failed to build the TensorRT engine

[soukaina11111]

Hi @jkjung-avt first, thank you for your amazing work,

I have tested this demo on jetson tx2 and it’s work! and i wanted to test it on jetson xavier, i succefully convert yolov4.weights to yolov4.onnx, but when I try to convert onnx to tensorrt without int8 and dla_core i got this error :

`python onnx_to_tensorrt.py -m yolov4

Loading the ONNX file...

Adding yolo_layer plugins.

Adding a concatenated output as "detections".

Naming the input tensort as "input".

Building the TensorRT engine. This would take a while...

(Use "--verbose" or "-v" to enable verbose logging.)

onnx_to_tensorrt.py:148: DeprecationWarning: Use set_memory_pool_limit instead.

config.max_workspace_size = 1 << 30

onnx_to_tensorrt.py:170: DeprecationWarning: Use build_serialized_network instead.

engine = builder.build_engine(network, config)

[07/26/2022-11:14:38] [TRT] [W] Convolution + generic activation fusion is disable due to incompatible driver or nvrtc

[07/26/2022-11:14:39] [TRT] [E] 1: [ltWrapper.cpp::CublasLtWrapper::66] Error Code 1: Cublas (Could not initialize cublasLt.

Please check CUDA installation.)

ERROR: failed to build the TensorRT engine!`

Any idea how to solve it? best regards.

[jkjung-avt]

This might be a problem of the specific combination CUDA 10.2 and TensorRT 7.2.1.x? Please refer to: https://github.com/NVIDIA/TensorRT/issues/866

Please test with the latest version of JetPack.

[soukaina11111]

Thank you for your answer .I’m using jetpack 5.0.1, cuda 11.4 and tensorrt 8.4.0.11

[jkjung-avt]

Then I have no idea. The error message points to a problem about the cuBLAS library. Please direct your question to NVIDIA instead.

[07/26/2022-11:14:39] [TRT] [E] 1: [ltWrapper.cpp::CublasLtWrapper::66] Error Code 1: Cublas (Could not initialize cublasLt.

Please check CUDA installation.)

[soukaina11111]

Problem solved by reflashing the os, and installing the newest version of jetpack 5.0.2. thanks again :).

[jkjung-avt]

Good to hear that.