TensorRT icon indicating copy to clipboard operation
TensorRT copied to clipboard

Published 20 hours ago • verified publisher icon NVIDIA

why plugin？

Open dingjingzhen opened this issue 2 years ago • 1 comments

I recently tried to write a plugin, and it was a pain to write it. Even after I wrote it, there were still a lot of problems when compiling it, and it might be different on different machines. Why can't I skip the arithmetic that can't turn trt and just use the original or use the cuda kernel written by the user himself? The threshold for using plugin is indeed a bit high.

dingjingzhen avatar Aug 04 '22 02:08 dingjingzhen

I'm developing a tool that can produce the TRT plugin from a simple config file, so the user only needs to focus on the kernel implementation of the operator :)

zerollzeng avatar Aug 04 '22 14:08 zerollzeng

closing since no activity for more than 3 weeks, please reopen if you still have question, thanks!

ttyio avatar Dec 06 '22 01:12 ttyio

the tool is public now: https://github.com/NVIDIA-AI-IOT/tensorrt_plugin_generator

zerollzeng avatar Dec 07 '22 13:12 zerollzeng

the tool is public now: https://github.com/NVIDIA-AI-IOT/tensorrt_plugin_generator

Thanks, I'll give it a try

dingjingzhen avatar Dec 08 '22 01:12 dingjingzhen