Why tacotron2 model separated into 3 parts?

AI & Data Science Deep Learning (Training & Inference) TensorRT
1

I am reading the source code of TensorRT TensorRT/demo/Tacotron2/tensorrt at main · NVIDIA/TensorRT · GitHub.
The Tacotron2 model has been split into three parts: Encoder, Deocder, Postnet. And convert into onnx and engine respcetivaly.
Why not convert the model into engine as a whole? Is there some reason to separate into three parts?

2

Hi,
Request you to share the ONNX model and the script if not shared already so that we can assist you better.
Alongside you can try few things:

  1. validating your model with the below snippet

check_model.py

import sys
import onnx
filename = yourONNXmodel
model = onnx.load(filename)
onnx.checker.check_model(model).
2) Try running your model with trtexec command.
https://github.com/NVIDIA/TensorRT/tree/master/samples/opensource/trtexec
In case you are still facing issue, request you to share the trtexec “”–verbose"" log for further debugging
Thanks!

4

Hi,

The onnx model is worked.
I just wonder why they convert Encoder, Postnet and Decoder respectivaly in the official sample,(see TensorRT github Repositories https://github.com/NVIDIA/TensorRT/blob/main/demo/Tacotron2/tensorrt/convert_tacotron22onnx.py).
Why they didn’t convert tacotron2 into one engine?

5

Hi,

The number of iterations that the decoder runs for is data-dependent. This causes the input shapes to Postnet become data-dependent which TRT does not support as of now.

Thank you.