This guide trains a neural network model to classify images of clothing, like sneakers and shirts, saves the trained model, and then serves it with TensorFlow Serving. The focus is on TensorFlow Serving, rather than the modeling and training in TensorFlow, so for a complete example which focuses on the modeling and training see the Basic Classification example.
This guide uses tf.keras, a high-level API to build and train models in TensorFlow.
import sys
# Confirm that we're using Python 3
assert sys.version_info.major is 3, 'Oops, not running Python 3. Use Runtime > Change runtime type'
# TensorFlow and tf.keras
print("Installing dependencies for Colab environment")
!pip install -Uq grpcio==1.26.0
import tensorflow as tf
from tensorflow import keras
# Helper libraries
import numpy as np
import matplotlib.pyplot as plt
import os
import subprocess
print('TensorFlow version: {}'.format(tf.__version__))
Installing dependencies for Colab environment [K |████████████████████████████████| 2.4MB 4.6MB/s [?25hInstalling TensorFlow TensorFlow 2.x selected. TensorFlow version: 2.1.0-rc1
Create your model
Import the Fashion MNIST dataset
This guide uses the Fashion MNIST dataset which contains 70,000 grayscale images in 10 categories. The images show individual articles of clothing at low resolution (28 by 28 pixels), as seen here:
![]() |
Figure 1. Fashion-MNIST samples (by Zalando, MIT License). |
Fashion MNIST is intended as a drop-in replacement for the classic MNIST dataset—often used as the "Hello, World" of machine learning programs for computer vision. You can access the Fashion MNIST directly from TensorFlow, just import and load the data.
fashion_mnist = keras.datasets.fashion_mnist
(train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data()
# scale the values to 0.0 to 1.0
train_images = train_images / 255.0
test_images = test_images / 255.0
# reshape for feeding into the model
train_images = train_images.reshape(train_images.shape[0], 28, 28, 1)
test_images = test_images.reshape(test_images.shape[0], 28, 28, 1)
class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']
print('\ntrain_images.shape: {}, of {}'.format(train_images.shape, train_images.dtype))
print('test_images.shape: {}, of {}'.format(test_images.shape, test_images.dtype))
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-labels-idx1-ubyte.gz 32768/29515 [=================================] - 0s 0us/step Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-images-idx3-ubyte.gz 26427392/26421880 [==============================] - 0s 0us/step Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-labels-idx1-ubyte.gz 8192/5148 [===============================================] - 0s 0us/step Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-images-idx3-ubyte.gz 4423680/4422102 [==============================] - 0s 0us/step train_images.shape: (60000, 28, 28, 1), of float64 test_images.shape: (10000, 28, 28, 1), of float64
Train and evaluate your model
Let's use the simplest possible CNN, since we're not focused on the modeling part.
model = keras.Sequential([
keras.layers.Conv2D(input_shape=(28,28,1), filters=8, kernel_size=3,
strides=2, activation='relu', name='Conv1'),
keras.layers.Flatten(),
keras.layers.Dense(10, activation=tf.nn.softmax, name='Softmax')
])
model.summary()
testing = False
epochs = 5
model.compile(optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
model.fit(train_images, train_labels, epochs=epochs)
test_loss, test_acc = model.evaluate(test_images, test_labels)
print('\nTest accuracy: {}'.format(test_acc))
Model: "sequential" _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= Conv1 (Conv2D) (None, 13, 13, 8) 80 _________________________________________________________________ flatten (Flatten) (None, 1352) 0 _________________________________________________________________ Softmax (Dense) (None, 10) 13530 ================================================================= Total params: 13,610 Trainable params: 13,610 Non-trainable params: 0 _________________________________________________________________ Train on 60000 samples Epoch 1/5 60000/60000 [==============================] - 11s 185us/sample - loss: 0.5466 - accuracy: 0.8087 Epoch 2/5 60000/60000 [==============================] - 5s 79us/sample - loss: 0.4032 - accuracy: 0.8580 Epoch 3/5 60000/60000 [==============================] - 5s 76us/sample - loss: 0.3613 - accuracy: 0.8712 Epoch 4/5 60000/60000 [==============================] - 5s 75us/sample - loss: 0.3406 - accuracy: 0.8797 Epoch 5/5 60000/60000 [==============================] - 4s 75us/sample - loss: 0.3247 - accuracy: 0.8848 10000/10000 [==============================] - 1s 73us/sample - loss: 0.3510 - accuracy: 0.8747 Test accuracy: 0.8747000098228455
Save your model
To load our trained model into TensorFlow Serving we first need to save it in SavedModel format. This will create a protobuf file in a well-defined directory hierarchy, and will include a version number. TensorFlow Serving allows us to select which version of a model, or "servable" we want to use when we make inference requests. Each version will be exported to a different sub-directory under the given path.
# Fetch the Keras session and save the model
# The signature definition is defined by the input and output tensors,
# and stored with the default serving key
import tempfile
MODEL_DIR = tempfile.gettempdir()
version = 1
export_path = os.path.join(MODEL_DIR, str(version))
print('export_path = {}\n'.format(export_path))
tf.keras.models.save_model(
model,
export_path,
overwrite=True,
include_optimizer=True,
save_format=None,
signatures=None,
options=None
)
print('\nSaved model:')
!ls -l {export_path}
export_path = /tmp/1 Warning:tensorflow:From /tensorflow-2.1.0/python3.6/tensorflow_core/python/ops/resource_variable_ops.py:1786: calling BaseResourceVariable.__init__ (from tensorflow.python.ops.resource_variable_ops) with constraint is deprecated and will be removed in a future version. Instructions for updating: If using Keras pass *_constraint arguments to layers. INFO:tensorflow:Assets written to: /tmp/1/assets Saved model: total 84 drwxr-xr-x 2 root root 4096 Jan 7 23:15 assets -rw-r--r-- 1 root root 74086 Jan 7 23:15 saved_model.pb drwxr-xr-x 2 root root 4096 Jan 7 23:15 variables
Examine your saved model
We'll use the command line utility saved_model_cli
to look at the MetaGraphDefs (the models) and SignatureDefs (the methods you can call) in our SavedModel. See this discussion of the SavedModel CLI in the TensorFlow Guide.
saved_model_cli show --dir {export_path} --all
MetaGraphDef with tag-set: 'serve' contains the following SignatureDefs: signature_def['__saved_model_init_op']: The given SavedModel SignatureDef contains the following input(s): The given SavedModel SignatureDef contains the following output(s): outputs['__saved_model_init_op'] tensor_info: dtype: DT_INVALID shape: unknown_rank name: NoOp Method name is: signature_def['serving_default']: The given SavedModel SignatureDef contains the following input(s): inputs['Conv1_input'] tensor_info: dtype: DT_FLOAT shape: (-1, 28, 28, 1) name: serving_default_Conv1_input:0 The given SavedModel SignatureDef contains the following output(s): outputs['Softmax'] tensor_info: dtype: DT_FLOAT shape: (-1, 10) name: StatefulPartitionedCall:0 Method name is: tensorflow/serving/predict WARNING:tensorflow:From /tensorflow-2.1.0/python3.6/tensorflow_core/python/ops/resource_variable_ops.py:1786: calling BaseResourceVariable.__init__ (from tensorflow.python.ops.resource_variable_ops) with constraint is deprecated and will be removed in a future version. Instructions for updating: If using Keras pass *_constraint arguments to layers. Defined Functions: Function Name: '__call__' Option #1 Callable with: Argument #1 Conv1_input: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='Conv1_input') Argument #2 DType: bool Value: True Argument #3 DType: NoneType Value: None Option #2 Callable with: Argument #1 inputs: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='inputs') Argument #2 DType: bool Value: True Argument #3 DType: NoneType Value: None Option #3 Callable with: Argument #1 inputs: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='inputs') Argument #2 DType: bool Value: False Argument #3 DType: NoneType Value: None Option #4 Callable with: Argument #1 Conv1_input: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='Conv1_input') Argument #2 DType: bool Value: False Argument #3 DType: NoneType Value: None Function Name: '_default_save_signature' Option #1 Callable with: Argument #1 Conv1_input: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='Conv1_input') Function Name: 'call_and_return_all_conditional_losses' Option #1 Callable with: Argument #1 inputs: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='inputs') Argument #2 DType: bool Value: True Argument #3 DType: NoneType Value: None Option #2 Callable with: Argument #1 inputs: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='inputs') Argument #2 DType: bool Value: False Argument #3 DType: NoneType Value: None Option #3 Callable with: Argument #1 Conv1_input: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='Conv1_input') Argument #2 DType: bool Value: False Argument #3 DType: NoneType Value: None Option #4 Callable with: Argument #1 Conv1_input: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='Conv1_input') Argument #2 DType: bool Value: True Argument #3 DType: NoneType Value: None
That tells us a lot about our model! In this case we just trained our model, so we already know the inputs and outputs, but if we didn't this would be important information. It doesn't tell us everything, like the fact that this is grayscale image data for example, but it's a great start.
Serve your model with TensorFlow Serving
Add TensorFlow Serving distribution URI as a package source:
We're preparing to install TensorFlow Serving using Aptitude since this Colab runs in a Debian environment. We'll add the tensorflow-model-server
package to the list of packages that Aptitude knows about. Note that we're running as root.
# This is the same as you would do from your command line, but without the [arch=amd64], and no sudo
# You would instead do:
# echo "deb [arch=amd64] http://storage.googleapis.com/tensorflow-serving-apt stable tensorflow-model-server tensorflow-model-server-universal" | sudo tee /etc/apt/sources.list.d/tensorflow-serving.list && \
# curl https://storage.googleapis.com/tensorflow-serving-apt/tensorflow-serving.release.pub.gpg | sudo apt-key add -
!echo "deb http://storage.googleapis.com/tensorflow-serving-apt stable tensorflow-model-server tensorflow-model-server-universal" | tee /etc/apt/sources.list.d/tensorflow-serving.list && \
curl https://storage.googleapis.com/tensorflow-serving-apt/tensorflow-serving.release.pub.gpg | apt-key add -
!apt update
deb http://storage.googleapis.com/tensorflow-serving-apt stable tensorflow-model-server tensorflow-model-server-universal % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 2943 100 2943 0 0 11496 0 --:--:-- --:--:-- --:--:-- 11496 OK Get:1 http://storage.googleapis.com/tensorflow-serving-apt stable InRelease [3,012 B] Get:2 https://cloud.r-project.org/bin/linux/ubuntu bionic-cran35/ InRelease [3,626 B] Ign:3 https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64 InRelease Ign:4 https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64 InRelease Hit:5 https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64 Release Get:6 https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64 Release [564 B] Get:7 https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64 Release.gpg [833 B] Hit:8 http://ppa.launchpad.net/graphics-drivers/ppa/ubuntu bionic InRelease Hit:9 http://archive.ubuntu.com/ubuntu bionic InRelease Get:10 http://security.ubuntu.com/ubuntu bionic-security InRelease [88.7 kB] Get:11 http://storage.googleapis.com/tensorflow-serving-apt stable/tensorflow-model-server amd64 Packages [354 B] Get:12 https://cloud.r-project.org/bin/linux/ubuntu bionic-cran35/ Packages [81.6 kB] Get:13 http://archive.ubuntu.com/ubuntu bionic-updates InRelease [88.7 kB] Get:14 http://storage.googleapis.com/tensorflow-serving-apt stable/tensorflow-model-server-universal amd64 Packages [364 B] Get:15 http://ppa.launchpad.net/marutter/c2d4u3.5/ubuntu bionic InRelease [15.4 kB] Get:17 https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64 Packages [30.4 kB] Get:18 http://archive.ubuntu.com/ubuntu bionic-backports InRelease [74.6 kB] Get:19 http://ppa.launchpad.net/marutter/c2d4u3.5/ubuntu bionic/main Sources [1,749 kB] Get:20 http://security.ubuntu.com/ubuntu bionic-security/universe amd64 Packages [796 kB] Get:21 http://archive.ubuntu.com/ubuntu bionic-updates/main amd64 Packages [1,073 kB] Get:22 http://security.ubuntu.com/ubuntu bionic-security/main amd64 Packages [776 kB] Get:23 http://security.ubuntu.com/ubuntu bionic-security/restricted amd64 Packages [21.3 kB] Get:24 http://archive.ubuntu.com/ubuntu bionic-updates/multiverse amd64 Packages [10.8 kB] Get:25 http://archive.ubuntu.com/ubuntu bionic-updates/universe amd64 Packages [1,324 kB] Get:26 http://archive.ubuntu.com/ubuntu bionic-updates/restricted amd64 Packages [35.5 kB] Get:27 http://ppa.launchpad.net/marutter/c2d4u3.5/ubuntu bionic/main amd64 Packages [844 kB] Fetched 7,019 kB in 4s (1,913 kB/s) Reading package lists... Done Building dependency tree Reading state information... Done 21 packages can be upgraded. Run 'apt list --upgradable' to see them.
Install TensorFlow Serving
This is all you need - one command line!
apt-get install tensorflow-model-server
Reading package lists... Done Building dependency tree Reading state information... Done The following package was automatically installed and is no longer required: libnvidia-common-430 Use 'apt autoremove' to remove it. The following NEW packages will be installed: tensorflow-model-server 0 upgraded, 1 newly installed, 0 to remove and 21 not upgraded. Need to get 140 MB of archives. After this operation, 0 B of additional disk space will be used. Get:1 http://storage.googleapis.com/tensorflow-serving-apt stable/tensorflow-model-server amd64 tensorflow-model-server all 2.0.0 [140 MB] Fetched 140 MB in 2s (78.8 MB/s) Selecting previously unselected package tensorflow-model-server. (Reading database ... 145674 files and directories currently installed.) Preparing to unpack .../tensorflow-model-server_2.0.0_all.deb ... Unpacking tensorflow-model-server (2.0.0) ... Setting up tensorflow-model-server (2.0.0) ...
Start running TensorFlow Serving
This is where we start running TensorFlow Serving and load our model. After it loads we can start making inference requests using REST. There are some important parameters:
rest_api_port
: The port that you'll use for REST requests.model_name
: You'll use this in the URL of REST requests. It can be anything.model_base_path
: This is the path to the directory where you've saved your model.
os.environ["MODEL_DIR"] = MODEL_DIR
nohup tensorflow_model_server \
--rest_api_port=8501 \
--model_name=fashion_model \
--model_base_path="${MODEL_DIR}" >server.log 2>&1
Starting job # 0 in a separate thread.
tail server.log
[warn] getaddrinfo: address family for nodename not supported [evhttp_server.cc : 238] NET_LOG: Entering the event loop ...
Make a request to your model in TensorFlow Serving
First, let's take a look at a random example from our test data.
def show(idx, title):
plt.figure()
plt.imshow(test_images[idx].reshape(28,28))
plt.axis('off')
plt.title('\n\n{}'.format(title), fontdict={'size': 16})
import random
rando = random.randint(0,len(test_images)-1)
show(rando, 'An Example Image: {}'.format(class_names[test_labels[rando]]))
Ok, that looks interesting. How hard is that for you to recognize? Now let's create the JSON object for a batch of three inference requests, and see how well our model recognizes things:
import json
data = json.dumps({"signature_name": "serving_default", "instances": test_images[0:3].tolist()})
print('Data: {} ... {}'.format(data[:50], data[len(data)-52:]))
Data: {"signature_name": "serving_default", "instances": ... [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0]]]]}
Make REST requests
Newest version of the servable
We'll send a predict request as a POST to our server's REST endpoint, and pass it three examples. We'll ask our server to give us the latest version of our servable by not specifying a particular version.
!pip install -q requests
import requests
headers = {"content-type": "application/json"}
json_response = requests.post('http://localhost:8501/v1/models/fashion_model:predict', data=data, headers=headers)
predictions = json.loads(json_response.text)['predictions']
show(0, 'The model thought this was a {} (class {}), and it was actually a {} (class {})'.format(
class_names[np.argmax(predictions[0])], np.argmax(predictions[0]), class_names[test_labels[0]], test_labels[0]))
A particular version of the servable
Now let's specify a particular version of our servable. Since we only have one, let's select version 1. We'll also look at all three results.
headers = {"content-type": "application/json"}
json_response = requests.post('http://localhost:8501/v1/models/fashion_model/versions/1:predict', data=data, headers=headers)
predictions = json.loads(json_response.text)['predictions']
for i in range(0,3):
show(i, 'The model thought this was a {} (class {}), and it was actually a {} (class {})'.format(
class_names[np.argmax(predictions[i])], np.argmax(predictions[i]), class_names[test_labels[i]], test_labels[i]))