After little success with using the SDK on an large GPU instance (Ubuntu EC2 instance) to convert a test model (yolo8n small)), I wanted to understand if anyone has actually created a pipeline for such.
On the ec2 instance, we convert from .pt to onnx just fine, but sadly the 2nd .hef conversion/optimisation hangs and fails (usually noticing gpu not found errors). On checking with the docker host and image, we can see the GUP librarys are available.
Ideally, for us to consider H8’s we will need such a pipeline thats easy to use and easy to deploy, I can’t say I have not been going round in circles for the last 24 hours trying to just convert!
Frustrations aside, if someone would like to point me in the current right direction, I will happily continue to build such a pipeline.