断言错误：某些对象具有未恢复的属性

Question

我正在通过关注官方 TensorFlow 网站此处 来训练关于文本预测的基本 LSTM 。我设法在 GTX 1050ti 上训练我的模型多达 40 个时期，并将 checkPoint 文件保存在一个单独的文件夹中。但是，当我现在尝试恢复模型时，出现了这个长错误：-

StreamExecutor device (0): GeForce GTX 1050 Ti, Compute Capability 6.1
WARNING:tensorflow:Entity <function standard_gru at 0x7f9e121324d0> could not be transformed and will be executed as-is. Please report this to the AutoGraph team. When filing the bug, set the verbosity to 10 (on Linux, `export AUTOGRAPH_VERBOSITY=10`) and attach the full output. Cause: converting <function standard_gru at 0x7f9e121324d0>: AttributeError: module 'gast' has no attribute 'Num'
WARNING:tensorflow:Entity <function cudnn_gru at 0x7f9e120c1d40> could not be transformed and will be executed as-is. Please report this to the AutoGraph team. When filing the bug, set the verbosity to 10 (on Linux, `export AUTOGRAPH_VERBOSITY=10`) and attach the full output. Cause: converting <function cudnn_gru at 0x7f9e120c1d40>: AttributeError: module 'gast' has no attribute 'Num'
WARNING:tensorflow:Entity <function standard_gru at 0x7f9e121324d0> could not be transformed and will be executed as-is. Please report this to the AutoGraph team. When filing the bug, set the verbosity to 10 (on Linux, `export AUTOGRAPH_VERBOSITY=10`) and attach the full output. Cause: converting <function standard_gru at 0x7f9e121324d0>: AttributeError: module 'gast' has no attribute 'Num'
WARNING:tensorflow:Entity <function cudnn_gru at 0x7f9e120c1d40> could not be transformed and will be executed as-is. Please report this to the AutoGraph team. When filing the bug, set the verbosity to 10 (on Linux, `export AUTOGRAPH_VERBOSITY=10`) and attach the full output. Cause: converting <function cudnn_gru at 0x7f9e120c1d40>: AttributeError: module 'gast' has no attribute 'Num'
WARNING:tensorflow:From /home/awesome_ruler/.local/lib/python3.7/site-packages/tensorflow/python/training/tracking/util.py:1200: NameBasedSaverStatus.__init__ (from tensorflow.python.training.tracking.util) is deprecated and will be removed in a future version.
Instructions for updating:
Restoring a name-based tf.train.Saver checkpoint using the object-based restore API. This mode uses global names to match variables, and so is somewhat fragile. It also adds new restore ops to the graph each time it is called when graph building. Prefer re-encoding training checkpoints in the object-based format: run save() on the object-based saver (the same one this message is coming from) and use that checkpoint in the future.
Traceback (most recent call last):
  File "main.py", line 95, in <module>
    model.load_weights(checkpoint_dir)
  File "/home/awesome_ruler/.local/lib/python3.7/site-packages/tensorflow/python/keras/engine/training.py", line 162, in load_weights
    return super(Model, self).load_weights(filepath, by_name)
  File "/home/awesome_ruler/.local/lib/python3.7/site-packages/tensorflow/python/keras/engine/network.py", line 1398, in load_weights
    status.assert_nontrivial_match()
  File "/home/awesome_ruler/.local/lib/python3.7/site-packages/tensorflow/python/training/tracking/util.py", line 917, in assert_nontrivial_match
    return self.assert_consumed()
  File "/home/awesome_ruler/.local/lib/python3.7/site-packages/tensorflow/python/training/tracking/util.py", line 894, in assert_consumed
    (unused_attributes,))
AssertionError: Some objects had attributes which were not restored: {<tf.Variable 'embedding_1/embeddings:0' shape=(65, 256) dtype=float32, numpy=
array([[-0.00044268, -0.02351714, -0.01139065, ..., -0.00327835,
         0.00074228, -0.00383734],
       [-0.02313181,  0.04697707, -0.02350216, ...,  0.040385  ,
         0.03087702,  0.02765551],
       [ 0.0410727 ,  0.00130001,  0.0051438 , ...,  0.02899202,
         0.04258115, -0.03773504],
       ...,
       [-0.03134514,  0.01370119,  0.00993627, ..., -0.02257681,
         0.02617678,  0.03761976],
       [-0.02954974,  0.02407967,  0.02768463, ..., -0.0056519 ,
        -0.01507735,  0.04617763],
       [-0.04113789, -0.03544737,  0.01056757, ...,  0.01236727,
        -0.01791535, -0.01635399]], dtype=float32)>: ['embedding_1/embeddings'], <tf.Variable 'dense_1/kernel:0' shape=(1024, 65) dtype=float32, numpy=
array([[-6.7811467e-02, -2.5536597e-02,  5.1763237e-02, ...,
        -6.9665730e-02,  3.9457709e-02, -5.3290475e-02],
       [ 1.5835620e-02, -3.0763537e-02, -7.4058644e-02, ...,
         3.8087368e-05, -9.1508478e-03,  5.5485427e-02],
       [ 3.8143486e-02,  8.8131428e-04, -2.3478847e-02, ...,
        -1.5135627e-02, -5.2146181e-02,  7.1185097e-02],
       ...,
       [-6.6591002e-02,  4.7627889e-02,  5.7474524e-02, ...,
         4.1528463e-02,  4.6467118e-02, -3.0670539e-02],
       [-5.0804108e-02,  5.4505378e-02, -1.5776977e-03, ...,
         2.1875933e-02, -2.9637258e-02,  2.0201296e-02],
       [-4.7325939e-02, -8.0013275e-03, -3.6348965e-02, ...,
        -7.0560835e-02, -4.9752403e-02,  1.0509960e-02]], dtype=float32)>: ['dense_1/kernel'], <tf.Variable 'dense_1/bias:0' shape=(65,) dtype=float32, numpy=
array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],
      dtype=float32)>: ['dense_1/bias'], <tf.Variable 'gru_1/kernel:0' shape=(256, 3072) dtype=float32, numpy=
array([[ 0.00432818,  0.03131782,  0.00038544, ..., -0.00559966,
         0.03458985, -0.03219106],
       [-0.00865119,  0.01648769, -0.00768028, ...,  0.01366192,
        -0.03043955, -0.01382086],
       [-0.01379537,  0.00547716, -0.00385967, ..., -0.00027269,
        -0.01285852,  0.0377048 ],
       ...,
       [-0.01940641,  0.01454895,  0.03349226, ..., -0.04234404,
        -0.02699661,  0.0376601 ],
       [ 0.00186675, -0.00547577, -0.02205843, ..., -0.01287581,
        -0.02314153,  0.04158166],
       [ 0.00954719, -0.02883693, -0.03259185, ..., -0.02587803,
         0.02906795, -0.00559821]], dtype=float32)>: ['gru_1/kernel'], <tf.Variable 'gru_1/recurrent_kernel:0' shape=(1024, 3072) dtype=float32, numpy=
array([[ 9.11542401e-03,  1.50135346e-02,  2.96630897e-02, ...,
         2.25223936e-02,  2.31253020e-02, -2.96920985e-02],
       [-2.21075956e-02, -8.46013427e-06, -2.16848943e-02, ...,
        -1.26914177e-02, -3.49153839e-02, -3.01396102e-02],
       [-3.59148793e-02,  9.98445973e-03,  2.60963626e-02, ...,
         3.15430500e-02,  1.28889643e-02,  3.37569825e-02],
       ...,
       [ 3.39106433e-02,  6.54980540e-03, -1.27352085e-02, ...,
        -4.14674729e-03,  3.53236459e-02, -1.36333425e-02],
       [-3.50691415e-02, -1.76392253e-02,  1.67468414e-02, ...,
        -2.06982102e-02, -1.06042419e-02,  2.26641595e-02],
       [-1.14825107e-02, -3.46554294e-02, -1.83847174e-03, ...,
         2.25809850e-02,  2.45791934e-02, -2.70933360e-02]], dtype=float32)>: ['gru_1/recurrent_kernel'], <tf.Variable 'gru_1/bias:0' shape=(2, 3072) dtype=float32, numpy=
array([[0., 0., 0., ..., 0., 0., 0.],
       [0., 0., 0., ..., 0., 0., 0.]], dtype=float32)>: ['gru_1/bias']}

我正在尝试加载文件ckpt_40.index，如您所见，这是最新的检查点。但是我不能。我正在使用此代码加载我的模型 ==>

checkpoint_dir = 'CheckPoints/ckpt_40.index'

model = build_model(vocab_size, embedding_dim, rnn_units, batch_size=1)

model.load_weights(checkpoint_dir)

model.summary()

并且正在使用generate_text网站上的功能来预测某些事情。

我认为Stack overflow here 上发布了一个类似的问题，但没有得到答复。我正在使用 Tf[GPU] 2.0-beta1，这是 GPU 的最新 tf 版本...

Answer 1

我犯了一个非常愚蠢的错误，这个错误太小了，我怀疑有人会发现它。在这一行：-

checkpoint_dir = 'CheckPoints/ckpt_40.index'

尽管该文件被命名为具有“.index”前缀，但出于某种原因将该扩展名附加到变量/调用函数已导致它出于某种原因而恐慌（可能是一个错误）。更有用的是指出错误的扩展名的错误。

因此，对于遇到此问题的其他任何人，只需将检查点目录更改为此 ===>

checkpoint_dir = 'CheckPoints/ckpt_40  # .index has been removed'