¿Preguntas sobre TFX? ¡Únase a nosotros en Google I / O!

Entrena y entrega un modelo de TensorFlow con TensorFlow Serving

Esta guía entrena un modelo de red neuronal para clasificar imágenes de ropa, como zapatillas y camisetas , guarda el modelo entrenado y luego lo muestra con TensorFlow Serving . La atención se centra en TensorFlow Serving, en lugar del modelado y el entrenamiento en TensorFlow, por lo que para obtener un ejemplo completo que se centra en el modelado y el entrenamiento, consulte el ejemplo de Clasificación básica .

Esta guía usa tf.keras , una API de alto nivel para crear y entrenar modelos en 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__))

Crea tu modelo

Importar el conjunto de datos Fashion MNIST

Esta guía utiliza el conjunto de datos Fashion MNIST que contiene 70.000 imágenes en escala de grises en 10 categorías. Las imágenes muestran prendas de vestir individuales a baja resolución (28 por 28 píxeles), como se ve aquí:

Sprite de moda MNIST
Figura 1. Muestras de moda-MNIST (por Zalando, licencia del MIT).

Fashion MNIST está pensado como un reemplazo directo del clásico conjunto de datos MNIST , que a menudo se usa como el "Hola, mundo" de los programas de aprendizaje automático para visión por computadora. Puede acceder a Fashion MNIST directamente desde TensorFlow, solo importe y cargue los datos.

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

Entrena y evalúa tu modelo

Usemos la CNN más simple posible, ya que no estamos enfocados en la parte del modelado.

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, name='Dense')
])
model.summary()

testing = False
epochs = 5

model.compile(optimizer='adam', 
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=[keras.metrics.SparseCategoricalAccuracy()])
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         
_________________________________________________________________
Dense (Dense)                (None, 10)                13530     
=================================================================
Total params: 13,610
Trainable params: 13,610
Non-trainable params: 0
_________________________________________________________________
Epoch 1/5
1875/1875 [==============================] - 13s 2ms/step - loss: 0.7546 - sparse_categorical_accuracy: 0.7457
Epoch 2/5
1875/1875 [==============================] - 3s 2ms/step - loss: 0.4254 - sparse_categorical_accuracy: 0.8521
Epoch 3/5
1875/1875 [==============================] - 3s 2ms/step - loss: 0.3812 - sparse_categorical_accuracy: 0.8668
Epoch 4/5
1875/1875 [==============================] - 3s 2ms/step - loss: 0.3557 - sparse_categorical_accuracy: 0.8770
Epoch 5/5
1875/1875 [==============================] - 3s 2ms/step - loss: 0.3415 - sparse_categorical_accuracy: 0.8795
313/313 [==============================] - 1s 2ms/step - loss: 0.3699 - sparse_categorical_accuracy: 0.8694

Test accuracy: 0.8694000244140625

Guarda tu modelo

Para cargar nuestro modelo entrenado en TensorFlow Serving, primero debemos guardarlo en formato SavedModel . Esto creará un archivo protobuf en una jerarquía de directorios bien definida e incluirá un número de versión. TensorFlow Serving nos permite seleccionar qué versión de un modelo o "servible" queremos usar cuando hacemos solicitudes de inferencia. Cada versión se exportará a un subdirectorio diferente en la ruta indicada.

# 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

INFO:tensorflow:Assets written to: /tmp/1/assets

Saved model:
total 88
drwxr-xr-x 2 kbuilder kbuilder  4096 Mar  9 10:10 assets
-rw-rw-r-- 1 kbuilder kbuilder 78123 Mar  9 10:10 saved_model.pb
drwxr-xr-x 2 kbuilder kbuilder  4096 Mar  9 10:10 variables

Examina tu modelo guardado

Usaremos la utilidad de línea de comando saved_model_cli para ver MetaGraphDefs (los modelos) y SignatureDefs (los métodos que puede llamar) en nuestro SavedModel. Consulta esta discusión sobre la CLI del modelo guardado en la Guía de TensorFlow.

saved_model_cli show --dir {export_path} --all
2021-03-09 10:10:12.685464: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.11.0

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['Dense'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1, 10)
        name: StatefulPartitionedCall:0
  Method name is: tensorflow/serving/predict

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: False
        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
          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 #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

  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: False
        Argument #3
          DType: NoneType
          Value: None
    Option #2
      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 #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
          inputs: TensorSpec(shape=(None, 28, 28, 1), dtype=tf.float32, name='inputs')
        Argument #2
          DType: bool
          Value: True
        Argument #3
          DType: NoneType
          Value: None

¡Eso nos dice mucho sobre nuestro modelo! En este caso, acabamos de entrenar nuestro modelo, por lo que ya conocemos las entradas y salidas, pero si no lo supiéramos, esta sería información importante. No nos dice todo, como el hecho de que se trata de datos de imagen en escala de grises, por ejemplo, pero es un gran comienzo.

Sirva su modelo con TensorFlow Serving

Agrega el URI de distribución de TensorFlow Serving como fuente del paquete:

Nos estamos preparando para instalar TensorFlow Serving con Aptitude, ya que este Colab se ejecuta en un entorno Debian. tensorflow-model-server el tensorflow-model-server a la lista de paquetes que conoce Aptitude. Tenga en cuenta que estamos ejecutando como root.

import sys
# We need sudo prefix if not on a Google Colab.
if 'google.colab' not in sys.modules:
  SUDO_IF_NEEDED = 'sudo'
else:
  SUDO_IF_NEEDED = ''
# 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" | {SUDO_IF_NEEDED} tee /etc/apt/sources.list.d/tensorflow-serving.list && \
curl https://storage.googleapis.com/tensorflow-serving-apt/tensorflow-serving.release.pub.gpg | {SUDO_IF_NEEDED} apt-key add -
!{SUDO_IF_NEEDED} 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  15822      0 --:--:-- --:--:-- --:--:-- 15822
OK
Hit:1 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic InRelease
Hit:2 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic-updates InRelease
Hit:3 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic-backports InRelease
Hit:4 http://packages.cloud.google.com/apt google-cloud-logging-wheezy InRelease
Get:5 https://packages.cloud.google.com/apt eip-cloud-bionic InRelease [5419 B]
Hit:6 https://nvidia.github.io/libnvidia-container/stable/ubuntu18.04/amd64  InRelease
Hit:7 https://nvidia.github.io/nvidia-container-runtime/ubuntu18.04/amd64  InRelease
Get:8 https://nvidia.github.io/nvidia-docker/ubuntu18.04/amd64  InRelease [1129 B]
Get:9 http://storage.googleapis.com/tensorflow-serving-apt stable InRelease [3012 B]
Hit:10 http://archive.canonical.com/ubuntu bionic InRelease
Hit:11 http://security.ubuntu.com/ubuntu bionic-security InRelease
Get:12 http://storage.googleapis.com/tensorflow-serving-apt stable/tensorflow-model-server amd64 Packages [340 B]
Get:13 http://storage.googleapis.com/tensorflow-serving-apt stable/tensorflow-model-server-universal amd64 Packages [348 B]
Fetched 10.2 kB in 1s (7051 B/s)



114 packages can be upgraded. Run 'apt list --upgradable' to see them.

Instalar publicación de TensorFlow

Esto es todo lo que necesita: ¡una línea de comando!

{SUDO_IF_NEEDED} apt-get install tensorflow-model-server
The following packages were automatically installed and are no longer required:
  adwaita-icon-theme ca-certificates-java dconf-gsettings-backend
  dconf-service default-jre default-jre-headless dkms fonts-dejavu-extra
  freeglut3 freeglut3-dev g++-6 glib-networking glib-networking-common
  glib-networking-services gsettings-desktop-schemas gtk-update-icon-cache
  hicolor-icon-theme humanity-icon-theme java-common libaccinj64-9.1
  libasound2 libasound2-data libasyncns0 libatk-bridge2.0-0
  libatk-wrapper-java libatk-wrapper-java-jni libatk1.0-0 libatk1.0-data
  libatspi2.0-0 libavahi-client3 libavahi-common-data libavahi-common3
  libcairo-gobject2 libcolord2 libcroco3 libcudart9.1 libcufft9.1 libcufftw9.1
  libcups2 libcurand9.1 libcusolver9.1 libcusparse9.1 libdconf1 libdrm-amdgpu1
  libdrm-dev libdrm-intel1 libdrm-nouveau2 libdrm-radeon1 libegl-mesa0 libegl1
  libegl1-mesa libepoxy0 libflac8 libfontenc1 libgbm1 libgdk-pixbuf2.0-0
  libgdk-pixbuf2.0-common libgif7 libgl1 libgl1-mesa-dev libgl1-mesa-dri
  libglapi-mesa libgles1 libgles2 libglu1-mesa libglu1-mesa-dev
  libglvnd-core-dev libglvnd-dev libglvnd0 libglx-mesa0 libglx0 libgtk-3-0
  libgtk-3-common libgtk2.0-0 libgtk2.0-common libice-dev libjansson4
  libjson-glib-1.0-0 libjson-glib-1.0-common liblcms2-2 libllvm9 libnppc9.1
  libnppial9.1 libnppicc9.1 libnppicom9.1 libnppidei9.1 libnppif9.1
  libnppig9.1 libnppim9.1 libnppist9.1 libnppisu9.1 libnppitc9.1 libnpps9.1
  libnvrtc9.1 libnvtoolsext1 libnvvm3 libogg0 libopengl0 libpciaccess0
  libpcsclite1 libproxy1v5 libpthread-stubs0-dev libpulse0 librest-0.7-0
  librsvg2-2 librsvg2-common libsensors4 libsm-dev libsndfile1
  libsoup-gnome2.4-1 libsoup2.4-1 libstdc++-6-dev libthrust-dev libvdpau-dev
  libvdpau1 libvorbis0a libvorbisenc2 libwayland-client0 libwayland-cursor0
  libwayland-egl1 libwayland-server0 libx11-dev libx11-xcb-dev libx11-xcb1
  libxau-dev libxcb-dri2-0 libxcb-dri2-0-dev libxcb-dri3-0 libxcb-dri3-dev
  libxcb-glx0 libxcb-glx0-dev libxcb-present-dev libxcb-present0 libxcb-randr0
  libxcb-randr0-dev libxcb-render0-dev libxcb-shape0 libxcb-shape0-dev
  libxcb-sync-dev libxcb-sync1 libxcb-xfixes0 libxcb-xfixes0-dev libxcb1-dev
  libxcomposite1 libxcursor1 libxdamage-dev libxdamage1 libxdmcp-dev
  libxext-dev libxfixes-dev libxfixes3 libxfont2 libxft2 libxi-dev libxi6
  libxinerama1 libxkbcommon0 libxkbfile1 libxmu-dev libxmu-headers libxnvctrl0
  libxrandr2 libxshmfence-dev libxshmfence1 libxt-dev libxtst6 libxv1
  libxxf86dga1 libxxf86vm-dev libxxf86vm1 linux-gcp-5.3-headers-5.3.0-1030
  linux-gcp-headers-5.0.0-1026 linux-headers-5.3.0-1030-gcp
  linux-image-5.3.0-1030-gcp linux-modules-5.3.0-1030-gcp
  linux-modules-extra-5.3.0-1030-gcp mesa-common-dev ocl-icd-libopencl1
  ocl-icd-opencl-dev opencl-c-headers openjdk-11-jre openjdk-11-jre-headless
  openjdk-8-jre openjdk-8-jre-headless pkg-config policykit-1-gnome
  python3-xkit screen-resolution-extra ubuntu-mono x11-utils x11-xkb-utils
  x11proto-core-dev x11proto-damage-dev x11proto-dev x11proto-fixes-dev
  x11proto-input-dev x11proto-xext-dev x11proto-xf86vidmode-dev
  xorg-sgml-doctools xserver-common xserver-xorg-core-hwe-18.04 xtrans-dev
Use 'sudo apt autoremove' to remove them.
The following NEW packages will be installed:
  tensorflow-model-server
0 upgraded, 1 newly installed, 0 to remove and 114 not upgraded.
Need to get 223 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 2.4.1 [223 MB]
Fetched 223 MB in 6s (40.3 MB/s)
Selecting previously unselected package tensorflow-model-server.
(Reading database ... 242337 files and directories currently installed.)
Preparing to unpack .../tensorflow-model-server_2.4.1_all.deb ...
Unpacking tensorflow-model-server (2.4.1) ...
Setting up tensorflow-model-server (2.4.1) ...

Comience a ejecutar TensorFlow Serving

Aquí es donde comenzamos a ejecutar TensorFlow Serving y cargamos nuestro modelo. Después de que se cargue, podemos comenzar a realizar solicitudes de inferencia usando REST. Hay algunos parámetros importantes:

  • rest_api_port : el puerto que usará para las solicitudes REST.
  • model_name : lo usará en la URL de las solicitudes REST. Puede ser cualquier cosa.
  • model_base_path : esta es la ruta al directorio donde ha guardado su modelo.
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
tail server.log

Realice una solicitud a su modelo en TensorFlow Serving

Primero, echemos un vistazo a un ejemplo aleatorio de nuestros datos de prueba.

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]]))

png

Ok, eso parece interesante. ¿Qué tan difícil es para usted reconocerlo? Ahora creemos el objeto JSON para un lote de tres solicitudes de inferencia y veamos qué tan bien reconoce nuestro modelo las cosas:

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]]]]}

Realizar solicitudes de REST

Versión más nueva del servidor

Enviaremos una solicitud de predicción como POST al punto final REST de nuestro servidor y le pasaremos tres ejemplos. Le pediremos a nuestro servidor que nos dé la última versión de nuestro servidor sin especificar una versión en particular.

!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]))

Una versión particular del servidor

Ahora especifiquemos una versión particular de nuestro servidor. Como solo tenemos uno, seleccionemos la versión 1. También veremos los tres resultados.

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]))

png

png

png