在 keras 模型中使用 `sentence-transformers`

Question

我想在更大的 Keras 模型中使用来自句子转换器的模型。

这是完整的示例：

import tensorflow as tf
from transformers import AutoTokenizer, TFAutoModel


MODEL_PATH = 'sentence-transformers/all-MiniLM-L6-v2'
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
model = TFAutoModel.from_pretrained(MODEL_PATH, from_pt=True)


class SBert(tf.keras.layers.Layer):
    def __init__(self, tokenizer, model):
        super(SBert, self).__init__()
        
        self.tokenizer = tokenizer
        self.model = model
        
    def tf_encode(self, inputs):
        def encode(inputs):
            return self.tokenizer(
                inputs, padding=True, truncation=True, return_tensors='tf'
            )
        return tf.py_function(func=encode, inp=[inputs], Tout=[tf.int64])
    
    def mean_pooling(model_output, attention_mask):
        token_embeddings = model_output[0]
        input_mask_expanded = tf.cast(
            tf.broadcast_to(tf.expand_dims(attention_mask, -1), token_embeddings.shape),
            tf.float32
        )
        a = tf.math.reduce_sum(token_embeddings * input_mask_expanded, axis=1)
        b = tf.clip_by_value(tf.math.reduce_sum(input_mask_expanded, axis=1), 1e-9, tf.float32.max)
        embeddings = a / b
        embeddings, _ = tf.linalg.normalize(embeddings, 2, axis=1)
        return embeddings

    def call(self, inputs):
        encoded_input = self.tf_encode(inputs)
        model_output = self.model(encoded_input)
        embeddings = self.mean_pooling(model_output, encoded_input['attention_mask'])
        return embeddings
    
    
sbert = SBert(tokenizer, model)
sbert(['some text', 'more text'])

我可以在 TF / Keras 之外使用模型和标记生成器，没有任何问题，当我们尝试构建图形并且 TF 将符号张量传递给标记生成器时，似乎会发生问题，从而生成错误 -这就是为什么我尝试包装 tf.py_function 但没有成功......

错误：

---------------------------------------------------------------------------
InvalidArgumentError                      Traceback (most recent call last)
<ipython-input-20-a0c4a906e456> in <module>
     44 
     45 sbert = SBert(tokenizer, model)
---> 46 sbert(['some text', 'more text'])

~/.pyenv/versions/3.7.8/lib/python3.7/site-packages/keras/utils/traceback_utils.py in error_handler(*args, **kwargs)
     65     except Exception as e:  # pylint: disable=broad-except
     66       filtered_tb = _process_traceback_frames(e.__traceback__)
---> 67       raise e.with_traceback(filtered_tb) from None
     68     finally:
     69       del filtered_tb

<ipython-input-20-a0c4a906e456> in call(self, inputs)
     36     def call(self, inputs):
     37         tf.print(inputs, output_stream=sys.stdout)
---> 38         encoded_input = self.tf_encode(inputs)
     39         tf.print(encoded_input, output_stream=sys.stdout)
     40         model_output = self.model(encoded_input)

<ipython-input-20-a0c4a906e456> in tf_encode(self, inputs)
     20                 inputs, padding=True, truncation=True, return_tensors='tf'
     21             )
---> 22         return tf.py_function(func=encode, inp=[inputs], Tout=[tf.int64])
     23 
     24     def mean_pooling(model_output, attention_mask):

InvalidArgumentError: Exception encountered when calling layer "s_bert_6" (type SBert).

ValueError: text input must of type `str` (single example), `List[str]` (batch or single pretokenized example) or `List[List[str]]` (batch of pretokenized examples).
Traceback (most recent call last):

  File "/Users/dennisyurkevich/.pyenv/versions/3.7.8/lib/python3.7/site-packages/tensorflow/python/ops/script_ops.py", line 269, in __call__
    return func(device, token, args)

  File "/Users/dennisyurkevich/.pyenv/versions/3.7.8/lib/python3.7/site-packages/tensorflow/python/ops/script_ops.py", line 147, in __call__
    outputs = self._call(device, args)

  File "/Users/dennisyurkevich/.pyenv/versions/3.7.8/lib/python3.7/site-packages/tensorflow/python/ops/script_ops.py", line 154, in _call
    ret = self._func(*args)

  File "/Users/dennisyurkevich/.pyenv/versions/3.7.8/lib/python3.7/site-packages/tensorflow/python/autograph/impl/api.py", line 642, in wrapper
    return func(*args, **kwargs)

  File "<ipython-input-20-a0c4a906e456>", line 20, in encode
    inputs, padding=True, truncation=True, return_tensors='tf'

  File "/Users/dennisyurkevich/.pyenv/versions/3.7.8/lib/python3.7/site-packages/transformers/tokenization_utils_base.py", line 2378, in __call__
    "text input must of type `str` (single example), `List[str]` (batch or single pretokenized example) "

ValueError: text input must of type `str` (single example), `List[str]` (batch or single pretokenized example) or `List[List[str]]` (batch of pretokenized examples).

 [Op:EagerPyFunc]

Call arguments received:
  • inputs=["'some text'", "'more text'"]

Answer 1

tf.py_function 似乎不适用于字典输出，这就是为什么您可以尝试返回三个单独的张量。另外，我正在解码输入以删除每个字符串前面的 b ：

import tensorflow as tf
from transformers import AutoTokenizer, TFAutoModel


MODEL_PATH = 'sentence-transformers/all-MiniLM-L6-v2'
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
model = TFAutoModel.from_pretrained(MODEL_PATH, from_pt=True)


class SBert(tf.keras.layers.Layer):
    def __init__(self, tokenizer, model):
        super(SBert, self).__init__()
        
        self.tokenizer = tokenizer
        self.model = model
        
    def tf_encode(self, inputs):
        def encode(inputs):
            inputs = [x.decode("utf-8") for x in inputs.numpy()]
            outputs = self.tokenizer(inputs, padding=True, truncation=True, return_tensors='tf')
            return outputs['input_ids'], outputs['token_type_ids'], outputs['attention_mask']
        return tf.py_function(func=encode, inp=[inputs], Tout=[tf.int32, tf.int32, tf.int32])
    
    def mean_pooling(self, model_output, attention_mask):
        token_embeddings = model_output[0]
        input_mask_expanded = tf.cast(
            tf.broadcast_to(tf.expand_dims(attention_mask, -1), tf.shape(token_embeddings)),
            tf.float32
        )
        a = tf.math.reduce_sum(token_embeddings * input_mask_expanded, axis=1)
        b = tf.clip_by_value(tf.math.reduce_sum(input_mask_expanded, axis=1), 1e-9, tf.float32.max)
        embeddings = a / b
        embeddings, _ = tf.linalg.normalize(embeddings, 2, axis=1)
        return embeddings
    def call(self, inputs):
        input_ids, token_type_ids, attention_mask = self.tf_encode(inputs)
        model_output = self.model({'input_ids': input_ids, 'token_type_ids': token_type_ids, 'attention_mask': attention_mask})
        embeddings = self.mean_pooling(model_output, attention_mask)
        return embeddings
    
    
sbert = SBert(tokenizer, model)
sbert(['some text', 'more text'])

如果您想使用 Keras 模型，则必须执行以下操作：

import tensorflow as tf
from transformers import AutoTokenizer, TFAutoModel

MODEL_PATH = 'sentence-transformers/all-MiniLM-L6-v2'
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
model = TFAutoModel.from_pretrained(MODEL_PATH, from_pt=True)

class SBert(tf.keras.layers.Layer):
    def __init__(self, tokenizer, model):
        super(SBert, self).__init__()
        
        self.tokenizer = tokenizer
        self.model = model
        
    def tf_encode(self, inputs):
        def encode(inputs):
            inputs = [x[0].decode("utf-8") for x in inputs.numpy()]
            outputs = self.tokenizer(inputs, padding=True, truncation=True, return_tensors='tf')
            return outputs['input_ids'], outputs['token_type_ids'], outputs['attention_mask']
        return tf.py_function(func=encode, inp=[inputs], Tout=[tf.int32, tf.int32, tf.int32])
    
    def process(self, i, t, a):
      def __call(i, t, a):
        model_output = self.model({'input_ids': i.numpy(), 'token_type_ids': t.numpy(), 'attention_mask': a.numpy()})
        return model_output[0]
      return tf.py_function(func=__call, inp=[i, t, a], Tout=[tf.float32])

    def mean_pooling(self, model_output, attention_mask):
        token_embeddings = tf.squeeze(tf.stack(model_output), axis=0)
        input_mask_expanded = tf.cast(
            tf.broadcast_to(tf.expand_dims(attention_mask, -1), tf.shape(token_embeddings)),
            tf.float32
        )
        a = tf.math.reduce_sum(token_embeddings * input_mask_expanded, axis=1)
        b = tf.clip_by_value(tf.math.reduce_sum(input_mask_expanded, axis=1), 1e-9, tf.float32.max)
        embeddings = a / b
        embeddings, _ = tf.linalg.normalize(embeddings, 2, axis=1)
        return embeddings
    def call(self, inputs):
        input_ids, token_type_ids, attention_mask = self.tf_encode(inputs)
        model_output = self.process(input_ids, token_type_ids, attention_mask)
        embeddings = self.mean_pooling(model_output, attention_mask)
        return embeddings
    
    
sbert = SBert(tokenizer, model)
inputs = tf.keras.layers.Input((1,), dtype=tf.string)
outputs = sbert(inputs)
model = tf.keras.Model(inputs, outputs)
model(tf.constant(['some text', 'more text']))

TensorShape([2, 384]).shape