Tensorflow OOM after freeze graph - tensorflow

Question

I'm running a seq2seq model with tf, the inference program runs well when loading parameters from checkpoint file using tf.train.Saver. But after exporting the graph with freeze_graph.py (using tf.framework.graph_util.convert_variables_to_constants()), and import with tf.import_graph_def in the inference program, it got OOM problem.
Here is a part of error log:
W tensorflow/core/common_runtime/bfc_allocator.cc:274] ****************************************************************************************************
W tensorflow/core/common_runtime/bfc_allocator.cc:275] Ran out of memory trying to allocate 4.0KiB. See logs for memory state.
W tensorflow/core/framework/op_kernel.cc:983] Internal: Dst tensor is not initialized.
E tensorflow/core/common_runtime/executor.cc:594] Executor failed to create kernel. Internal: Dst tensor is not initialized.
[[Node: embedding_attention_seq2seq/embedding_attention_decoder/attention_decoder/AttnV_0 = Const[dtype=DT_FLOAT, value=Tensor<type: float shape: [1024] values: -0.016628871 -0.2054652 -0.045054652...>, _device="/job:localhost/replica:0/task:0/gpu:0"]()]]
Traceback (most recent call last):
File "inference.py", line 88, in console_main
result = list(inference(source_sentence))
File "inference.py", line 54, in inference
for sequence in result:
File "/data/experiment/decoder.py", line 115, in search_best_sequence
State.batch_predict(self.session, self.model, self.context, beam)
File "/data/experiment/decoder.py", line 82, in batch_predict
state_list[0].depth)
File "/data/experiment/seq2seq_model.py", line 452, in batch_feed_decoder
log_softmax, attns, state = session.run(output_fetch, input_feed)
File "/home/.conda/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 767, in run
run_metadata_ptr)
File "/home/.conda/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 966, in _run
feed_dict_string, options, run_metadata)
File "/home/.conda/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1016, in _do_run
target_list, options, run_metadata)
File "/home/.conda/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1036, in _do_call
raise type(e)(node_def, op, message)
InternalError: Dst tensor is not initialized.
[[Node: embedding_attention_seq2seq/embedding_attention_decoder/attention_decoder/AttnV_0 = Const[dtype=DT_FLOAT, value=Tensor<type: float shape: [1024] values: -0.016628871 -0.2054652 -0.045054652...>, _device="/job:localhost/replica:0/task:0/gpu:0"]()]]
Caused by op u'embedding_attention_seq2seq/embedding_attention_decoder/attention_decoder/AttnV_0', defined at:
File "inference.py", line 169, in <module>
tf.app.run()
File "/home/.conda/lib/python2.7/site-packages/tensorflow/python/platform/app.py", line 44, in run
_sys.exit(main(_sys.argv[:1] + flags_passthrough))
File "inference.py", line 165, in main
console_main(session)
File "inference.py", line 66, in console_main
model = create_model(session, False)
File "/data/experiment/model.py", line 145, in create_model
tensor_name_pickle=tensor_name_pickle)
File "/data/experiment/seq2seq_model.py", line 106, in __init__
tf.import_graph_def(graph_def, name="")
File "/home/.conda/lib/python2.7/site-packages/tensorflow/python/framework/importer.py", line 287, in import_graph_def
op_def=op_def)
File "/home/.conda/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 2395, in create_op
original_op=self._default_original_op, op_def=op_def)
File "/home/.conda/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 1264, in __init__
self._traceback = _extract_stack()
InternalError (see above for traceback): Dst tensor is not initialized.
[[Node: embedding_attention_seq2seq/embedding_attention_decoder/attention_decoder/AttnV_0 = Const[dtype=DT_FLOAT, value=Tensor<type: float shape: [1024] values: -0.016628871 -0.2054652 -0.045054652...>, _device="/job:localhost/replica:0/task:0/gpu:0"]()]]
I thought it might cause by the memory issue of tf.Constant. Does someone have experience with this problem?

Answer 1

I had the same issue but when trying to load and run the inference from a C++ application using the C API. After a lot of twiddling and testing it appeared the culprit was the frozen graph and freeze_graph.py itself. It's probably a bug of some kind. There are actually multiple issue reports on github's TF repo, but they were just closed due to lack of activity, e.g. here and here. I guess apparent bugs of model freezing aren't of any priority.
In my case the model .pb file was around 500mb and it took around 10Gb of RAM while running a session. Not only did it occupy an insane amount of RAM, it was actually orders of magnitudes slower that way.
When I switched to loading just a SavedModel directory everything went to normal. I'm not sure how to achieve that in python, but for C code I replaced a TF_GraphImportGraphDef() call with TF_LoadSessionFromSavedModel().
I used TF v1.14.0. The library is built with Bazel by me, not the stock version. I could provide some details here and there if anybody was interested. Just not sure where to start, I had many trials and errors.