TensorFlow - InvalidArgumentError：所有输入的形状必须匹配

Question

我目前正在使用 Jupyter Notebook 遵循有关 seq2seq NMT 模型的 TensorFlow 指南/教程 (https://www.tensorflow.org/text/tutorials/nmt_with_attention)。 运行以下代码后，

# Setup the loop variables.
next_token, done, state = decoder.get_initial_state(ex_context)
tokens = []

for n in range(10):
  # Run one step.
  next_token, done, state = decoder.get_next_token(
      ex_context, next_token, done, state, temperature=1.0)
  # Add the token to the output.
  tokens.append(next_token)

# Stack all the tokens together.
tokens = tf.concat(tokens, axis=-1) # (batch, t)

# Convert the tokens back to a a string
result = decoder.tokens_to_text(tokens)
result[:3].numpy()

我收到如下 InvalidArgument 错误：

---------------------------------------------------------------------------
InvalidArgumentError                      Traceback (most recent call last)
Cell In[31], line 2
      1 # Setup the loop variables.
----> 2 next_token, done, state = decoder.get_initial_state(ex_context)
      3 tokens = []
      5 for n in range(10):
      6   # Run one step.

Cell In[28], line 8
      6 embedded = self.embedding(start_tokens)
      7 print(embedded)
----> 8 return start_tokens, done, self.rnn.get_initial_state(embedded)[0]

File ~/Library/Python/3.11/lib/python/site-packages/keras/src/layers/rnn/rnn.py:309, in RNN.get_initial_state(self, batch_size)
    307 get_initial_state_fn = getattr(self.cell, "get_initial_state", None)
    308 if get_initial_state_fn:
--> 309     init_state = get_initial_state_fn(batch_size=batch_size)
    310 else:
    311     return [
    312         ops.zeros((batch_size, d), dtype=self.cell.compute_dtype)
    313         for d in self.state_size
    314     ]

File ~/Library/Python/3.11/lib/python/site-packages/keras/src/layers/rnn/gru.py:326, in GRUCell.get_initial_state(self, batch_size)
    324 def get_initial_state(self, batch_size=None):
    325     return [
--> 326         ops.zeros((batch_size, self.state_size), dtype=self.compute_dtype)
    327     ]

File ~/Library/Python/3.11/lib/python/site-packages/keras/src/ops/numpy.py:5968, in zeros(shape, dtype)
   5957 @keras_export(["keras.ops.zeros", "keras.ops.numpy.zeros"])
   5958 def zeros(shape, dtype=None):
   5959     """Return a new tensor of given shape and type, filled with zeros.
   5960 
   5961     Args:
   (...)
   5966         Tensor of zeros with the given shape and dtype.
   5967     """
-> 5968     return backend.numpy.zeros(shape, dtype=dtype)

File ~/Library/Python/3.11/lib/python/site-packages/keras/src/backend/tensorflow/numpy.py:619, in zeros(shape, dtype)
--> 617 return tf.zeros(shape, dtype=dtype)

File ~/Library/Python/3.11/lib/python/site-packages/tensorflow/python/util/traceback_utils.py:153, in filter_traceback.<locals>.error_handler(*args, **kwargs)
    151 except Exception as e:
    152   filtered_tb = _process_traceback_frames(e.__traceback__)
--> 153   raise e.with_traceback(filtered_tb) from None
    154 finally:
    155   del filtered_tb

File ~/Library/Python/3.11/lib/python/site-packages/tensorflow/python/framework/ops.py:5983, in raise_from_not_ok_status(e, name)
   5981 def raise_from_not_ok_status(e, name) -> NoReturn:
   5982   e.message += (" name: " + str(name if name is not None else ""))
-> 5983   raise core._status_to_exception(e) from None

InvalidArgumentError: {{function_node __wrapped__Pack_N_2_device_/job:localhost/replica:0/task:0/device:CPU:0}} Shapes of all inputs must match: values[0].shape = [64,1,256] != values[1].shape = [] [Op:Pack] name: 

有什么想法吗？我很确定我正在遵循这封信的指南。

Answer 1

原来我唯一的问题是版本控制 - 使用 python 3.10 和tensorflow 2.11 和tensorflow-text 2.11。使用 pyenv 创建虚拟环境似乎可以解决该问题。张量流和张量流文本包的轮子可以在 PyPi 上找到，因为 pip 没有这些版本。

我使用的其他软件包是：

• einops==0.6.0

• matplotlib==3.6.1

• numpy==1.23.3