/usr/local/bin/ncsdk/Controllers/FileIO.py:65: UserWarning: You are using a large type. Consider reducing your data sizes for best performance
I've changed all the tensors in my graph to tf.float16 and I'm still getting the above warning during compile. 'input:0' and 'output:0' tensors both print as dtype = 'float16'
I've looped through all the tensors in the graph and printed their dtype. The only odd thing I noticed was that dtype was listed as "float16_ref' for many variables.
Maybe related, but no matter what type I make my placeholder, the API complains unless I pass it a float32. Maybe this is expected behavior in V2?