运行 run_squad.py 微调 Google BERT 模型时加载（恢复）TensorFlow 检查点失败（官方 tensorflow 预训练模型）

Question

我是深度学习和 NLP 的新手，现在尝试开始使用预训练的 Google BERT 模型。由于我打算用 BERT 构建一个 QA 系统，所以我决定从 SQuAD 相关的微调开始。

我按照the official Google BERT GitHub repository 中的 README.md 中的说明进行操作。

我输入的代码如下：

export BERT_BASE_DIR=/home/bert/Dev/venv/uncased_L-12_H-768_A-12/
export SQUAD_DIR=/home/bert/Dev/venv/squad
python run_squad.py \
  --vocab_file=$BERT_BASE_DIR/vocab.txt \
  --bert_config_file=$BERT_BASE_DIR/bert_config.json \
  --init_checkpoint=$BERT_BASE_DIR/bert_model.ckpt \
  --do_train=True \
  --train_file=$SQUAD_DIR/train-v1.1.json \
  --do_predict=True \
  --predict_file=$SQUAD_DIR/dev-v1.1.json \
  --train_batch_size=12 \
  --learning_rate=3e-5 \
  --num_train_epochs=2.0 \
  --max_seq_length=384 \
  --doc_stride=128 \
  --output_dir=/tmp/squad_base/

几分钟后（培训开始时），我得到了这个：

a lot of output omitted
INFO:tensorflow:start_position: 53
INFO:tensorflow:end_position: 54
INFO:tensorflow:answer: february 1848
INFO:tensorflow:***** Running training *****
INFO:tensorflow:  Num orig examples = 87599
INFO:tensorflow:  Num split examples = 88641
INFO:tensorflow:  Batch size = 12
INFO:tensorflow:  Num steps = 14599
INFO:tensorflow:Calling model_fn.
INFO:tensorflow:Running train on CPU
INFO:tensorflow:*** Features ***
INFO:tensorflow:  name = end_positions, shape = (12,)
INFO:tensorflow:  name = input_ids, shape = (12, 384)
INFO:tensorflow:  name = input_mask, shape = (12, 384)
INFO:tensorflow:  name = segment_ids, shape = (12, 384)
INFO:tensorflow:  name = start_positions, shape = (12,)
INFO:tensorflow:  name = unique_ids, shape = (12,)
INFO:tensorflow:Error recorded from training_loop: Unsuccessful TensorSliceReader constructor: Failed to find any matching files for /home/bert/Dev/venv/uncased_L-12_H-768_A-12//bert_model.ckpt
INFO:tensorflow:training_loop marked as finished
WARNING:tensorflow:Reraising captured error
Traceback (most recent call last):
  File "run_squad.py", line 1283, in <module>
    tf.app.run()
  File "/home/bert/Dev/venv/lib/python3.5/site-packages/tensorflow/python/platform/app.py", line 125, in run
    _sys.exit(main(argv))
  File "run_squad.py", line 1215, in main
    estimator.train(input_fn=train_input_fn, max_steps=num_train_steps)
  File "/home/bert/Dev/venv/lib/python3.5/site-packages/tensorflow/contrib/tpu/python/tpu/tpu_estimator.py", line 2400, in train
    rendezvous.raise_errors()
  File "/home/bert/Dev/venv/lib/python3.5/site-packages/tensorflow/contrib/tpu/python/tpu/error_handling.py", line 128, in raise_errors
    six.reraise(typ, value, traceback)
  File "/home/bert/Dev/venv/lib/python3.5/site-packages/six.py", line 693, in reraise
    raise value
  File "/home/bert/Dev/venv/lib/python3.5/site-packages/tensorflow/contrib/tpu/python/tpu/tpu_estimator.py", line 2394, in train
    saving_listeners=saving_listeners
  File "/home/bert/Dev/venv/lib/python3.5/site-packages/tensorflow/python/estimator/estimator.py", line 356, in train
    loss = self._train_model(input_fn, hooks, saving_listeners)
  File "/home/bert/Dev/venv/lib/python3.5/site-packages/tensorflow/python/estimator/estimator.py", line 1181, in _train_model
    return self._train_model_default(input_fn, hooks, saving_listeners)
  File "/home/bert/Dev/venv/lib/python3.5/site-packages/tensorflow/python/estimator/estimator.py", line 1211, in _train_model_default
    features, labels, model_fn_lib.ModeKeys.TRAIN, self.config)
  File "/home/bert/Dev/venv/lib/python3.5/site-packages/tensorflow/contrib/tpu/python/tpu/tpu_estimator.py", line 2186, in _call_model_fn
    features, labels, mode, config)
  File "/home/bert/Dev/venv/lib/python3.5/site-packages/tensorflow/python/estimator/estimator.py", line 1169, in _call_model_fn
    model_fn_results = self._model_fn(features=features, **kwargs)
  File "/home/bert/Dev/venv/lib/python3.5/site-packages/tensorflow/contrib/tpu/python/tpu/tpu_estimator.py", line 2470, in _model_fn
    features, labels, is_export_mode=is_export_mode)
  File "/home/bert/Dev/venv/lib/python3.5/site-packages/tensorflow/contrib/tpu/python/tpu/tpu_estimator.py", line 1250, in call_without_tpu
    return self._call_model_fn(features, labels, is_export_mode=is_export_mode)
  File "/home/bert/Dev/venv/lib/python3.5/site-packages/tensorflow/contrib/tpu/python/tpu/tpu_estimator.py", line 1524, in _call_model_fn
    estimator_spec = self._model_fn(features=features, **kwargs)
  File "run_squad.py", line 623, in model_fn
    ) = modeling.get_assignment_map_from_checkpoint(tvars, init_checkpoint)
  File "/home/bert/Dev/venv/bert/modeling.py", line 330, in get_assignment_map_from_checkpoint
    init_vars = tf.train.list_variables(init_checkpoint)
  File "/home/bert/Dev/venv/lib/python3.5/site-packages/tensorflow/python/training/checkpoint_utils.py", line 95, in list_variables
    reader = load_checkpoint(ckpt_dir_or_file)
  File "/home/bert/Dev/venv/lib/python3.5/site-packages/tensorflow/python/training/checkpoint_utils.py", line 64, in load_checkpoint
    return pywrap_tensorflow.NewCheckpointReader(filename)
  File "/home/bert/Dev/venv/lib/python3.5/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 314, in NewCheckpointReader
    return CheckpointReader(compat.as_bytes(filepattern), status)
  File "/home/bert/Dev/venv/lib/python3.5/site-packages/tensorflow/python/framework/errors_impl.py", line 526, in __exit__
    c_api.TF_GetCode(self.status.status))
tensorflow.python.framework.errors_impl.NotFoundError: Unsuccessful TensorSliceReader constructor: Failed to find any matching files for /home/bert/Dev/venv/uncased_L-12_H-768_A-12//bert_model.ckpt

似乎tensorflow找不到检查点文件，但据我所知，一个tensorflow检查点“文件”实际上是三个文件，这是正确的调用方式（带有路径和前缀）。

我相信我将文件放置在正确的位置：

(venv) bert@bert-System-Product-Name:~/Dev/venv/uncased_L-12_H-768_A-12$ pwd
/home/bert/Dev/venv/uncased_L-12_H-768_A-12
(venv) bert@bert-System-Product-Name:~/Dev/venv/uncased_L-12_H-768_A-12$ ls
bert_config.json  bert_model.ckpt.data-00000-of-00001  bert_model.ckpt.index  bert_model.ckpt.meta  vocab.txt

我在 Ubuntu 16.04 LTS 上运行 , 使用 NVIDIA GTX 1080 Ti (CUDA 9.0) , 使用 Anaconda python 3.5 发行版 ，在虚拟环境中使用 tensorflow-gpu 1.11.0。

我希望代码能够顺利运行并开始训练（微调），因为它是官方代码，并且我将文件作为说明放置。

Answer 1

我正在回答我自己的问题。

我刚刚通过删除$BERT_BASE_DIR 中的斜杠(/) 解决了这个问题，因此变量从'/home/bert/Dev/venv/uncased_L-12_H-768_A-12/' 更改为'/home/bert/Dev/venv/uncased_L-12_H-768_A-12'。

所以前缀"/home/bert/Dev/venv/uncased_L-12_H-768_A-12//bert_model.ckpt" 中不再有双斜线。

似乎单斜杠或双斜杠被 tensorflow 中的检查点恢复功能认为是不同的，因为我相信 bash 将它们解释为相同。