Zero Zeng

Could you please share the reproduce script then I can debug it. Thanks!

I have to say I didn't put much effort on the python api and I didn't add accuracy test for it. So it might have bugs. Source code in https://github.com/zerollzeng/tiny-tensorrt/blob/master/src/PyTrt.cpp

Sorry I miss this :-) You can include a kernel head files, just like you do in your development.

Have you solve it? Sorry I miss the issue, you can provide a reproduce and I can help take a check.

Do you mean you met error during enqueue? should be caused by the kernel implementation?

I would suggest to debug it using cuda-gdb

> I am not able to figure out how the array is passed to enqeue function. What the size of array is? It's the input of you plugin, so it...

> e.g. should I access the tpg_input_1 like > > ``` > const float* var1 = reinterpret_cast(inputs[0]); > and then access individual elements like var1[0][18256][31] ? > ``` Yes.

@jeethesh-pai You can pull the latest change, I made the default input/output data format in linear format.

trtexec use random inputs. There is also a `--loadInputs` option that let you load a real input in binary.