Addestra e servi un modello TensorFlow con TensorFlow Serving

Mantieni tutto organizzato con le raccolte Salva e classifica i contenuti in base alle tue preferenze.

Questa guida si allena un modello di rete neurale per classificare le immagini di abbigliamento, come scarpe da ginnastica e camicie , salva il modello addestrato, e poi lo serve con tensorflow servizio . Il focus è sulla tensorflow Serve, piuttosto che la modellazione e la formazione in tensorflow, così per un esempio completo che si concentra sulla modellazione e la formazione vedono l' esempio di classificazione di base .

Questa guida utilizza tf.keras , un'API di alto livello ai modelli costruire e formare in tensorflow.

import sys

# Confirm that we're using Python 3
assert sys.version_info.major == 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 il tuo modello

Importa il set di dati Fashion MNIST

Questa guida utilizza la moda MNIST set di dati che contiene 70.000 immagini in scala di grigi in 10 categorie. Le immagini mostrano singoli capi di abbigliamento a bassa risoluzione (28 x 28 pixel), come si vede qui:

Sprite di moda MNIST
Figura 1. Campioni di modo-MNIST (per Zalando, licenza MIT).

Moda MNIST è inteso come un rimpiazzo per il classico MNIST set di dati, spesso utilizzato come "Ciao, mondo" di programmi di apprendimento della macchina per la visione del computer. Puoi accedere al Fashion MNIST direttamente da TensorFlow, importa e carica i dati.

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

Allena e valuta il tuo modello

Usiamo la CNN più semplice possibile, dal momento che non ci concentriamo sulla parte di modellazione.

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))
2021-12-04 10:29:34.128871: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcusolver.so.10'; dlerror: libcusolver.so.10: cannot open shared object file: No such file or directory
2021-12-04 10:29:34.129907: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1757] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices...
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 [==============================] - 4s 2ms/step - loss: 0.7204 - sparse_categorical_accuracy: 0.7549
Epoch 2/5
1875/1875 [==============================] - 4s 2ms/step - loss: 0.3997 - sparse_categorical_accuracy: 0.8611
Epoch 3/5
1875/1875 [==============================] - 4s 2ms/step - loss: 0.3580 - sparse_categorical_accuracy: 0.8754
Epoch 4/5
1875/1875 [==============================] - 4s 2ms/step - loss: 0.3399 - sparse_categorical_accuracy: 0.8780
Epoch 5/5
1875/1875 [==============================] - 4s 2ms/step - loss: 0.3232 - sparse_categorical_accuracy: 0.8849
313/313 [==============================] - 0s 1ms/step - loss: 0.3586 - sparse_categorical_accuracy: 0.8738

Test accuracy: 0.8737999796867371

Salva il tuo modello

Per caricare il nostro modello addestrato in tensorflow Serve in primo luogo abbiamo bisogno di salvarlo in SavedModel formato. Questo creerà un file protobuf in una gerarchia di directory ben definita e includerà un numero di versione. Tensorflow dose ci permette di selezionare quale versione di un modello, o "servibile" vogliamo usare quando facciamo le richieste di inferenza. Ogni versione verrà esportata in una sottodirectory diversa nel percorso specificato.

# 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
2021-12-04 10:29:53.392905: W tensorflow/python/util/util.cc:348] Sets are not currently considered sequences, but this may change in the future, so consider avoiding using them.
INFO:tensorflow:Assets written to: /tmp/1/assets

Saved model:
total 88
drwxr-xr-x 2 kbuilder kbuilder  4096 Dec  4 10:29 assets
-rw-rw-r-- 1 kbuilder kbuilder 78055 Dec  4 10:29 saved_model.pb
drwxr-xr-x 2 kbuilder kbuilder  4096 Dec  4 10:29 variables

Esamina il tuo modello salvato

Useremo il comando utilità della riga saved_model_cli di guardare ai MetaGraphDefs (i modelli) e SignatureDefs (i metodi che si possono chiamare) nel nostro SavedModel. Vedere questa discussione del SavedModel CLI nella Guida tensorflow.

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['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

Questo ci dice molto sul nostro modello! In questo caso abbiamo appena addestrato il nostro modello, quindi conosciamo già gli input e gli output, ma se non lo facessimo questa sarebbe un'informazione importante. Non ci dice tutto, come ad esempio il fatto che si tratta di dati di immagini in scala di grigi, ma è un ottimo inizio.

Servi il tuo modello con TensorFlow Serving

Aggiungi l'URI di distribuzione TensorFlow Serving come origine del pacchetto:

Ci stiamo preparando per l'installazione tensorflow Serving utilizzando Aptitude poiché questo Colab viene eseguito in un ambiente di Debian. Aggiungeremo il tensorflow-model-server pacchetto per la lista dei pacchetti che Aptitude conosce. Nota che stiamo eseguendo come 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  15571      0 --:--:-- --:--:-- --:--:-- 15571
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 https://nvidia.github.io/libnvidia-container/stable/ubuntu18.04/amd64  InRelease
Get:5 https://nvidia.github.io/nvidia-container-runtime/ubuntu18.04/amd64  InRelease [1481 B]
Get:6 https://nvidia.github.io/nvidia-docker/ubuntu18.04/amd64  InRelease [1474 B]
Ign:7 http://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64  InRelease
Get:8 http://storage.googleapis.com/tensorflow-serving-apt stable InRelease [3012 B]
Hit:9 http://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64  Release
Get:10 http://security.ubuntu.com/ubuntu bionic-security InRelease [88.7 kB]
Get:11 https://packages.cloud.google.com/apt eip-cloud-bionic InRelease [5419 B]
Get:12 http://packages.cloud.google.com/apt google-cloud-logging-wheezy InRelease [5483 B]
Hit:13 http://archive.canonical.com/ubuntu bionic InRelease
Err:11 https://packages.cloud.google.com/apt eip-cloud-bionic InRelease
  The following signatures couldn't be verified because the public key is not available: NO_PUBKEY FEEA9169307EA071 NO_PUBKEY 8B57C5C2836F4BEB
Get:15 http://storage.googleapis.com/tensorflow-serving-apt stable/tensorflow-model-server amd64 Packages [339 B]
Err:12 http://packages.cloud.google.com/apt google-cloud-logging-wheezy InRelease
  The following signatures couldn't be verified because the public key is not available: NO_PUBKEY FEEA9169307EA071 NO_PUBKEY 8B57C5C2836F4BEB
Get:16 http://storage.googleapis.com/tensorflow-serving-apt stable/tensorflow-model-server-universal amd64 Packages [348 B]
Fetched 106 kB in 1s (103 kB/s)



119 packages can be upgraded. Run 'apt list --upgradable' to see them.
W: An error occurred during the signature verification. The repository is not updated and the previous index files will be used. GPG error: https://packages.cloud.google.com/apt eip-cloud-bionic InRelease: The following signatures couldn't be verified because the public key is not available: NO_PUBKEY FEEA9169307EA071 NO_PUBKEY 8B57C5C2836F4BEB
W: An error occurred during the signature verification. The repository is not updated and the previous index files will be used. GPG error: http://packages.cloud.google.com/apt google-cloud-logging-wheezy InRelease: The following signatures couldn't be verified because the public key is not available: NO_PUBKEY FEEA9169307EA071 NO_PUBKEY 8B57C5C2836F4BEB
W: Failed to fetch https://packages.cloud.google.com/apt/dists/eip-cloud-bionic/InRelease  The following signatures couldn't be verified because the public key is not available: NO_PUBKEY FEEA9169307EA071 NO_PUBKEY 8B57C5C2836F4BEB
W: Failed to fetch http://packages.cloud.google.com/apt/dists/google-cloud-logging-wheezy/InRelease  The following signatures couldn't be verified because the public key is not available: NO_PUBKEY FEEA9169307EA071 NO_PUBKEY 8B57C5C2836F4BEB
W: Some index files failed to download. They have been ignored, or old ones used instead.

Installa TensorFlow Serving

Questo è tutto ciò di cui hai bisogno: una riga di comando!

{SUDO_IF_NEEDED} apt-get install tensorflow-model-server
The following packages were automatically installed and are no longer required:
  linux-gcp-5.4-headers-5.4.0-1040 linux-gcp-5.4-headers-5.4.0-1043
  linux-gcp-5.4-headers-5.4.0-1044 linux-gcp-5.4-headers-5.4.0-1049
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 119 not upgraded.
Need to get 335 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 2.7.0 [335 MB]
Fetched 335 MB in 7s (45.2 MB/s)
Selecting previously unselected package tensorflow-model-server.
(Reading database ... 264341 files and directories currently installed.)
Preparing to unpack .../tensorflow-model-server_2.7.0_all.deb ...
Unpacking tensorflow-model-server (2.7.0) ...
Setting up tensorflow-model-server (2.7.0) ...

Inizia a eseguire TensorFlow Serving

È qui che iniziamo a eseguire TensorFlow Serving e carichiamo il nostro modello. Dopo il caricamento, possiamo iniziare a fare richieste di inferenza usando REST. Ci sono alcuni parametri importanti:

  • rest_api_port : la porta che verranno utilizzate per le richieste REST.
  • model_name : Userete questa nell'URL delle richieste REST. Può essere qualsiasi cosa.
  • model_base_path : Questo è il percorso della directory in cui avete salvato il vostro modello.
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

Fai una richiesta al tuo modello in TensorFlow Serving

Per prima cosa, diamo un'occhiata a un esempio casuale dai nostri dati di test.

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, sembra interessante. Quanto è difficile per te riconoscerlo? Ora creiamo l'oggetto JSON per un batch di tre richieste di inferenza e vediamo quanto bene il nostro modello riconosce le cose:

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]]]]}

Fai richieste REST

Versione più recente del server

Invieremo una richiesta di previsione come POST all'endpoint REST del nostro server e gli passeremo tre esempi. Chiederemo al nostro server di fornirci l'ultima versione del nostro server non specificando una versione particolare.

# docs_infra: no_execute
!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 versione particolare del servibile

Ora specifichiamo una versione particolare del nostro servable. Dato che ne abbiamo solo uno, selezioniamo la versione 1. Esamineremo anche tutti e tre i risultati.

# docs_infra: no_execute
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]))