从tensorflow hub加载wiki40b嵌入时出错

我正在尝试使用此模块（）加载带有

KerasLayer

，但不确定产生此错误的原因

import tensorflow as tf
import tensorflow_hub as hub
import tensorflow_text

hub_url = "https://tfhub.dev/google/wiki40b-lm-nl/1"
embed = hub.KerasLayer(hub_url, input_shape=[], 
                   dtype=tf.string)

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-37-4e8ab0d5082c> in <module>()
      5 hub_url = "https://tfhub.dev/google/wiki40b-lm-nl/1"
      6 embed = hub.KerasLayer(hub_url, input_shape=[], 
----> 7                        dtype=tf.string)

1 frames
/usr/local/lib/python3.7/dist-packages/tensorflow_hub/keras_layer.py in _get_callable(self)
    300     if self._signature not in self._func.signatures:
    301       raise ValueError("Unknown signature %s in %s (available signatures: %s)."
--> 302                        % (self._signature, self._handle, self._func.signatures))
    303     f = self._func.signatures[self._signature]
    304     if not callable(f):

ValueError: Unknown signature default in https://tfhub.dev/google/wiki40b-lm-nl/1 (available signatures: _SignatureMap({'neg_log_likelihood': <ConcreteFunction pruned(text) at 0x7F3044A93210>, 'tokenization': <ConcreteFunction pruned(text) at 0x7F3040B7D190>, 'token_neg_log_likelihood': <ConcreteFunction pruned(token) at 0x7F3040D14810>, 'word_embeddings': <ConcreteFunction pruned(text) at 0x7F303D3FF2D0>, 'activations': <ConcreteFunction pruned(text) at 0x7F303D3FFF50>, 'prediction': <ConcreteFunction pruned(mem_4, mem_5, mem_6, mem_7, mem_8, mem_9, mem_10, mem_11, input_tokens, mem_0, mem_1, mem_2, mem_3) at 0x7F303C189090>, 'detokenization': <ConcreteFunction pruned(token_ids) at 0x7F3039860790>, 'token_word_embeddings': <ConcreteFunction pruned(token) at 0x7F3038FC2110>, 'token_activations': <ConcreteFunction pruned(token) at 0x7F303BAF9150>})).

---

我的问题是，如何使用

str

作为输入的嵌入，正如他们在模块页面（输入部分）中指出的那样？

将封装在

tf.constant

中的文本传递到

embed（）

并设置

output\u键应该可以工作：

import tensorflow as tf
import tensorflow_hub as hub
import tensorflow_text

embed = hub.KerasLayer("https://tfhub.dev/google/wiki40b-lm-nl/1",
                       signature="word_embeddings",
                       output_key="word_embeddings")
embed(tf.constant("\n_START_ARTICLE_\n1001 vrouwen uit de Nederlandse "
                  "geschiedenis\n_START_SECTION_\nSelectie van vrouwen"
                  "\n_START_PARAGRAPH_\nDe 'oudste' biografie in het boek "
                  "is gewijd aan de beschermheilige"))

（使用TF 2.4.1和tensorflow_hub 0.11.0进行测试）
谢谢。我不是这种嵌入的专家，但在玩它的时候，我注意到每个输入文本都会产生一个三维输出，在第二维度上有不同的形状，这取决于标记的数量。很抱歉提出另一个问题，但是将嵌入平均为每个输入文本嵌入一行embed_dim是否不合适？
import tensorflow as tf
import tensorflow_hub as hub
import tensorflow_text

embed = hub.KerasLayer("https://tfhub.dev/google/wiki40b-lm-nl/1",
                       signature="word_embeddings",
                       output_key="word_embeddings")
embed(tf.constant("\n_START_ARTICLE_\n1001 vrouwen uit de Nederlandse "
                  "geschiedenis\n_START_SECTION_\nSelectie van vrouwen"
                  "\n_START_PARAGRAPH_\nDe 'oudste' biografie in het boek "
                  "is gewijd aan de beschermheilige"))