The same onnx file produces engine files with different size under same configurations

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

the Tensor-RT version used is 8.0.1.6 Win10, CUDA 11.1.1, cudnn v8.2, I generate engine file as following
.\trtexec --onnx=lanenet_free_graphy.onnx --saveEngine=lanenet_trt8.0.1.6.engine --verbose

the first time, the size of lanenet_trt8.0.1.6.engine is 9.991KB
the second time, the size of lanenet_trt8.0.1.6.engine is 9.962KB
the third time, the size of lanenet_trt8.0.1.6.engine is 10434KB

It seems that the size of engine file fluctuates around 10KB,not a fixed number,why?
Thanks in advance.

#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.

In case you are still facing issue, request you to share the trtexec “”–verbose"" log for further debugging
Thanks!

#4

I use your code to check if the onnx model is valid, the result is valid

the engine files are :
lanenet_trt8.0.1.6v1.engine (9.8 MB)
lanenet_trt8.0.1.6v2.engine (9.7 MB)
lanenet_trt8.0.1.6v3.engine (10.2 MB)
lanenet_trt8.0.1.6v4.engine (9.8 MB)

here are my onnx model and verbose log

lanenet_free_graphy.onnx (9.0 MB)

and the log
lanenet_trt8.0.1.6.engine.log (2.3 MB)

#5

Hi,

We couldn’t reproduce the same issue.
Could you please try on the latest TebsorRT version 8.4.

https://developer.nvidia.com/nvidia-tensorrt-8x-download

Also please confirm with us, are you using the same machine for everytime build?

Thank you.

#6

OK,I will have a try on TensorRT v8.4