Community
Register Help
cancel
Turn on suggestions
Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.
Showing results for 
Search instead for 
Did you mean: 
idata
Community Manager
‎10-18-2018 08:43 AM
414 Views

How to compile a model trianed with tensorflow estimator

I have trained a tensorflow model using the code for tensorflow triplet loss. The model was trained using a Tensorflow estimator. I tried to freeze the graph and run mvNCCompile but I'm getting errors. The list of operations is the following:

 

global_step/Initializer/zeros/shape_as_tensor

 

global_step/Initializer/zeros/Const

 

global_step/Initializer/zeros

 

global_step

 

global_step/Assign

 

global_step/read

 

global_step/IsVariableInitialized

 

global_step/cond/Switch

 

global_step/cond/switch_t

 

global_step/cond/switch_f

 

global_step/cond/pred_id

 

global_step/cond/read

 

global_step/cond/read/Switch

 

global_step/cond/Switch_1

 

global_step/cond/Merge

 

global_step/add/y

 

global_step/add

 

filenames

 

compression_type

 

buffer_size

 

buffer_size_1

 

seed

 

seed2

 

count

 

batch_size

 

buffer_size_2

 

Iterator

 

TFRecordDataset

 

MapDataset

 

ShuffleDataset

 

RepeatDataset

 

BatchDataset

 

PrefetchDataset

 

MakeIterator

 

IteratorToStringHandle

 

IteratorGetNext

 

Reshape/shape

 

Reshape

 

model/block_1/conv2d/kernel/Initializer/random_uniform/shape

 

model/block_1/conv2d/kernel/Initializer/random_uniform/min

 

model/block_1/conv2d/kernel/Initializer/random_uniform/max

 

model/block_1/conv2d/kernel/Initializer/random_uniform/RandomUniform

 

model/block_1/conv2d/kernel/Initializer/random_uniform/sub

 

model/block_1/conv2d/kernel/Initializer/random_uniform/mul

 

model/block_1/conv2d/kernel/Initializer/random_uniform

 

model/block_1/conv2d/kernel

 

model/block_1/conv2d/kernel/Assign

 

model/block_1/conv2d/kernel/read

 

model/block_1/conv2d/bias/Initializer/zeros/shape_as_tensor

 

model/block_1/conv2d/bias/Initializer/zeros/Const

 

model/block_1/conv2d/bias/Initializer/zeros

 

model/block_1/conv2d/bias

 

model/block_1/conv2d/bias/Assign

 

model/block_1/conv2d/bias/read

 

model/block_1/conv2d/dilation_rate

 

model/block_1/conv2d/Conv2D

 

model/block_1/conv2d/BiasAdd

 

model/block_1/Relu

 

model/block_1/max_pooling2d/MaxPool

 

[other convolutions…]

 

model/block_4/max_pooling2d/MaxPool

 

model/Flatten/flatten/Shape

 

model/Flatten/flatten/strided_slice/stack

 

model/Flatten/flatten/strided_slice/stack_1

 

model/Flatten/flatten/strided_slice/stack_2

 

model/Flatten/flatten/strided_slice

 

model/Flatten/flatten/Reshape/shape/1

 

model/Flatten/flatten/Reshape/shape

 

model/Flatten/flatten/Reshape

 

model/fc_1/dense/kernel/Initializer/random_uniform/shape

 

[fully connected operations…]

 

model/fc_2/dense/bias/read

 

model/fc_2/dense/MatMul

 

model/fc_2/dense/BiasAdd

 

[other operations]

 

I am confused about what the input node to be passed to mvNCCompile should be, since there is not a "input" node.

Tags (1)
0 Kudos
Reply

Link Copied

0 Replies

For more complete information about compiler optimizations, see our Optimization Notice.