When using TF-TRT, is still possible to get a .plan file?


I am reading the documentation of NVIDIA and some Githubs.
When using ONNX and UFF parsers it is possible to serialize the engine to a .plan file. I would like to know if the same can be done with TF-TRT? Will it make any difference in speed or something?

thank you


You can serialize the TF-TRT engine also.
The performance difference is similar to TensorRT vs TensorFlow-TRT.