Precision missing while writing the GRU layer in c++

C++
1

Hi,

 I extracted the GRU weights from the trained model and followed the instructions in pytorch document.

But, while comparing the results of model’s GRU outputs and c++ GRU output precision is missing by 0.1(maximum of all results). Can you suggest me to overcome this.
Advance thanks.

2

Could you post the full Python and C++ code for your experiment?

3

IIUC, is this resolved in GRU calculations for gates and output in testing mode?

4

Yes, it is resolved. I made an small error in matmuls.

Thanks,
M. Dinesh

5

Can you give me suggestions on my current problem I am facing with bidirectional multilayered GRU?