Closed Hperigo closed 3 months ago
Engine file should be created for your gpu. So if I put engine in repo, it wouldn't work for most people. It is recommend to build the engine for your gpu and cuda version. Yes it is possible to load onnx file and create an engine without using trtexec.exe. Early version of the code has this functionality but we removed it to simplify the code. You may check at early commits.
Add the TensorRT library files to your system PATH. To do so, copy the DLL files from tensorrt_install_path/lib to your CUDA installation directory, for example, C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\vX.Y\bin, where vX.Y is your CUDA version. The CUDA installer should have already added the CUDA path to your system PATH.
Or you can add tensorrt dlls in lib dir to bin dir where trtexec locates.
Printing the error message here could help.
Hello, thank you for taking the time to answer.
I started following your suggestion of implementing the old code and doing the conversion from inside c++. When I got that compiled I got an error saying it was unable to load the dll from cudnn 8.8
I installed cudnn and the error got fixed. Probably nvinfer_plugin.dll
was trying to load something from there and it was failing without giving any errors indicating that..
Note that I had to run export_to_onxx.py
using python 3.10 ( It's just the other version I had installed, maybe 3.11 can work as well) but 3.12 failed to install the onnx package via pip
gonna close the issue :)
Hi, I added build engine functionality in recent version:
Hello,
I'm trying to run the command to convert the
.onxx
model to the.engine
extension but I'm running in this issue where thenvinfer_plugin.dll
cannot be open.I triple checked things and the DLL is in the correct PATH ( I tried both adding it to the PATH and also copying the files to the cuda bin and lib folder) but I always get the same error message.
Any guidance would be greatly appreciated.
I tried with both
TensorRT-8.6.1.6.Windows10.x86_64.cuda-12.0
andTensorRT-8.6.0.12.Windows10.x86_64.cuda-12.0
; as well asCuda 12.0
andCuda 12.4
a couple of extra questions:
.engine
model in the repo ?thank you