Error during PyTorch model deployment in Triton Server: PytorchStreamReader failed locating file constants.pkl

[Manishthakur2503]

When attempting to load a PyTorch model (model.pth) in Triton Server (version 23.01-py3), the following error occurs:
E0718 11:47:30.586692 1 model_lifecycle.cc:597] failed to load 'CombinedFaceAudioModel' version 1: Internal: failed to load model 'CombinedFaceAudioModel': PytorchStreamReader failed locating file constants.pkl: file not found

Steps to Reproduce:

• Download the model file from https://github.com/rishiswethan/Video-Audio-Face-Emotion-Recognition/blob/main/models/audio_face_combined/audio_face_combined_model.pth
• Triton Version: 23.01-py3
  Docker Environment: Windows 10
  Model Repository Structure:

model_repository/
  └── CombinedFaceAudioModel/
      ├── 1/
      │   └── model.pth
      └── config.pbtxt

• My config.pbtxt is

name: "CombinedFaceAudioModel"
platform: "pytorch_libtorch"
max_batch_size: 8
default_model_filename: "model.pth"
input [
  {
    name: "audio_input"
    data_type: TYPE_FP32
    dims: [ 3, 224, 224 ]
  },
  {
    name: "image_input"
    data_type: TYPE_FP32
    dims: [ 3, 224, 224 ]
  }
]
output [
  {
    name: "output_label"
    data_type: TYPE_FP32
    dims: [ 5 ]
  }
]
instance_group [
  {
    count: 1
    kind: KIND_CPU
  }
]

• Run Triton Server with the following configuration: docker run --rm -p8000:8000 -p8001:8001 -p8002:8002 -v C:\Users\manis\Desktop\Triton_Emotion_Model\model_repository:/model_repository nvcr.io/nvidia/tritonserver:23.01-py3 tritonserver --model-repository=/model_repository

Expected Behavior:
The model CombinedFaceAudioModel should load successfully without any errors.

Actual Behavior:
The following error is reported: Internal: failed to load model 'CombinedFaceAudioModel': PytorchStreamReader failed locating file constants.pkl: file not found and if you want to check the full error message then please check the attacher ErrorMessage.txt file
ErrorMessage.txt

Additional Context:
It appears that Triton is expecting the file constants.pkl, which is referenced by the PyTorch model during loading. However, this file is not present in the model directory. Is there a specific requirement or configuration needed to handle constants.pkl in PyTorch models within Triton Server? Or is there anything wring with the configurations I am using?

Thanks
Manish Thakur