O Dia da Comunidade de ML é dia 9 de novembro! Junte-nos para atualização de TensorFlow, JAX, e mais Saiba mais

Salve e carregue um modelo usando uma estratégia de distribuição

Ver no TensorFlow.org Executar no Google Colab Ver fonte no GitHub Baixar caderno

Visão geral

É comum salvar e carregar um modelo durante o treinamento. Existem dois conjuntos de APIs para salvar e carregar um modelo keras: uma API de alto nível e uma API de baixo nível. Este tutorial demonstra como você pode usar as APIs SavedModel ao usar tf.distribute.Strategy . Para saber mais sobre SavedModel e serialização, em geral, leia o guia de modelo salvo , eo guia de modelo de serialização Keras . Vamos começar com um exemplo simples:

Dependências de importação:

import tensorflow_datasets as tfds

import tensorflow as tf

Prepare os dados e modelar usando tf.distribute.Strategy :

mirrored_strategy = tf.distribute.MirroredStrategy()

def get_data():
  datasets, ds_info = tfds.load(name='mnist', with_info=True, as_supervised=True)
  mnist_train, mnist_test = datasets['train'], datasets['test']

  BUFFER_SIZE = 10000

  BATCH_SIZE_PER_REPLICA = 64
  BATCH_SIZE = BATCH_SIZE_PER_REPLICA * mirrored_strategy.num_replicas_in_sync

  def scale(image, label):
    image = tf.cast(image, tf.float32)
    image /= 255

    return image, label

  train_dataset = mnist_train.map(scale).cache().shuffle(BUFFER_SIZE).batch(BATCH_SIZE)
  eval_dataset = mnist_test.map(scale).batch(BATCH_SIZE)

  return train_dataset, eval_dataset

def get_model():
  with mirrored_strategy.scope():
    model = tf.keras.Sequential([
        tf.keras.layers.Conv2D(32, 3, activation='relu', input_shape=(28, 28, 1)),
        tf.keras.layers.MaxPooling2D(),
        tf.keras.layers.Flatten(),
        tf.keras.layers.Dense(64, activation='relu'),
        tf.keras.layers.Dense(10)
    ])

    model.compile(loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
                  optimizer=tf.keras.optimizers.Adam(),
                  metrics=[tf.metrics.SparseCategoricalAccuracy()])
    return model
INFO:tensorflow:Using MirroredStrategy with devices ('/job:localhost/replica:0/task:0/device:GPU:0',)

Treine o modelo:

model = get_model()
train_dataset, eval_dataset = get_data()
model.fit(train_dataset, epochs=2)
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
2021-10-26 01:26:36.109959: W tensorflow/core/grappler/optimizers/data/auto_shard.cc:461] The `assert_cardinality` transformation is currently not handled by the auto-shard rewrite and will be removed.
Epoch 1/2
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
938/938 [==============================] - 13s 3ms/step - loss: 0.2015 - sparse_categorical_accuracy: 0.9410
Epoch 2/2
938/938 [==============================] - 3s 3ms/step - loss: 0.0663 - sparse_categorical_accuracy: 0.9807
<keras.callbacks.History at 0x7fa92037bc90>

Salve e carregue o modelo

Agora que você tem um modelo simples para trabalhar, vamos dar uma olhada nas APIs de salvamento / carregamento. Existem dois conjuntos de APIs disponíveis:

As APIs Keras

Aqui está um exemplo de como salvar e carregar um modelo com as APIs Keras:

keras_model_path = "/tmp/keras_save"
model.save(keras_model_path)
2021-10-26 01:26:52.520058: 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/keras_save/assets
INFO:tensorflow:Assets written to: /tmp/keras_save/assets

Restaurar o modelo sem tf.distribute.Strategy :

restored_keras_model = tf.keras.models.load_model(keras_model_path)
restored_keras_model.fit(train_dataset, epochs=2)
Epoch 1/2
938/938 [==============================] - 2s 2ms/step - loss: 0.0491 - sparse_categorical_accuracy: 0.9851
Epoch 2/2
938/938 [==============================] - 2s 2ms/step - loss: 0.0356 - sparse_categorical_accuracy: 0.9890
<keras.callbacks.History at 0x7fa8dc6d6690>

Depois de restaurar o modelo, você pode continuar a treinar nele, mesmo sem a necessidade de chamadas compile() novamente, uma vez que já é compilado antes de salvar. O modelo é salvo no padrão da TensorFlow SavedModel formato proto. Para mais informações, consulte o guia para saved_model formato .

Agora, para carregar o modelo e treiná-lo usando um tf.distribute.Strategy :

another_strategy = tf.distribute.OneDeviceStrategy("/cpu:0")
with another_strategy.scope():
  restored_keras_model_ds = tf.keras.models.load_model(keras_model_path)
  restored_keras_model_ds.fit(train_dataset, epochs=2)
2021-10-26 01:26:57.965185: W tensorflow/core/grappler/optimizers/data/auto_shard.cc:461] The `assert_cardinality` transformation is currently not handled by the auto-shard rewrite and will be removed.
2021-10-26 01:26:58.004038: W tensorflow/core/framework/dataset.cc:679] Input of GeneratorDatasetOp::Dataset will not be optimized because the dataset does not implement the AsGraphDefInternal() method needed to apply optimizations.
Epoch 1/2
938/938 [==============================] - 9s 9ms/step - loss: 0.0493 - sparse_categorical_accuracy: 0.9846
Epoch 2/2
938/938 [==============================] - 8s 9ms/step - loss: 0.0345 - sparse_categorical_accuracy: 0.9898

Como você pode ver, as obras de carga como esperado com tf.distribute.Strategy . A estratégia usada aqui não precisa ser a mesma estratégia usada antes de salvar.

Os tf.saved_model APIs

Agora, vamos dar uma olhada nas APIs de nível inferior. Salvar o modelo é semelhante à API keras:

model = get_model()  # get a fresh model
saved_model_path = "/tmp/tf_save"
tf.saved_model.save(model, saved_model_path)
INFO:tensorflow:Assets written to: /tmp/tf_save/assets
INFO:tensorflow:Assets written to: /tmp/tf_save/assets

Carregando pode ser feito com tf.saved_model.load() . No entanto, como é uma API de nível inferior (e, portanto, tem uma gama mais ampla de casos de uso), ela não retorna um modelo Keras. Em vez disso, ele retorna um objeto que contém funções que podem ser usadas para fazer inferência. Por exemplo:

DEFAULT_FUNCTION_KEY = "serving_default"
loaded = tf.saved_model.load(saved_model_path)
inference_func = loaded.signatures[DEFAULT_FUNCTION_KEY]

O objeto carregado pode conter várias funções, cada uma associada a uma chave. O "serving_default" é a chave padrão para a função de inferência com um modelo Keras salvo. Para fazer uma inferência com esta função:

predict_dataset = eval_dataset.map(lambda image, label: image)
for batch in predict_dataset.take(1):
  print(inference_func(batch))
{'dense_3': <tf.Tensor: shape=(64, 10), dtype=float32, numpy=
array([[-0.11688858, -0.05038287, -0.2585946 ,  0.04893515,  0.27253783,
         0.1022947 , -0.06840641, -0.33529347, -0.07071295,  0.06517357],
       [ 0.10904782, -0.23611397, -0.16135186,  0.10045648,  0.26082516,
        -0.02260189,  0.0424989 , -0.09468129,  0.05540806,  0.10558474],
       [-0.0491788 , -0.04070761, -0.23004392,  0.17719601,  0.20461476,
        -0.05333536, -0.02240408, -0.21509385, -0.05161493,  0.12337525],
       [ 0.00487803, -0.05770147, -0.23551641,  0.05988425,  0.15881103,
        -0.05608599, -0.04135028, -0.3390705 , -0.07579579, -0.08983649],
       [-0.04663972, -0.13439807, -0.19048163,  0.13628994,  0.05608338,
        -0.06012772, -0.03063064, -0.32014394, -0.16421723,  0.08930477],
       [ 0.02328245,  0.05272574, -0.34110764,  0.12926938,  0.33982378,
         0.12486804, -0.04870659, -0.45755434, -0.05433567,  0.14137071],
       [ 0.06421333, -0.20211999, -0.14309192,  0.00360708,  0.23210834,
         0.00101324, -0.01692696, -0.15713055,  0.00623474, -0.02222142],
       [ 0.08059486,  0.0456247 , -0.15926833,  0.05546484,  0.09179395,
         0.06136999, -0.07209414, -0.2553306 , -0.04975087,  0.06797761],
       [ 0.05864911, -0.10561213, -0.23619679,  0.11069187,  0.13890924,
         0.04969782, -0.05587994, -0.26131746, -0.0363602 ,  0.02788973],
       [ 0.0296779 ,  0.06670297, -0.12159262,  0.06834705,  0.19103828,
         0.14597046,  0.00285575, -0.19362533, -0.06905006,  0.097047  ],
       [ 0.05100356, -0.03875454, -0.31727186,  0.01787528,  0.20725562,
        -0.01677462, -0.00129463, -0.17944467,  0.05812614,  0.04979762],
       [-0.03301986, -0.10880841, -0.21802825,  0.0578297 ,  0.41345048,
         0.10376748,  0.03452782, -0.27389282, -0.06923576,  0.14353925],
       [-0.02203556, -0.08816119, -0.15965816,  0.07572726,  0.018046  ,
        -0.10299203,  0.01126328, -0.21401492, -0.17861444,  0.05669294],
       [-0.0245089 , -0.03849422, -0.2968499 ,  0.23396973,  0.22189453,
         0.00512835, -0.00468208, -0.29407114, -0.14926936, -0.02818882],
       [-0.02376807, -0.05931192, -0.31774518,  0.15711312,  0.31248903,
        -0.04320139, -0.08301807, -0.4610513 , -0.10252888, -0.03784092],
       [-0.03953424, -0.08268867, -0.3604463 ,  0.14048189,  0.33057037,
         0.01373108, -0.12093162, -0.38173944,  0.01771745, -0.07451382],
       [-0.05658644,  0.0519563 , -0.20794927,  0.10203589,  0.2135886 ,
         0.14241108, -0.04007911, -0.26177728, -0.08082938,  0.00216334],
       [-0.06207625, -0.01838757, -0.21708131,  0.10756977,  0.25599915,
         0.03101911,  0.05593228, -0.25550944, -0.11642678,  0.09014311],
       [ 0.05197014,  0.03873106, -0.1469059 ,  0.08044868,  0.12293777,
        -0.00388163,  0.00324975, -0.08145286, -0.12639561, -0.03596487],
       [-0.10676757, -0.05767517, -0.20481907,  0.14739943,  0.17379019,
        -0.08260865, -0.09114882, -0.38688654, -0.1448748 ,  0.03397277],
       [-0.03770879,  0.04663504, -0.30894646,  0.05933709,  0.09536786,
         0.1006383 ,  0.00984312, -0.3204393 , -0.01170056, -0.03391666],
       [ 0.0231554 ,  0.12106506, -0.255493  ,  0.04387057,  0.12491666,
         0.03297757, -0.03934925, -0.17047551,  0.00603533,  0.02295396],
       [-0.0137163 , -0.08226999, -0.3219023 ,  0.1111999 ,  0.15005693,
        -0.10358538, -0.04351711, -0.24015021, -0.08079101,  0.01281704],
       [ 0.08698535, -0.17155564, -0.19832517, -0.0417797 ,  0.24460419,
        -0.00698967,  0.08663791, -0.20004068,  0.02847612,  0.12739052],
       [ 0.0248102 , -0.07629397, -0.10130948,  0.00225735,  0.14270194,
         0.01750292,  0.03144339, -0.1429488 , -0.02819812,  0.24307509],
       [-0.06557162, -0.06485987, -0.36512223,  0.18774748,  0.25643086,
         0.0340823 , -0.01398754, -0.19010906, -0.07261477,  0.05117159],
       [ 0.04187369,  0.0132397 , -0.16233045,  0.10300563,  0.06598518,
         0.05728842, -0.02450454, -0.22889516, -0.03530695,  0.08300389],
       [ 0.15359762, -0.06493542, -0.22839671,  0.05915322,  0.26544052,
         0.15312935, -0.05132065, -0.34682024, -0.0181414 ,  0.08866596],
       [-0.06705338, -0.05590982, -0.21037713,  0.05252159,  0.22411834,
         0.06072947, -0.01180699, -0.31283215, -0.06644081, -0.02687445],
       [-0.01673558, -0.04322004, -0.22221681,  0.11640421,  0.27585298,
        -0.00789917, -0.03705985, -0.12847525, -0.14132528, -0.01258589],
       [ 0.05363014, -0.11879475, -0.08204994,  0.16474688,  0.09248446,
        -0.09719495, -0.07723137, -0.23136492, -0.05618468,  0.10164495],
       [-0.02539362, -0.14454898, -0.32296312,  0.2053542 ,  0.18563472,
        -0.0445538 , -0.13633929, -0.12712947, -0.06732591,  0.05459897],
       [-0.02403368, -0.09293792, -0.22012895,  0.09356467,  0.3415923 ,
        -0.09844425, -0.04539915, -0.28688133, -0.14435257,  0.05483858],
       [ 0.03492264,  0.04167182, -0.08564096,  0.01466741,  0.14968738,
         0.01946784, -0.04962645, -0.09357765, -0.03180797,  0.03431095],
       [ 0.04553585, -0.06386177, -0.159064  ,  0.09195592,  0.20032357,
         0.05248308,  0.05274323, -0.09328806, -0.02849531,  0.10636853],
       [-0.08788846, -0.05706687, -0.27519208,  0.12941426,  0.1730625 ,
         0.00562337,  0.03862702, -0.3364083 ,  0.01087172,  0.03377784],
       [-0.08110045, -0.06666276, -0.34764278,  0.25369477,  0.26242447,
         0.03672977,  0.07488421, -0.11382174,  0.03446682,  0.20799701],
       [-0.02429771, -0.0130821 , -0.28549588,  0.09956603,  0.19093114,
         0.09172641, -0.01084431, -0.26826024, -0.09550276, -0.09001306],
       [-0.0405377 ,  0.02302578, -0.16092977,  0.12650998,  0.10584372,
         0.0598565 ,  0.0370068 , -0.13375495, -0.05769489,  0.04597083],
       [-0.08379065, -0.12666067, -0.23740488,  0.08539408,  0.19100066,
        -0.19001569, -0.03504099, -0.2954648 , -0.00778607, -0.10035929],
       [-0.06841633, -0.02935523, -0.27325606,  0.07019119,  0.13153824,
         0.03444952, -0.07040955, -0.16061744, -0.05776489, -0.02386798],
       [ 0.02282005, -0.03760834, -0.17803052,  0.09008945,  0.15709753,
        -0.02815568, -0.01385967, -0.2636196 , -0.06011615, -0.04417434],
       [ 0.05103182, -0.0073192 , -0.2492007 ,  0.09097242,  0.2589297 ,
         0.03582668, -0.05287637, -0.1023304 , -0.10472505, -0.02360192],
       [-0.04446318, -0.00104156, -0.22680247,  0.0975772 ,  0.25874364,
         0.07281871,  0.14879908, -0.21233654, -0.11104408,  0.1596871 ],
       [-0.16542982, -0.02617702, -0.2530758 ,  0.09354755,  0.19404459,
         0.0228528 , -0.03458656, -0.3274249 , -0.08492248,  0.07104953],
       [-0.04432368, -0.01551367, -0.30958706,  0.08279304,  0.15877493,
         0.14097705,  0.0056034 , -0.2121813 , -0.10417398,  0.13372038],
       [ 0.00872401,  0.02290398, -0.18306321,  0.11926699,  0.0969364 ,
        -0.04007095,  0.01660407, -0.28434896, -0.15929542,  0.01083255],
       [ 0.07433248, -0.14991361, -0.2220522 ,  0.00625274,  0.39078072,
         0.03646233,  0.10941336, -0.20384778, -0.02929106,  0.03544597],
       [-0.00069001, -0.0680518 , -0.11302898,  0.11793397,  0.11893341,
        -0.05947986, -0.02543334, -0.24527295, -0.09240474, -0.00762735],
       [ 0.01683525,  0.03738175, -0.18935157,  0.07978748,  0.23876491,
         0.15589894, -0.00638897, -0.25770593, -0.11232982, -0.0446422 ],
       [-0.01690136, -0.19515185, -0.2338915 , -0.00964288,  0.17318843,
        -0.02175554,  0.07482283, -0.19234088, -0.0229656 ,  0.11406161],
       [-0.00661898,  0.00870193, -0.11167589,  0.15103012,  0.06432639,
        -0.12180559,  0.04999296, -0.2667799 , -0.17659347, -0.04285187],
       [-0.01717829,  0.02375691, -0.14970137,  0.1191919 ,  0.10172842,
        -0.07352136,  0.02696884, -0.11598936, -0.1331213 , -0.00928868],
       [-0.05850236,  0.03356444, -0.24372646,  0.14034908,  0.22228894,
         0.04799255, -0.01023421, -0.23915118, -0.07773915,  0.01665494],
       [-0.04828071, -0.00198432, -0.21945187,  0.14940068,  0.26243302,
         0.04732714, -0.03919668, -0.3767312 , -0.04807761,  0.04837478],
       [ 0.08090632,  0.02816604, -0.31061617,  0.04813545,  0.17886776,
         0.10947818,  0.0324835 , -0.22861008, -0.01619428, -0.00963937],
       [ 0.01237603, -0.07633115, -0.20681188,  0.08626392,  0.16251579,
         0.05692254,  0.00641025, -0.027444  ,  0.05301347,  0.00296039],
       [-0.03114549, -0.03946134, -0.20575103,  0.158873  ,  0.19106835,
        -0.00628418, -0.06812906, -0.29752672, -0.12863883,  0.00519179],
       [-0.02839492,  0.00197193, -0.38123846,  0.12928526,  0.4360217 ,
         0.06745887, -0.01924693, -0.3610945 ,  0.02880143,  0.00938179],
       [-0.10277586,  0.01430387, -0.24793717, -0.02120358,  0.20257095,
         0.10856566,  0.08017994, -0.21743834,  0.02736677,  0.01270235],
       [ 0.00209297, -0.04658009, -0.10872659,  0.00873713,  0.12002683,
        -0.01763269,  0.00062436, -0.07574805,  0.00423002,  0.09696378],
       [-0.0030484 ,  0.00373926, -0.20884912,  0.03331832,  0.37477142,
         0.14008212,  0.031428  , -0.40348598, -0.02555457,  0.05203115],
       [ 0.06917666, -0.07515088, -0.15344585,  0.08451273,  0.16555418,
        -0.00663652, -0.03506049, -0.19360425, -0.01485892, -0.1411201 ],
       [ 0.08957651, -0.0336723 , -0.16066113,  0.09386282,  0.21388392,
        -0.01653587, -0.02893457, -0.04395334, -0.03723653,  0.07710503]],
      dtype=float32)>}
2021-10-26 01:27:16.715879: W tensorflow/core/kernels/data/cache_dataset_ops.cc:768] The calling iterator did not fully read the dataset being cached. In order to avoid unexpected truncation of the dataset, the partially cached contents of the dataset  will be discarded. This can happen if you have an input pipeline similar to `dataset.cache().take(k).repeat()`. You should use `dataset.take(k).cache().repeat()` instead.

Você também pode carregar e fazer inferências de maneira distribuída:

another_strategy = tf.distribute.MirroredStrategy()
with another_strategy.scope():
  loaded = tf.saved_model.load(saved_model_path)
  inference_func = loaded.signatures[DEFAULT_FUNCTION_KEY]

  dist_predict_dataset = another_strategy.experimental_distribute_dataset(
      predict_dataset)

  # Calling the function in a distributed manner
  for batch in dist_predict_dataset:
    another_strategy.run(inference_func,args=(batch,))
INFO:tensorflow:Using MirroredStrategy with devices ('/job:localhost/replica:0/task:0/device:GPU:0',)
INFO:tensorflow:Using MirroredStrategy with devices ('/job:localhost/replica:0/task:0/device:GPU:0',)
2021-10-26 01:27:16.888897: W tensorflow/core/grappler/optimizers/data/auto_shard.cc:461] The `assert_cardinality` transformation is currently not handled by the auto-shard rewrite and will be removed.
WARNING:tensorflow:Using MirroredStrategy eagerly has significant overhead currently. We will be working on improving this in the future, but for now please wrap `call_for_each_replica` or `experimental_run` or `run` inside a tf.function to get the best performance.
WARNING:tensorflow:Using MirroredStrategy eagerly has significant overhead currently. We will be working on improving this in the future, but for now please wrap `call_for_each_replica` or `experimental_run` or `run` inside a tf.function to get the best performance.
WARNING:tensorflow:Using MirroredStrategy eagerly has significant overhead currently. We will be working on improving this in the future, but for now please wrap `call_for_each_replica` or `experimental_run` or `run` inside a tf.function to get the best performance.
WARNING:tensorflow:Using MirroredStrategy eagerly has significant overhead currently. We will be working on improving this in the future, but for now please wrap `call_for_each_replica` or `experimental_run` or `run` inside a tf.function to get the best performance.
WARNING:tensorflow:Using MirroredStrategy eagerly has significant overhead currently. We will be working on improving this in the future, but for now please wrap `call_for_each_replica` or `experimental_run` or `run` inside a tf.function to get the best performance.
WARNING:tensorflow:Using MirroredStrategy eagerly has significant overhead currently. We will be working on improving this in the future, but for now please wrap `call_for_each_replica` or `experimental_run` or `run` inside a tf.function to get the best performance.
WARNING:tensorflow:Using MirroredStrategy eagerly has significant overhead currently. We will be working on improving this in the future, but for now please wrap `call_for_each_replica` or `experimental_run` or `run` inside a tf.function to get the best performance.
WARNING:tensorflow:Using MirroredStrategy eagerly has significant overhead currently. We will be working on improving this in the future, but for now please wrap `call_for_each_replica` or `experimental_run` or `run` inside a tf.function to get the best performance.
WARNING:tensorflow:Using MirroredStrategy eagerly has significant overhead currently. We will be working on improving this in the future, but for now please wrap `call_for_each_replica` or `experimental_run` or `run` inside a tf.function to get the best performance.
WARNING:tensorflow:Using MirroredStrategy eagerly has significant overhead currently. We will be working on improving this in the future, but for now please wrap `call_for_each_replica` or `experimental_run` or `run` inside a tf.function to get the best performance.

Chamar a função restaurada é apenas um passe para frente no modelo salvo (previsão). E se você quiser continuar treinando a função carregada? Ou incorporar a função carregada em um modelo maior? Uma prática comum é envolver esse objeto carregado em uma camada Keras para fazer isso. Felizmente, TF Hub tem hub.KerasLayer para este fim, mostrado aqui:

import tensorflow_hub as hub

def build_model(loaded):
  x = tf.keras.layers.Input(shape=(28, 28, 1), name='input_x')
  # Wrap what's loaded to a KerasLayer
  keras_layer = hub.KerasLayer(loaded, trainable=True)(x)
  model = tf.keras.Model(x, keras_layer)
  return model

another_strategy = tf.distribute.MirroredStrategy()
with another_strategy.scope():
  loaded = tf.saved_model.load(saved_model_path)
  model = build_model(loaded)

  model.compile(loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
                optimizer=tf.keras.optimizers.Adam(),
                metrics=[tf.metrics.SparseCategoricalAccuracy()])
  model.fit(train_dataset, epochs=2)
INFO:tensorflow:Using MirroredStrategy with devices ('/job:localhost/replica:0/task:0/device:GPU:0',)
INFO:tensorflow:Using MirroredStrategy with devices ('/job:localhost/replica:0/task:0/device:GPU:0',)
2021-10-26 01:27:18.637232: W tensorflow/core/grappler/optimizers/data/auto_shard.cc:461] The `assert_cardinality` transformation is currently not handled by the auto-shard rewrite and will be removed.
Epoch 1/2
938/938 [==============================] - 5s 3ms/step - loss: 0.2057 - sparse_categorical_accuracy: 0.9392
Epoch 2/2
938/938 [==============================] - 3s 3ms/step - loss: 0.0688 - sparse_categorical_accuracy: 0.9802

Como você pode ver, hub.KerasLayer envolve a parte de trás resultado carregado a partir tf.saved_model.load() em uma camada Keras que pode ser usado para construir um outro modelo. Isso é muito útil para a aprendizagem por transferência.

Qual API devo usar?

Para salvar, se você estiver trabalhando com um modelo keras, é quase sempre recomendado o uso do Keras model.save() API. Se o que você está salvando não é um modelo Keras, a API de nível inferior é sua única opção.

Para carregar, qual API você usa depende do que você deseja obter da API de carregamento. Se você não pode (ou não quer) obter um modelo Keras, em seguida, usar tf.saved_model.load() . Caso contrário, o uso tf.keras.models.load_model() . Observe que você pode obter um modelo Keras de volta somente se você salvou um modelo Keras.

É possível misturar e combinar as APIs. Você pode salvar um modelo Keras com model.save , e carregar um modelo não-Keras com o baixo nível API, tf.saved_model.load .

model = get_model()

# Saving the model using Keras's save() API
model.save(keras_model_path) 

another_strategy = tf.distribute.MirroredStrategy()
# Loading the model using lower level API
with another_strategy.scope():
  loaded = tf.saved_model.load(keras_model_path)
INFO:tensorflow:Assets written to: /tmp/keras_save/assets
INFO:tensorflow:Assets written to: /tmp/keras_save/assets
INFO:tensorflow:Using MirroredStrategy with devices ('/job:localhost/replica:0/task:0/device:GPU:0',)
INFO:tensorflow:Using MirroredStrategy with devices ('/job:localhost/replica:0/task:0/device:GPU:0',)

Salvando / carregando do dispositivo local

Quando salvar e carregar a partir de um dispositivo io locais durante a execução remotamente, por exemplo usando um TPU nuvem, a opção experimental_io_device deve ser usado para definir o dispositivo io para localhost.

model = get_model()

# Saving the model to a path on localhost.
saved_model_path = "/tmp/tf_save"
save_options = tf.saved_model.SaveOptions(experimental_io_device='/job:localhost')
model.save(saved_model_path, options=save_options)

# Loading the model from a path on localhost.
another_strategy = tf.distribute.MirroredStrategy()
with another_strategy.scope():
  load_options = tf.saved_model.LoadOptions(experimental_io_device='/job:localhost')
  loaded = tf.keras.models.load_model(saved_model_path, options=load_options)
INFO:tensorflow:Assets written to: /tmp/tf_save/assets
INFO:tensorflow:Assets written to: /tmp/tf_save/assets
INFO:tensorflow:Using MirroredStrategy with devices ('/job:localhost/replica:0/task:0/device:GPU:0',)
INFO:tensorflow:Using MirroredStrategy with devices ('/job:localhost/replica:0/task:0/device:GPU:0',)

Ressalvas

Um caso especial é quando você tem um modelo Keras que não possui entradas bem definidas. Por exemplo, um modelo sequencial pode ser criado sem quaisquer formas de entrada ( Sequential([Dense(3), ...] ). Modelos subclasse também não têm entradas bem definidas após a inicialização. Neste caso, você deve ficar com a APIs de nível inferior para salvar e carregar, caso contrário, você receberá um erro.

Para verificar se o seu modelo tem entradas bem definidos, apenas verificar se model.inputs é None . Se não for None , você é tudo de bom. Formas de entrada são definidas automaticamente quando o modelo é usado em .fit , .evaluate , .predict , ou ao chamar o modelo ( model(inputs) ).

Aqui está um exemplo:

class SubclassedModel(tf.keras.Model):

  output_name = 'output_layer'

  def __init__(self):
    super(SubclassedModel, self).__init__()
    self._dense_layer = tf.keras.layers.Dense(
        5, dtype=tf.dtypes.float32, name=self.output_name)

  def call(self, inputs):
    return self._dense_layer(inputs)

my_model = SubclassedModel()
# my_model.save(keras_model_path)  # ERROR! 
tf.saved_model.save(my_model, saved_model_path)
WARNING:tensorflow:Skipping full serialization of Keras layer <__main__.SubclassedModel object at 0x7fa4f68ee5d0>, because it is not built.
WARNING:tensorflow:Skipping full serialization of Keras layer <__main__.SubclassedModel object at 0x7fa4f68ee5d0>, because it is not built.
WARNING:tensorflow:Skipping full serialization of Keras layer <keras.layers.core.Dense object at 0x7fa4f68ee490>, because it is not built.
WARNING:tensorflow:Skipping full serialization of Keras layer <keras.layers.core.Dense object at 0x7fa4f68ee490>, because it is not built.
INFO:tensorflow:Assets written to: /tmp/tf_save/assets
INFO:tensorflow:Assets written to: /tmp/tf_save/assets