ]
     },
     {
       "cell_type": "code",
-      "execution_count": 0,
+      "execution_count": null,
       "metadata": {
         "colab": {},
         "colab_type": "code",
@@ -452,7 +456,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 0,
+      "execution_count": null,
       "metadata": {
         "colab": {},
         "colab_type": "code",
@@ -483,7 +487,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 0,
+      "execution_count": null,
       "metadata": {
         "colab": {},
         "colab_type": "code",
@@ -515,7 +519,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 0,
+      "execution_count": null,
       "metadata": {
         "colab": {},
         "colab_type": "code",
@@ -550,7 +554,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 0,
+      "execution_count": null,
       "metadata": {
         "colab": {},
         "colab_type": "code",
@@ -595,7 +599,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 0,
+      "execution_count": null,
       "metadata": {
         "colab": {},
         "colab_type": "code",
@@ -647,7 +651,7 @@
     },
     {
       "cell_type": "code",
-      "execution_count": 0,
+      "execution_count": null,
       "metadata": {
         "colab": {},
         "colab_type": "code",