Skip to content

Instantly share code, notes, and snippets.

@rozeappletree

rozeappletree/swift-keras-tflite-mlir.ipynb

Created January 4, 2020 01:21
Show Gist options
  • Save rozeappletree/6f388189724199554629f203542770e4 to your computer and use it in GitHub Desktop.
Save rozeappletree/6f388189724199554629f203542770e4 to your computer and use it in GitHub Desktop.
Download ZIP
swift-keras-tflite-mlir.ipynb
Display the source blob
Display the rendered blob
Raw
swift-keras-tflite-mlir.ipynb
{
"nbformat": 4,
"nbformat_minor": 0,
"metadata": {
"colab": {
"name": "swift-keras-tflite-mlir.ipynb",
"provenance": [],
"collapsed_sections": [],
"include_colab_link": true
},
"kernelspec": {
"name": "swift",
"display_name": "Swift"
}
},
"cells": [
{
"cell_type": "markdown",
"metadata": {
"id": "view-in-github",
"colab_type": "text"
},
"source": [
"<a href=\"https://colab.research.google.com/gist/rakesh4real/6f388189724199554629f203542770e4/swift-keras-tflite-mlir.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
]
},
{
"cell_type": "code",
"metadata": {
"id": "DBX6ou638P3w",
"colab_type": "code",
"outputId": "4017f10d-53c4-4f75-ffbd-59c2d4eb8f0b",
"colab": {
"base_uri": "https://localhost:8080/",
"height": 578
}
},
"source": [
"%install-location $cwd/swift-install\n",
"%install '.package(url: \"https://github.com/tensorflow/swift-models\", .branch(\"master\"))' Datasets"
],
"execution_count": 1,
"outputs": [
{
"output_type": "stream",
"text": [
"Installing packages:\n",
"\t.package(url: \"https://github.com/tensorflow/swift-models\", .branch(\"master\"))\n",
"\t\tDatasets\n",
"With SwiftPM flags: []\n",
"Working in: /tmp/tmpci8y9p4v/swift-install\n",
"Fetching https://github.com/tensorflow/swift-models\n",
"Fetching https://github.com/kylef/Commander.git\n",
"Fetching https://github.com/kylef/Spectre.git\n",
"Cloning https://github.com/kylef/Commander.git\n",
"Resolving https://github.com/kylef/Commander.git at 0.9.1\n",
"Cloning https://github.com/kylef/Spectre.git\n",
"Resolving https://github.com/kylef/Spectre.git at 0.9.0\n",
"Cloning https://github.com/tensorflow/swift-models\n",
"Resolving https://github.com/tensorflow/swift-models at master\n",
"[1/7] Compiling ModelSupport Stderr.swift\n",
"[2/7] Compiling ModelSupport FileManagement.swift\n",
"[3/7] Compiling ModelSupport Image.swift\n",
"[4/8] Merging module ModelSupport\n",
"[5/14] Wrapping AST for ModelSupport for debugging\n",
"[6/14] Compiling Datasets CIFAR10.swift\n",
"[7/14] Compiling Datasets DatasetUtilities.swift\n",
"[8/14] Compiling Datasets ImageClassificationDataset.swift\n",
"[9/14] Compiling Datasets Imagenette.swift\n",
"[10/14] Compiling Datasets LabeledExample.swift\n",
"[11/14] Compiling Datasets MNIST.swift\n",
"[12/15] Merging module Datasets\n",
"[13/16] Wrapping AST for Datasets for debugging\n",
"[14/16] Compiling jupyterInstalledPackages jupyterInstalledPackages.swift\n",
"[15/17] Merging module jupyterInstalledPackages\n",
"[16/17] Wrapping AST for jupyterInstalledPackages for debugging\n",
"[17/17] Linking libjupyterInstalledPackages.so\n",
"Initializing Swift...\n",
"Installation complete!\n"
],
"name": "stdout"
}
]
},
{
"cell_type": "code",
"metadata": {
"id": "e32OUw3UxpuH",
"colab_type": "code",
"colab": {}
},
"source": [
"import Python\n",
"import TensorFlow"
],
"execution_count": 0,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "Hrh7BrNR31Nx",
"colab_type": "code",
"colab": {}
},
"source": [
"// ----------------------------------\n",
"// Train CIFAR model\n",
"// ----------------------------------\n",
"\n",
"struct CIFAR_Keras: Layer {\n",
" typealias Input = Tensor<Float>\n",
" typealias Output = Tensor<Float>\n",
"\n",
" var conv1a = Conv2D<Float>(filterShape: (3, 3, 3, 32), padding: .same, activation: relu)\n",
" var conv1b = Conv2D<Float>(filterShape: (3, 3, 32, 32), activation: relu)\n",
" var pool1 = MaxPool2D<Float>(poolSize: (2, 2), strides: (2, 2))\n",
" // var dropout1 = Dropout<Float>(probability: 0.25)\n",
"\n",
" var conv2a = Conv2D<Float>(filterShape: (3, 3, 32, 64), padding: .same, activation: relu)\n",
" var conv2b = Conv2D<Float>(filterShape: (3, 3, 64, 64), activation: relu)\n",
" var pool2 = MaxPool2D<Float>(poolSize: (2, 2), strides: (2, 2))\n",
" // var dropout2 = Dropout<Float>(probability: 0.25)\n",
"\n",
" var flatten = Flatten<Float>()\n",
" var dense1 = Dense<Float>(inputSize: 64 * 6 * 6, outputSize: 512, activation: relu)\n",
" //var dropout3 = Dropout<Float>(probability: 0.5)\n",
" var dense2 = Dense<Float>(inputSize: 512, outputSize: 512, activation: relu)\n",
" var output = Dense<Float>(inputSize: 512, outputSize: 10, activation: softmax)\n",
"\n",
" @differentiable\n",
" func callAsFunction(_ input: Input) -> Output {\n",
" let conv1 = input.sequenced(through: conv1a, conv1b, pool1) //, dropout1)\n",
" let conv2 = conv1.sequenced(through: conv2a, conv2b, pool2) //, dropout2)\n",
" return conv2.sequenced(through: flatten, dense1, dense2, output) //dropout3, dense2)\n",
" }\n",
"}"
],
"execution_count": 0,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "ngqpop4c5h80",
"colab_type": "code",
"outputId": "863a24af-f85d-4c13-8b48-1c08c3590c49",
"colab": {
"base_uri": "https://localhost:8080/",
"height": 320
}
},
"source": [
"// TRAIN\n",
"\n",
"import Datasets\n",
"\n",
"let batchSize = 100\n",
"\n",
"let dataset = CIFAR10()\n",
"let testBatches = dataset.testDataset.batched(batchSize)\n",
"\n",
"var model = CIFAR_Keras()\n",
"let optimizer = RMSProp(for: model, learningRate: 0.0001, decay: 1e-6)\n",
"\n",
"print(\"Starting training...\")\n",
"\n",
"for epoch in 1...20 {\n",
" Context.local.learningPhase = .training\n",
" var trainingLossSum: Float = 0\n",
" var trainingBatchCount = 0\n",
" let trainingShuffled = dataset.trainingDataset.shuffled(\n",
" sampleCount: dataset.trainingExampleCount, randomSeed: Int64(epoch))\n",
" for batch in trainingShuffled.batched(batchSize) {\n",
" let (labels, images) = (batch.label, batch.data)\n",
" let (loss, gradients) = valueWithGradient(at: model) { model -> Tensor<Float> in\n",
" let logits = model(images)\n",
" return softmaxCrossEntropy(logits: logits, labels: labels)\n",
" }\n",
" trainingLossSum += loss.scalarized()\n",
" trainingBatchCount += 1\n",
" optimizer.update(&model, along: gradients)\n",
" }\n",
"\n",
" Context.local.learningPhase = .inference\n",
" var testLossSum: Float = 0\n",
" var testBatchCount = 0\n",
" var correctGuessCount = 0\n",
" var totalGuessCount = 0\n",
" for batch in testBatches {\n",
" let (labels, images) = (batch.label, batch.data)\n",
" let logits = model(images)\n",
" testLossSum += softmaxCrossEntropy(logits: logits, labels: labels).scalarized()\n",
" testBatchCount += 1\n",
"\n",
" let correctPredictions = logits.argmax(squeezingAxis: 1) .== labels\n",
" correctGuessCount = correctGuessCount\n",
" + Int(\n",
" Tensor<Int32>(correctPredictions).sum().scalarized())\n",
" totalGuessCount = totalGuessCount + batchSize\n",
" }\n",
"\n",
" let accuracy = Float(correctGuessCount) / Float(totalGuessCount)\n",
" print(\n",
" \"\"\"\n",
" [Epoch \\(epoch)] \\\n",
" Accuracy: \\(correctGuessCount)/\\(totalGuessCount) (\\(accuracy)) \\\n",
" Loss: \\(testLossSum / Float(testBatchCount))\n",
" \"\"\"\n",
" )\n",
"}"
],
"execution_count": 4,
"outputs": [
{
"output_type": "stream",
"text": [
"Preparing CIFAR dataset...\r\n",
"Archive missing, downloading...\n",
"Archive downloaded, processing...\n",
"Unarchiving completed\n",
"Starting training...\n",
"[Epoch 1] Accuracy: 4236/10000 (0.4236) Loss: 2.0377867\n",
"[Epoch 2] Accuracy: 4666/10000 (0.4666) Loss: 1.9923123\n",
"[Epoch 3] Accuracy: 4818/10000 (0.4818) Loss: 1.9769447\n",
"[Epoch 4] Accuracy: 5272/10000 (0.5272) Loss: 1.9336563\n",
"[Epoch 5] Accuracy: 5492/10000 (0.5492) Loss: 1.9105333\n"
],
"name": "stdout"
},
{
"output_type": "error",
"ename": "",
"evalue": "ignored",
"traceback": [
"Current stack trace:",
"\tframe #30: 0x00007f780dac7443 $__lldb_expr30`AD__$s14__lldb_expr_2911CIFAR_KerasV14callAsFunctiony10TensorFlow0H0VySfGAHF__pullback_src_0_wrt_1 at <Cell 3>:27:27",
"\tframe #32: 0x00007f780da2aa7c $__lldb_expr36`partial apply for AD__$s14__lldb_expr_3510TensorFlow0C0VySfG02__a1_B3_2911CIFAR_KerasVcfU___pullback_src_0_wrt_0 [inlined] AD__$s14__lldb_expr_3510TensorFlow0C0VySfG02__a1_B3_2911CIFAR_KerasVcfU___pullback_src_0_wrt_0 at <Cell 4>:24:26",
"\tframe #40: 0x00007f780da206c6 $__lldb_expr36`main at <Cell 4>:23:33"
]
}
]
},
{
"cell_type": "code",
"metadata": {
"id": "nn7yv1cDGayR",
"colab_type": "code",
"colab": {}
},
"source": [
"// `model` is the swift trained model\n",
"// Note: var model = CIFAR_Keras() in `TRAIN` section\n",
"// `model` has all the required weights which will\n",
"// be copied to python imported keras model\n",
"let swift_model = model"
],
"execution_count": 0,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "kZRlD4utdPuX",
"colab_type": "code",
"colab": {}
},
"source": [
"func save_swift_kerash5(){\n",
" let keras = Python.import(\"keras\")\n",
" let model = keras.Sequential()\n",
"\n",
" let _conv2D = keras.layers.Conv2D\n",
" let _maxPool2D = keras.layers.MaxPooling2D\n",
" let _flatten = keras.layers.Flatten\n",
" let _dense = keras.layers.Dense\n",
"\n",
" // 1\n",
" model.add(\n",
" _conv2D(\n",
" filters: 32,\n",
" kernel_size: [3, 3],\n",
" padding: \"same\",\n",
" activation: \"relu\",\n",
" input_shape: [32, 32, 3]\n",
" )\n",
" )\n",
" model.add(\n",
" _conv2D(\n",
" filters: 32,\n",
" kernel_size: [3, 3],\n",
" activation: \"relu\"\n",
" )\n",
" )\n",
" model.add(\n",
" _maxPool2D(\n",
" pool_size: [2, 2],\n",
" strides: [2, 2]\n",
" )\n",
" )\n",
"\n",
" // 2\n",
" model.add(\n",
" _conv2D(\n",
" filters: 64,\n",
" kernel_size: [3, 3],\n",
" padding: \"same\",\n",
" activation: \"relu\"\n",
" )\n",
" )\n",
" model.add(\n",
" _conv2D(\n",
" filters: 64,\n",
" kernel_size: [3, 3],\n",
" activation: \"relu\"\n",
" )\n",
" )\n",
" model.add(\n",
" _maxPool2D(\n",
" pool_size: [2, 2],\n",
" strides: [2, 2]\n",
" )\n",
" )\n",
"\n",
" // then\n",
" model.add(_flatten())\n",
" model.add(_dense(512, activation: \"relu\", input_shape: [2304]))\n",
" model.add(_dense(512, activation: \"relu\"))\n",
" model.add(_dense(10, activation: \"softmax\"))\n",
"\n",
" // print summary\n",
" print(model.summary())\n",
"\n",
" model.layers[0].set_weights([swift_model.conv1a.filter.makeNumpyArray(), swift_model.conv1a.bias.makeNumpyArray()])\n",
" model.layers[1].set_weights([swift_model.conv1b.filter.makeNumpyArray(), swift_model.conv1b.bias.makeNumpyArray()])\n",
"\n",
" model.layers[3].set_weights([swift_model.conv2a.filter.makeNumpyArray(), swift_model.conv2a.bias.makeNumpyArray()])\n",
" model.layers[4].set_weights([swift_model.conv2b.filter.makeNumpyArray(), swift_model.conv2b.bias.makeNumpyArray()])\n",
"\n",
" model.layers[7].set_weights([swift_model.dense1.weight.makeNumpyArray(), swift_model.dense1.bias.makeNumpyArray()])\n",
" model.layers[8].set_weights([swift_model.dense2.weight.makeNumpyArray(), swift_model.dense2.bias.makeNumpyArray()])\n",
" model.layers[9].set_weights([swift_model.output.weight.makeNumpyArray(), swift_model.output.bias.makeNumpyArray()])\n",
"\n",
" model.save(\"cifar.h5\")\n",
"}"
],
"execution_count": 0,
"outputs": []
},
{
"cell_type": "code",
"metadata": {
"id": "3uiYJKn1xfc6",
"colab_type": "code",
"colab": {
"base_uri": "https://localhost:8080/",
"height": 887
},
"outputId": "1f84ec40-7aef-410e-a048-c705d08ac2ff"
},
"source": [
"save_swift_kerash5()"
],
"execution_count": 14,
"outputs": [
{
"output_type": "stream",
"text": [
"Using TensorFlow backend.\n",
"WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:66: The name tf.get_default_graph is deprecated. Please use tf.compat.v1.get_default_graph instead.\n",
"\n",
"WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:541: The name tf.placeholder is deprecated. Please use tf.compat.v1.placeholder instead.\n",
"\n",
"WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:4432: The name tf.random_uniform is deprecated. Please use tf.random.uniform instead.\n",
"\n",
"WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:4267: The name tf.nn.max_pool is deprecated. Please use tf.nn.max_pool2d instead.\n",
"\n",
"Model: \"sequential_1\"\n",
"_________________________________________________________________\n",
"Layer (type) Output Shape Param # \n",
"=================================================================\n",
"conv2d_1 (Conv2D) (None, 32, 32, 32) 896 \n",
"_________________________________________________________________\n",
"conv2d_2 (Conv2D) (None, 30, 30, 32) 9248 \n",
"_________________________________________________________________\n",
"max_pooling2d_1 (MaxPooling2 (None, 15, 15, 32) 0 \n",
"_________________________________________________________________\n",
"conv2d_3 (Conv2D) (None, 15, 15, 64) 18496 \n",
"_________________________________________________________________\n",
"conv2d_4 (Conv2D) (None, 13, 13, 64) 36928 \n",
"_________________________________________________________________\n",
"max_pooling2d_2 (MaxPooling2 (None, 6, 6, 64) 0 \n",
"_________________________________________________________________\n",
"flatten_1 (Flatten) (None, 2304) 0 \n",
"_________________________________________________________________\n",
"dense_1 (Dense) (None, 512) 1180160 \n",
"_________________________________________________________________\n",
"dense_2 (Dense) (None, 512) 262656 \n",
"_________________________________________________________________\n",
"dense_3 (Dense) (None, 10) 5130 \n",
"=================================================================\n",
"Total params: 1,513,514\n",
"Trainable params: 1,513,514\n",
"Non-trainable params: 0\n",
"_________________________________________________________________\n",
"None\n",
"WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:190: The name tf.get_default_session is deprecated. Please use tf.compat.v1.get_default_session instead.\n",
"\n",
"WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:197: The name tf.ConfigProto is deprecated. Please use tf.compat.v1.ConfigProto instead.\n",
"\n",
"WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:203: The name tf.Session is deprecated. Please use tf.compat.v1.Session instead.\n",
"\n",
"WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:207: The name tf.global_variables is deprecated. Please use tf.compat.v1.global_variables instead.\n",
"\n",
"WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:216: The name tf.is_variable_initialized is deprecated. Please use tf.compat.v1.is_variable_initialized instead.\n",
"\n",
"WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:223: The name tf.variables_initializer is deprecated. Please use tf.compat.v1.variables_initializer instead.\n",
"\n"
],
"name": "stdout"
}
]
}
]
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment