YoloV4 Jupyter Notebook trained models with Triton errors

Please provide the following information when requesting support.

• Hardware (T4/V100/Xavier/Nano/etc) TAO over Ubuntu and triton on Docker image (triton 23.07_py3).
• Network Type (Detectnet_v2/Faster_rcnn/Yolo_v4/LPRnet/Mask_rcnn/Classification/etc) Yolo_v4
• TLT Version (Please run “tlt info --verbose” and share “docker_tag” here) TAO 5.0
• Training spec file(If have, please share here) Jupyter Notebook example
• How to reproduce the issue ? (This is for errors. Please share the command line and the detailed log here.)


I trained the example model with the Kitty dataset, pruned it and retrained, then I followed some more steps because I wanted to benchmark different approaches, so at the end I have this:

  • yolov4_resnet18_epoch_080.onnx
  • trt.engine
  • trt.engine.fp16

I am trying to load this models with an existing Triton inference server that has some other models running, but it gives different errors.

Are these models supposed to be loaded directly in the Triton server? The config.pbtxt for each try:

name: "test1"
platform: "tensorrt_plan"
max_batch_size: 1
default_model_filename: "trt.engine"

The model is in 1 folder and the file is trt.engine. The error says:

UNAVAILABLE: Internal: unable to load plan file to auto complete config: /models/test1/1/trt.engine

name: "test2"
platform: "onnxruntime_onnx"
max_batch_size: 1

The model is in 1 folder and the file is model.onnx. The error says:

UNAVAILABLE: Internal: onnx runtime error 10: Load model from /models/test1/1/model.onnx failed:This is an invalid model. In Node, (“BatchedNMS_N”, BatchedNMSDynamic_TRT, “”, -1) : (“box”: tensor(float),“cls”: tensor(float),) → (“BatchedNMS”: tensor(int32),“BatchedNMS_1”: tensor(float),“BatchedNMS_2”: tensor(float),“BatchedNMS_3”: tensor(float),) , Error No Op registered for BatchedNMSDynamic_TRT with domain_version of 12

What am I doing wrong?

Can you delete test2 folder and try to run again? Can test1’s engine be loaded?

Also, please try to follow GitHub - NVIDIA-AI-IOT/tao-toolkit-triton-apps: Sample app code for deploying TAO Toolkit trained models to Triton as getting started.


The tests were conducted one by one, so no, test1 is also not being loaded.

Please change the name to model.plan.


File renamed to model.plan, this is the error now:

UNAVAILABLE: Internal: unable to load plan file to auto complete config: /models/test1/1/model.plan

I trained a Detectnet_v2 example model and the .onnx result model loads perfectly:

resnet18_detector | 1       | READY

I’m on my way to following the guide you mentioned before, thank you for the kind support.

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

Yes, please try to run the official trition-tao-app firstly as getting started.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.