使用pytorch版本scibert环境配置与使用

1.环境配置：python3.6

conda install pytorch=1.3.1

pip install --upgrade pip

pip install transformers

详细见https://github.com/huggingface/transformers

sudo apt-get install git

pip install -r requirements.txt 

pip install --upgrade .

详细见https://github.com/allenai/scibert

 

2.下载训练好的模型

https://s3-us-west-2.amazonaws.com/ai2-s2-research/scibert/pytorch_models/scibert_scivocab_uncased.tar

详细见https://github.com/allenai/scibert

 

3.运行代码

更改scibert/train_allennlp_local.sh的路径变量

命令行运行

./scibert/scripts/train_allennlp_local.sh [serialization-directory]

其中[serialization-directory]为输出路径

详细见https://github.com/allenai/scibert