qualcomm/YOLOv8-Detection · finetuning models

DazMashaly

Feb 27

is there a demo for finetuning the quantized model? Is it even possible?

shreyajn

Qualcomm org Feb 27

This comment has been hidden

gustlars

Qualcomm org Feb 27

Great question! We currently don't have a guide for this, but I'll outline what you can do here:

If you want to finetune the model, you can check out the original source from https://github.com/ultralytics/ultralytics and follow their guides to train yourself or finetune.

With the new checkpoint, you can instantiate the AI Hub Models class as:

checkpoint = "/path/to/new/checkpoint.pt"
from qai_hub_models.models.yolov8_det import Model 
model = Model.from_pretrained(checkpoint)

Now you can use the functions (compile, profile, inference) on that instance.

As for the quantized model, we currently do not offer Yolo-v8-Detection as quantized. Please stay tuned since we hope to expand the models that offer quantization.