Number of operations in a TensorRT model

Hello all,

Is there a way to calculate the number of floating-point operations for executing a model using TensorRT?

Tensoflow has a very convenient way to see how many fp operations a network has (https://stackoverflow.com/a/51082939/2794625):

flops = tf.profiler.profile(graph, options=tf.profiler.ProfileOptionBuilder.float_operation())
print('FLOP = ', flops.total_float_ops)

Is there any similar approach for an optimized TensorRT model too?

Thanks,

Any updates on this?

We don’t have such a utility as of today.
Anyway, FLOPS should not change going from TF to TRT (assuming network does not have redundant branches which don’t contribute to the network outputs).

Thanks