Error Code 1: Serialization + device GPU, failed to create CUDA engine

[jpbaehr]

Man I am heaps full of problems today.

Background:
I cloned my working disk IMG to a higher capacity drive so I have more room to install things. Everything seems to be working good, up until I try running my previously working script that run an object detection model I trained.

Error:
[TRT] 1: [stdArchiveReader.cpp::StdArchiveReader::54] Error Code 1: Serialization (Serialization assertion sizeRead == static_cast<uint64_t>(mEnd - mCurrent) failed.Size specified in header does not match archive size)
[TRT] 4: [runtime.cpp::deserializeCudaEngine::50] Error Code 4: Internal Error (Engine deserialization failed.)
[TRT] device GPU, failed to create CUDA engine
[TRT] failed to create TensorRT engine for /home/jb/catkin_ws/src/jetson-inference/python/training/detection/ssd/models/eraser/ssd-mobilenet.onnx, device GPU
[TRT] detectNet – failed to initialize.
jetson.inference – detectNet failed to load network

Thank you again for everyone that helps, posting my problems benefits me as I can record and learn from it efficiently, while helping others in the future.

[kayccc]

Which JetPack version you’re using?

[jpbaehr]

4.4

[AastaLLL]

Hi,

The error indicates that the engine file is serialized with a different TensorRT version.

The TensorRT engine doesn’t support portability.
Both hardware (GPU architecture) and software (library version) need to be the same in the serializing and deserializing stage.

Thanks.

[jpbaehr]

Is there a method I can use to put them back in the same serializing and deserializing stage? Can you elaborate what that means specifically? Thanks

[AastaLLL]

Hi,

You can serialize the engine from ONNX or other model formats again.
Thanks.