torchao.sparsity

Created On: Dec 17, 2025 | Last Updated On: Dec 17, 2025

sparsify_	Convert the weight of linear modules in the model with apply_tensor_subclass.
semi_sparse_weight	alias of SemiSparseWeightConfig
apply_fake_sparsity	This function simulates 2:4 sparsity on all linear layers in a model.
WandaSparsifier	Wanda sparsifier
PerChannelNormObserver	A custom observer that computes the L2 norm of each channel and stores it in a buffer.