Fixed IntelTensorFlow_AMX_BF16_Inference (#2397)

* using new tf & keras version

* deleted models file, will download it from the Hub

Author: Ankur-singh

AI-and-Analytics/Features-and-Functionality/IntelTensorFlow_AMX_BF16_Inference/IntelTensorFlow_AMX_BF16_Inference.ipynb

@@ -32,7 +32,9 @@
     "from datetime import datetime\n",
     "import requests\n",
     "from copy import deepcopy\n",
-    "print(\"We are using Tensorflow version: \", tf.__version__)"
+    "\n",
+    "print(\"TF version:\", tf.__version__)\n",
+    "print(\"Hub version:\", hub.__version__)"
    ]
   },
   {
@@ -158,8 +160,34 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# import pre-trained fp_32 model\n",
-    "fp32_model = tf.keras.models.load_model('models/my_saved_model_fp32')"
+    "IMAGE_SIZE = (224, 224, 3)\n",
+    "model_handle = \"https://www.kaggle.com/models/google/resnet-v1/TensorFlow2/50-feature-vector/2\"\n",
+    "\n",
+    "print(\"Building model with\", model_handle)\n",
+    "fp32_model = tf.keras.Sequential([\n",
+    "    # Explicitly define the input shape so the model can be properly\n",
+    "    # loaded by the TFLiteConverter\n",
+    "    tf.keras.layers.InputLayer(input_shape=IMAGE_SIZE),\n",
+    "    hub.KerasLayer(model_handle, trainable=False),\n",
+    "    tf.keras.layers.Dropout(rate=0.2),\n",
+    "    tf.keras.layers.Dense(len(class_names),\n",
+    "                          kernel_regularizer=tf.keras.regularizers.l2(0.0001))\n",
+    "])\n",
+    "fp32_model.build((None,)+IMAGE_SIZE)\n",
+    "fp32_model.summary()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "a9e75958",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "fp32_model.compile(\n",
+    "  optimizer=tf.keras.optimizers.SGD(learning_rate=0.005, momentum=0.9), \n",
+    "  loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),\n",
+    "  metrics=['accuracy'])"
    ]
   },
   {
@@ -204,6 +232,24 @@
     "    break"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "id": "e1f94d18",
+   "metadata": {},
+   "source": [
+    "Lets save the model . . . "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "924ab886",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "fp32_model.save(\"models/my_saved_model_fp32\")"
+   ]
+  },
   {
    "cell_type": "markdown",
    "id": "c1d45604",
@@ -318,7 +364,7 @@
     "plt.title(\"Resnet50 Inference Time\")\n",
     "plt.xlabel(\"Test Case\")\n",
     "plt.ylabel(\"Inference Time (seconds)\")\n",
-    "plt.bar([\"FP32\", \"BF16 with AVX512\", \"BF16 with Intel® AMX\"], [fp32_inference_time, bf16_noAmx_inference_time, bf16_withAmx_inference_time])"
+    "plt.bar([\"FP32\", \"BF16 with AVX512\", \"BF16 with Intel® AMX\"], [fp32_inference_time, bf16_noAmx_inference_time, bf16_withAmx_inference_time]);"
    ]
   },
   {
@@ -347,7 +393,7 @@
     "plt.title(\"Intel® AMX Speedup\")\n",
     "plt.xlabel(\"Test Case\")\n",
     "plt.ylabel(\"Speedup\")\n",
-    "plt.bar([\"FP32\", \"BF16 with AVX512\", \"BF16 with Intel® AMX\"], [1, speedup_bf16_noAMX_from_fp32, speedup_bf16_withAMX_from_fp32])"
+    "plt.bar([\"FP32\", \"BF16 with AVX512\", \"BF16 with Intel® AMX\"], [1, speedup_bf16_noAMX_from_fp32, speedup_bf16_withAMX_from_fp32]);"
    ]
   },
   {
@@ -372,7 +418,7 @@
     "plt.title(\"Resnet50 Inference Accuracy\")\n",
     "plt.xlabel(\"Test Case\")\n",
     "plt.ylabel(\"Inference Accuracy\")\n",
-    "plt.bar([\"FP32\", \"BF16 with AVX512\", \"BF16 with Intel® AMX\"], [fp32_inference_accuracy, bf16_noAmx_inference_accuracy, bf16_withAmx_inference_accuracy])"
+    "plt.bar([\"FP32\", \"BF16 with AVX512\", \"BF16 with Intel® AMX\"], [fp32_inference_accuracy, bf16_noAmx_inference_accuracy, bf16_withAmx_inference_accuracy]);"
    ]
   },
   {