FeatureCloud Pruning

Model Compression with Pruning

The FC Pruning package offers a streamlined approach to model compression using advanced pruning techniques. With support for federated learning frameworks and integration with the Torch-Pruning library, this package enables efficient distributed training, suitable for a wide range of machine learning tasks.

1. Train Local Model

2. Configure Quantization:

   • Set up initial quantization settings.

       self.configure_pruning(pruning_ratio, model, reference_model, imp,ex_input,
                              ignored_layers)        

3. Send Data to Coordinator and perform Pruning:

   • Send the prepared model data to the coordinator.

    self.send_data_to_coordinator(model, use_pruning=True, use_smpc=False, use_dp=False)