Skip To Main Content
Support Knowledge Base

Does Filter Pruning Compression Algorithm in Neural Network Compression Framework (NNCF) Reduce Model Size?

Content Type: Product Information & Documentation   |   Article ID: 000092425   |   Last Reviewed: 03/01/2023

Description

Resolution

  • NNCF filter pruning algorithm only puts zeros inside convolutional and linear layer parameters of a model and doesn't reduce the size of model.
  • To remove channels/rows from the model, an additional nGraph pruning transformation is required. To do this, add --transform=Pruning to MO conversion command during model conversion to Intermediate Representation (IR) format.

Related Products

This article applies to 1 products.