Optimize a custom model manually

General
1

Hi, I wonder there is a way that optimizing a custom model manually not automatically when ‘runner.optimize(calib_data)’. It seems that when the number of calibration data is over than 1024, then hailo compiler optimize the network automatically using quantization aware training, knowledge distillation, or 4-bit weight quantization so on.

I want to test the level of the optimization for my research. but I could not find a way controlling the level of the optimization.

It would be good for any answers for me.

Thanks :slight_smile:

2

Have you checked out the documentation for model optimisation? There are a bunch of additional configurations you can add to the model script (load_model_script) which will alter the optimization approach, for example model_optimization_flavor()

link to the docs here

1 Like
3

Thank you!

I did not know that the document exists.