We’re running a YOLOv5s model converted to a TensorRT engine in a Triton Server. Sending the same image to the Triton server for inference returns different results each time. Is this normal or is there something we could do to make it deterministic? We are using a Jetson Nano 2GB and built the TensorRT engine on the Jetson Nano itself.
TensorRT Version: 7.1
Operating System + Version: Jetson Nano 2GB / Tegra
Run the inference with the same image using the same TensorRT engine running on a Triton Server multiple times and the results are different each time.