how to use the training results

[eejlny]

Hello,

I would appreciate very much if you could give some further insights on how to take the output of the training process into the pynq hardware.

Basically I have the LSTM running correctly in a Zynq Board with the design files provided in https://github.com/Xilinx/LSTM-PYNQ.

Now I have used the setup in here to retrain the system for a different problem. In the checkpoints directory there are the tar files with the best word error rate. I would like to use these training results in the board but I cannot find some instructions on how to do that. Could you advise, please ?

THanks,

[volcacius]

Hi,

As the README says, you can invoke the main with the --export flag together with the desired number of SIMD and PE (--simd_factor and --pe), together with a pretrained model (--input). If you look at how the function call propagates through the code you can get a picture of how _r_model_fwbw.h is generated.

Alessandro