无法加载tensorflow BERT预训练模型

Question

我试图加载BERT预训练模型来执行NER任务。但是系统找不到预训练的模型文件。

我在终端中使用了以下代码，该文件夹包含model.ckpt-1000000，model.ckpt-1000000.index，model.ckpt-1000000.meta文件。

python run_ner.py \
    --do_train=true \
    --do_eval=true \
    --vocab_file=vocab.txt \
    --bert_config_file=bert_config.json \
    --init_checkpoint=model.ckpt-1000000 \
    --num_train_epochs=10.0 \
    --data_dir=NCBI-disease/ \
    --output_dir=epoch1

错误消息是

2019-08-04 23:26:41.272281: W tensorflow/core/framework/op_kernel.cc:1401] OP_REQUIRES failed at save_restore_v2_ops.cc:184 : Not found: model.ckpt-1000000.data-00000-of-00001; No such file or directory

  File "/Users/anaconda/envs/BIOBERT/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1328, in _do_run
    run_metadata)
  File "/Users/anaconda/envs/BIOBERT/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1348, in _do_call
    raise type(e)(node_def, op, message)
tensorflow.python.framework.errors_impl.NotFoundError: model.ckpt-1000000.data-00000-of-00001; No such file or directory
     [[node checkpoint_initializer_161 (defined at run_ner.py:422) ]]

注：模型文件的原始名称为model.ckpt-1000000.data-00000-of-00001，model.ckpt-1000000.index和model.ckpt-1000000.meta。我也试过

python run_ner.py \
    --do_train=true \
    --do_eval=true \
    --vocab_file=vocab.txt \
    --bert_config_file=bert_config.json \
    --init_checkpoint=model.ckpt-1000000.data-00000-of-00001 \
    --num_train_epochs=10.0 \
    --data_dir=NCBI-disease/ \
    --output_dir=epoch1

然后错误将是

  File "/Users/anaconda/envs/BIOBERT/lib/python3.6/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 326, in NewCheckpointReader
    return CheckpointReader(compat.as_bytes(filepattern), status)
  File "/Users/SichengZhou/anaconda/envs/BIOBERT/lib/python3.6/site-packages/tensorflow/python/framework/errors_impl.py", line 528, in __exit__
    c_api.TF_GetCode(self.status.status))
tensorflow.python.framework.errors_impl.DataLossError: Unable to open table file ./model.ckpt-1000000.data-00000-of-00001: Data loss: not an sstable (bad magic number): perhaps your file is in a different file format and you need to use a different restore operator?

Answer 1

模型文件的原始名称应为model.ckpt-1000000.data-00000-of-00001，model.ckpt-1000000.index和model.ckpt-1000000.meta。

您收到的错误是因为正在运行的ckpt文件和python文件（run_ner.py）位于不同的路径。

python run_ner.py \
--do_train=true \
--do_eval=true \
--vocab_file=vocab.txt \
--bert_config_file=bert_config.json \
--init_checkpoint=<path to folder where ckpt files are saved>/model.ckpt-1000000 \
--num_train_epochs=10.0 \
--data_dir=NCBI-disease/ \
--output_dir=epoch1

运行--init_checkpoint=model.ckpt-1000000时请勿将--init_checkpoint=model.ckpt-1000000.data-00000-of-00001更改为run_ner.py。>

希望这会有所帮助。