BERT实践详解_bert_base_chinese

pip install transformers

from transformers import BertTokenizer

tokenizer = BertTokenizer.from_pretrained('bert-base-chinese')


max_length_test = 20
test_sentence = '曝梅西已通知巴萨他想离开'

# add special tokens
test_sentence_with_special_tokens = '[CLS]' + test_sentence + '[SEP]'
tokenized = tokenizer.tokenize(test_sentence_with_special_tokens)
print('tokenized', tokenized)

# convert tokens to ids in WordPiece
input_ids = tokenizer.convert_tokens_to_ids(tokenized)

# precalculation of pad length, so that we can reuse it later on
padding_length = max_length_test - len(input_ids)

# map tokens to WordPiece dictionary and add pad token for those text shorter than our max length
input_ids = input_ids + ([0] * padding_length)

# attention should focus just on sequence with non padded tokens
attention_mask = [1] * len(input_ids)

# do not focus attention on padded tokens
attention_mask = attention_mask + ([0] * padding_length)

# token types, needed for example for question answering, for our purpose we will just set 0 as we have just one sequence
token_type_ids = [0] * max_length_test
bert_input = {
    "token_ids": input_ids,
    "token_type_ids": token_type_ids,
    "attention_mask": attention_mask
}
print(bert_input)

import tensorflow as tf
import tensorflow_hub as hub
import tensorflow_text as text # A dependency of the preprocessing model

text_input = tf.keras.layers.Input(shape=(), dtype=tf.string)
preprocessor = hub.KerasLayer(
    "D:\\wyf\\workspace\\20211029_智能客服\\bert_zh_preprocess_3")
    #"https://tfhub.dev/tensorflow/bert_zh_preprocess/3")
encoder_inputs = preprocessor(text_input)
encoder = hub.KerasLayer(
    "D:\\wyf\\workspace\\20211029_智能客服\\bert_zh_L-12_H-768_A-12_4",
    #"https://tfhub.dev/tensorflow/bert_zh_L-12_H-768_A-12/4"
    trainable=True)
outputs = encoder(encoder_inputs)
pooled_output = outputs["pooled_output"]      # [batch_size, 768].
sequence_output = outputs["sequence_output"]  # [batch_size, seq_length, 768].

#embedding_model = tf.keras.Model(text_input, pooled_output)
embedding_model = tf.keras.Model(text_input, sequence_output)
sentences = tf.constant(["你好漂亮"])
#print(preprocessor(sentences))
print(embedding_model(sentences))