Weightspace parameters and organisation

I’m trying to understand the parameters of weightspace.
Does LinLayer (0, 1, 2, 3) correspond to Weights in (c, i, f, o) format?
Does LinLayer (4, 5, 6, 7) correspond to Unit Weights in (c, i, f, o) format?

Finally, cudnnGetRNNWeightParams & cudnnGetTensorNdDescriptor do help me answer these questions but is the format actually {1, rows, columns } ?
If that is true I will have to create a kernel that walks through and copy each value from my weights.
Is there any other way around this?


Hi @edp ,
Apologies for delayed response, we are currently checking on this, and will update you.