TopK - This version of TensorRT only supports input K as an initializer

AI & Data Science Deep Learning (Training & Inference) TensorRT
, ,
5

Hi,

Yes. The dynamic K value is not supported in the TensorRT now. We have a plan for supporting in the future but we are currently not sure about the ETA.
This looks error looks due to the dynamic K value.
You can try running constant folding with Polygraphy or use GitHub - daquexian/onnx-simplifier: Simplify your onnx model as a workaround.
polygraphy surgeon sanitize model.onnx --fold-constants --output model_folded.onnx

Thank you.