Add custom tactics for TensorRT

Through the TensorRT verbose log and profile result, I’m not satisfied with some layers’ performance. And I’ve implemented faster CUDA kernels for these layers. Can I use my kernels as the external tactics while building the TensorRT engine?

Hi,
Please refer to below links related custom plugin implementation and sample:
https://github.com/NVIDIA/TensorRT/tree/master/samples/opensource/sampleOnnxMnistCoordConvAC

Thanks!

1 Like

Hi @439290087,

Hope following will help you. Please let us know if you still have a query.
https://github.com/NVIDIA/TensorRT/issues/1049

Thank you.