tensorflow - 来自 TF.Estimator 的 .pbtxt 的 freeze_graph MemoryError

Question

我正在尝试冻结 Multi-GPU TF Estimator 输出的检查点的图表。这是我使用的命令。

python -m tensorflow.python.tools.freeze_graph --input_graph models/graph.pbtxt --input_checkpoint models/model.ckpt-40886 --output_graph hello.pb --output_node_names Decoder/Layer8/filter --input_binary=false

它出现以下 MemoryError 错误。graph.pbtxt 文件大约 1.5GBs，我的机器有 64Gigs 的 RAM。

  .....
  .....
  .....
  File "/home/will/pyenv3/lib/python3.6/site-packages/google/protobuf/text_format.py", line 837, in _MergeMessageField
self._MergeField(tokenizer, sub_message)
  File "/home/will/pyenv3/lib/python3.6/site-packages/google/protobuf/text_format.py", line 749, in _MergeField
tokenizer.Consume(':')
  File "/home/will/pyenv3/lib/python3.6/site-packages/google/protobuf/text_format.py", line 1100, in Consume
if not self.TryConsume(token):
  File "/home/will/pyenv3/lib/python3.6/site-packages/google/protobuf/text_format.py", line 1087, in TryConsume
self.NextToken()
  File "/home/will/pyenv3/lib/python3.6/site-packages/google/protobuf/text_format.py", line 1336, in NextToken
match = self._TOKEN.match(self._current_line, self._column)
  MemoryError

有没有办法让模型冻结来处理 TF Estimator 的训练检查点输出？

标签： tensorflowtensorflow-estimator