search

NVIDIA / TensorRT

NVIDIA® TensorRT™ is an SDK for high-performance deep learning inference on NVIDIA GPUs. This repository contains the open source components of TensorRT.
https://developer.nvidia.com/tensorrt
Apache License 2.0
10.75k stars 2.13k forks source link

Internal Error (Assertion paramWeights.count() == aspectWeights.count() failed.) when convert to engine #1637

Closed twmht closed 2 years ago

twmht commented 2 years ago

Description

I use the quantization toolkit to quantize my model.

but TensorRT threw the exceptions when convert the model to tensorrt engine

[11/24/2021-18:03:50] [E] Error[2]: [layerAspects.cpp::flipAspects::24] Error Code 2: Internal Error (Assertion paramWeights.count() == aspectWeights.count() failed.)
[11/24/2021-18:03:50] [E] Error[2]: [builder.cpp::buildSerializedNetwork::417] Error Code 2: Internal Error (Assertion enginePtr != nullptr failed.)

the model is first trained from mmdetection and then quantize by the toolkit.

the onnx file is provided as below.

Environment

TensorRT Version: 8.0.1.6 NVIDIA GPU: 1080 ti NVIDIA Driver Version: 460.73.01 CUDA Version: 11.1 CUDNN Version: 8.2.1 Operating System: ubuntu 18.04 Python Version (if applicable): 3.7

Relevant Files

log.txt

Here are the onnx file https://drive.google.com/file/d/1kLv4NPM8gBFg_wlyj5hKElhxr_GAJF-j/view?usp=sharing

Steps To Reproduce

../TensorRT-8.0.1.6/bin/trtexec --onnx=epoch_12_calib.onnx --int8 --verbose
zerollzeng commented 2 years ago

Could you try TRT 8.2?

twmht commented 2 years ago

@zerollzeng

I can't since it's bundle with jetpack 4.6 in the production use.

liuanhua110 commented 2 years ago

meet the same erro any update?

oxana-nvidia commented 2 years ago

@twmht @liuanhua110 Could you please check if you can repro this issue with TensorRT 8.4? I've tested on my local set up and issue is not reproducible with provided model.

oxana-nvidia commented 2 years ago

Closing as no response provided in >=14 days. Please reopen if there are any updates.