未验证 提交 4d4e26dc 编写于 作者: J Jacob Devlin 提交者: GitHub

Merge pull request #137 from JasonJPu/master

Fix squad 2.0 command to use 2.0 train and dev files
......@@ -462,9 +462,9 @@ python run_squad.py \
--bert_config_file=$BERT_LARGE_DIR/bert_config.json \
--init_checkpoint=$BERT_LARGE_DIR/bert_model.ckpt \
--do_train=True \
--train_file=$SQUAD_DIR/train-v1.1.json \
--train_file=$SQUAD_DIR/train-v2.0.json \
--do_predict=True \
--predict_file=$SQUAD_DIR/dev-v1.1.json \
--predict_file=$SQUAD_DIR/dev-v2.0.json \
--train_batch_size=24 \
--learning_rate=3e-5 \
--num_train_epochs=2.0 \
......@@ -498,9 +498,9 @@ python run_squad.py \
--bert_config_file=$BERT_LARGE_DIR/bert_config.json \
--init_checkpoint=$BERT_LARGE_DIR/bert_model.ckpt \
--do_train=False \
--train_file=$SQUAD_DIR/train-v1.1.json \
--train_file=$SQUAD_DIR/train-v2.0.json \
--do_predict=True \
--predict_file=$SQUAD_DIR/dev-v1.1.json \
--predict_file=$SQUAD_DIR/dev-v2.0.json \
--train_batch_size=24 \
--learning_rate=3e-5 \
--num_train_epochs=2.0 \
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册