Pruning model

Hello everyone.
I have created and trained a MobileNet model (.pth).
The question is can I apply some pruning method to reduce the size of this and run it on the Jetson Nano? My aim is to have better performance with the new model.



You can try some PyTorch samples to do the pruning and run the output model on Jetson.

We do also have a library that can reduce the model complexity but it has its own model format.
If retraining is an option, you can also check the library:


Thanks for the reply @AastaLLL

I had already read in this forum about the existence of TAO Toolkit. As far as I understand, does this framework build a model in .tlt format that it pruning? Isn’t there a tool that converts the .tlt (or .etlt) format to onnx? Thank you.

You can try our tool for Neural network compression and acceleration for Jetson.Nano

I believe you can get up to 3-4 times acceleration.
Do you want to try it for free for your model? Please email me