You can follow deepstream_tao_apps/TRT-OSS at master · NVIDIA-AI-IOT/deepstream_tao_apps · GitHub to build the libnvinfer_plugin.so. Then generate the trt engine via the tool tlt-converter.
Then let triton inference server be able to recognize TRT engine.
Reference:
Integrating TAO Models into DeepStream — TAO Toolkit 3.22.05 documentation
Using TensorRT Inference Server with TLT models - #6 by andrliu