Could not find any implementation for node {foreignNode[onnx::scatterND 345 ... Transpose 19675]}

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

Description

I have converted my pytorch model to onnx, and onnx.checker.check_model() was used to check the onnx model, there is no error.

But when i convert the onnx to trt with Python API or trtexec, i got the same error – “could not find any implementation for node {foreignNode[onnx::scatterND 345 … Transpose 19675]}.”

Environment

TensorRT Version: 8.6.1.6
GPU Type: Jetson
Nvidia Driver Version:
CUDA Version: 12.0
CUDNN Version: 8.6
Operating System + Version: linux ubuntu20.04
Python Version (if applicable): 3.8
TensorFlow Version (if applicable):
PyTorch Version (if applicable): 1.11
Baremetal or Container (if container which image + tag):

Relevant Files

My onnx model is here:
Uploading: small.zip…

Steps To Reproduce

my command is here:
trtexec --onnx=small.onnx --saveEngine==small.trt --fp16 --verbose

the error is:

error
error1920×1080 294 KB

2

Hi @846443037 ,
This is a known issue by CUDA bug and fixed in the latest SW.
Thanks

3

Thanks for your reply !

The work run in Xvaier NX, NX was installed with JetPack5.1.3 by default, and cuda supports up to 11.4. I installed cuda12.0 and 12.4 manually. The same error occurred under both versions of cuda.

4

Yes, with the best of my knowledge, the bug has been addressed to be resolved in latest SW.
however if the issue still persist, i would recommend you to reach out to JEtson Xavier Forum.

Thanks

5

OK, thank you !

6

Have you solved this problem？