Converting an .etlt to .engine Is deterministic?

Hi Ive been exporting tao trained yolov4 with tao-converter:

./tao-converter -k key -d 3,480,864 -p Input,1x3x480x864,1x3x480x864,1x3x480x864 -e trt_jpk.engine yolov4_resnet18_epoch_010.etlt

how come I use the exact same command twice and get 2 different file sizes? is the converter not determinstic?
image

It should be deterministic. Is the size different every time?

Yes.
I tried a couple of times and get different sizes.

After checking, the converter is not deterministic.
So, the result is normal.

1 Like

Is it possible to elaborate why this happens?
I want to have 2 engine files on different platforms,
but I want them to be exactly the same.

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

It is related to tactic.
If using trtexec to load tactic, it is deterministic.

There will be onnx file available in next release, so user can use trtexec.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.