bert-as-service配置

环境配置

conda create -n bert-service python=3.8
conda activate bert-service
pip install --user nvidia-pyindex
pip install --user nvidia-tensorflow[horovod]
pip install bert-serving-server
pip install protobuf==3.20.0

测试是否安装成功

import tensorflow as tf
tf.test.is_gpu_available()

开启服务

# bert-serving-start -model_dir F:/Bert/wwm_cased_L-24_H-1024_A-16/ -max_seq_len=None -num_worker=1
bert-serving-start -model_dir /mnt/f/model/Bert/biobert_v11_pubmed -ckpt_name=biobert_model_pubmed_v11 -num_worker=1

调用服务

from bert_serving.client import BertClient
bc = BertClient()
bc.encode(['str'])