YoloV10 on Ryzen

[mht-sharma]

Description

I was trying to run the Yolov10 model on Ryzen. I quantized the ONNX model using the VitisAI ONNX Quantizer. I used the following configuration as per the VitisAI ONNX Quantizer documentation.

To my surprise, only 2% of the operators ran on the DPU. Upon checking the Vitisai_ep_report.json file, I found that the Conv was running on the CPU instead of the DPU.

Models

You can find the quantized models and other artifacts here: Yolov10s model

Note: I wanted to check only if the model runs on Ryzen and not worry about accuracy, hence the model is not quantized with accuracy in mind.

Expected Behavior

• The Conv operator should run on the DPU for better performance.

Any help or insights into resolving this issue would be greatly appreciated.

@uday610

[uday610]

Hi @mht-sharma , looking into this

[uday610]

Hi @mht-sharma , currently yolov10 model is not supported.

[uday610]

closing it