prune

Modules

modelopt.torch.prune.config

Default configurations for prune modes.

modelopt.torch.prune.fastnas

Module implementing fasnas pruning algorithm for search.

modelopt.torch.prune.gradnas

Module implementing gradnas pruning algorithm for search.

modelopt.torch.prune.mcore_gpt_minitron

Module implementing mcore_gpt_minitron pruning algorithm for NVIDIA Megatron-Core / NeMo models.

modelopt.torch.prune.mode

Module implementing and describing modes that can be used during the NAS convert process.

modelopt.torch.prune.plugins

Handles prune plugins for third-party modules.

modelopt.torch.prune.pruning

High-level API to automatically prune and optimize your model with various algorithms.

A simplified API for modelopt.torch.nas for pruning algorithms.

This module provides a simplified API for pruning that is based on the NAS infrastructure but simplifies the overall workflow to accommodate for the simpler nature of pruning algorithms.