Tensorrt8.5 inference different with origin onnx model

[spolisetty]

Hi @liwenjudetiankong,

@spolisetty

@liwenjudetiankong

i have trained a model with pytorch and export it to onnx format model, this onnx model predicts correct value. but when i convert it to tensorrt engine, it gives wrong inference.

Could you please elaborate more on the above highlighted.
Are you looking for 100% numerical accuracy in comparison with ONNX-Runtime results?

@spolisetty

Our engineers think the polygraphy issue is just a tolerance issue. With “–atol 1e-4 --rtol 1e-4”, the polygraphy check could get passed, while the default value is “–atol 1e-5 --rtol 1e-5”.

Could you also please give more info on requested.

Thank you.