neural_compressor.compression.pruner.wanda.prune

Module Contents

Functions

prune_wanda(model, dataloader, sparsity_ratio[, ...])

Prune the model using wanda

neural_compressor.compression.pruner.wanda.prune.prune_wanda(model, dataloader, sparsity_ratio, prune_n=0, prune_m=0, nsamples=128, use_variant=False, device=None, low_mem_usage=None, dsnot=False)[source]

Prune the model using wanda Sij = |Wij| · ||Xj||2.

See the original paper: https://arxiv.org/pdf/2306.11695.pdf