def gen():
tokenizer = FullTokenizer(vocab_file=os.path.join(self.args.bert_model_dir, 'vocab.txt'))
# Windows does not support logger in MP environment, thus get a new logger
# inside the process for better compatibility
logger = set_logger(colored('WORKER-%d' % self.worker_id, 'yellow'), self.verbose)
poller = zmq.Poller()
for sock in socks:
poller.register(sock, zmq.POLLIN)
logger.info('ready and listening!')
while not self.exit_flag.is_set():
events = dict(poller.poll())
for sock_idx, sock in enumerate(socks):
if sock in events:
#接收来自客户端的消息
client_id, raw_msg = sock.recv_multipart()
msg = jsonapi.loads(raw_msg)
logger.info('new job\tsocket: %d\tsize: %d\tclient: %s' % (sock_idx, len(msg), client_id))
# check if msg is a list of list, if yes consider the input is already tokenized
# 对接收到的字符进行切词，并且转化为id格式
# logger.info('get msg:%s, type:%s' % (msg[0], type(msg[0])))
is_tokenized = all(isinstance(el, list) for el in msg)
tmp_f = list(convert_lst_to_features(msg, self.max_seq_len, tokenizer, logger,
is_tokenized, self.mask_cls_sep))
#print([f.input_ids for f in tmp_f])
yield {
'client_id': client_id,
'input_ids': [f.input_ids for f in tmp_f],
'input_mask': [f.input_mask for f in tmp_f],
'input_type_ids': [f.input_type_ids for f in tmp_f]
}
def localTestPb():
sentence = [
'重庆市南川中学副校长张竞说，寒假放假前高三年级还有10%左右的内容没复习完，学校开学受疫情影响延期不少，“老师们压力比较大，怕耽误复习进度。”'
]
is_tokenized = all(isinstance(el, list) for el in sentence)
tokenizer = FullTokenizer(vocab_file=os.path.join(
"D:\\LiuXianXian\\pycharm--code\\BertServiceBase_classify\\checkpoints",
'vocab.txt'))
logger = set_logger('WORKER-%d')
tmp_f = list(
convert_lst_to_features(sentence, 128, tokenizer, logger, is_tokenized,
False))
ids = tmp_f[0].input_ids
mask = tmp_f[0].input_mask
segment_ids = tmp_f[0].input_type_ids
export_dir = 'D:\\LiuXianXian\\Pycharm--code\\BertServiceBase_classify\\pbs'
sess = tf.Session()
sess.run(tf.global_variables_initializer())
meta_graph_def = tf.saved_model.loader.load(
sess, [tf.saved_model.tag_constants.SERVING], export_dir)
signature = meta_graph_def.signature_def
feed_dict = {
"input_ids": [ids],
"input_mask": [mask],
"segment_ids": [segment_ids]
}
logits = predict(sess, feed_dict=feed_dict, signature=signature)
print(logits)
