Load Pytorch Model Weights in Python Backend

[maxi-w]

Hey, I have a model deployed as python backend and was wondering if and how I can suply additional files like a model.pth with that python backend?

One way to solve this would be to deploy the model as e.g. ONNX but that comes with some extra work of seperating out the pre-/post-processing and combining the steps.

[hasnainkothawala]

Not sure if I understand you correctly but this how i did it ,

I wrote this in my model.py

def initialize(self, args):
	try:
		self.image_model_path = Path("/mnt/models/detectron-service/1/resources/v1.0/model_final.pth")
		self.image_confidence_threshold = 0.8
		self.detectron_model_weights_file = ("COCO-Detection/faster_rcnn_R_50_FPN_3x.yaml")
		self.detectron2_config = None

And added the model.pth file in the same location :