Can not run with another TensorRT Engine in one program.

[Water2style]

I got a good model_trt by torch2trt, and it can run alone in a program.
But when i add an TensorRT engine in the program, it seems that program can not run an TensorRT engine and a model_trt together. But they can run correctly separately.

If I want to run two models in one program, what should i do? Thanks!

[jaybdub]

Hi Water2style,

Thanks for reaching out!

Hmm. I haven't seen this before personally. Do you mind sharing the code you used to produce this issue?

Also, the following information may also be helpful

• PyTorch Version
• TensorRT Version
• Platform (ie: Jetson Xavier)

Best,
John

[Water2style]

Hi Water2style,

Thanks for reaching out!

Hmm. I haven't seen this before personally. Do you mind sharing the code you used to produce this issue?

Also, the following information may also be helpful

• PyTorch Version
• TensorRT Version
• Platform (ie: Jetson Xavier)

Best,
John

Sorry i cannot share the code, but it just a strange try(tensoRT engine + trt.pth) , ocuurs : Cudnn Error configure:7 (CUDNN_STATUS_MAPPING_ERROR) and CUDNN_STATUS_EXECUTION_FAILED
maybe i will give up this strange combination. Thank you all the same John

[BarryKCL]

I encountered the same problem.when I use torch2trt and another TensorRT engine, an illegal memory access was encountered,but they can run correctly separately。

[xuatpham]

@BarryKCL excuse me, have you solved it? I'm facing the same problem when running multi tensorRT_engines in one program.
Thanks!

[fabiozappo]

Have anyone solved it? I'm facing the same problem when running a tensorRT_engine (yolov5) and another classification network (resnet_trt.pth) in the same program.

Thanks!

[xuatpham]

@fabiozappo yes, I was able to solve it bro. All you need to do is just a simple trick:
initialize a tensor before loading the model. Good luck!

[fabiozappo]

@fabiozappo yes, I was able to solve it bro. All you need to do is just a simple trick:
initialize a tensor before loading the model. Good luck!

it works!

[JosephKKim]

@xuatpham @fabiozappo
I am facing the same issue... can you please tell me how to initialize a tensor???

[xuatpham]

@xuatpham @fabiozappo I am facing the same issue... can you please tell me how to initialize a tensor???

before loading the yolo, just initialize a tensor
example: torch.randn(1).cuda()
after that, you are good to go.

good luck bro