Closed hansely closed 5 years ago
Yes, it means multiplying each element in the tensor with a scalar, and it is essentially the same as a 0-d tensor. The fact that in this case it happens to be 1.0 is just a coincidence in this particular example, it was like that in the original model from which it was converted, but it could be any other scalar.
Constants are not saved to a binary file, they are directly included inside the graph, but they have the same effect.
I get it. Thanks!
Hi, I was looking at the sample NNEF models on the website and found some instructions with scalar values for the tensor. For example in the vgg19, it has
mul_5 = mul(1.0, variable_37) where mul operation requires two tensors as inputs. Does above operation mean multiplying 1.0 to each elements in the tensor? If yes, what is the point of multiplying 1?
In addition, in NNEF, each tensor variables are saved as a binary data file. Does the 0-d tensor also saved as a binary data file and loaded when load_graph function is called?