How to read the PLAN file?

AI & Data Science Deep Learning (Training & Inference) TensorRT
1

Hi,

I’ve export the serialized engine into a PLAN file using C++ following the answer of this thread:

[url]https://devtalk.nvidia.com/default/topic/1029737/jetson-tx1/tensorrt-saving-models-as-portable-plan-files/[/url]

However, the exported PLAN file is just the cache and can only be directly imported by the tensorRT api.

So, can we decode the PLAN file to give meaningful information to humans?

Thanks.

2

I have the same question. Thanks.

3

It seems that the PLAN file can’t be decoded explicitly.
For me, I was using the INT8 calibration and wanted to get the INT8 quantization information. Finally, I found that the file “CalibrationTable” is what I really needed.

4

Any idea about how to convert a FP32 PLAN model to a FP16 one?