h @SSKrSSKJr SSKJrJr SSKrSSKJr SSK J r SSK J r SSK JrJrJr SSKJrJr SS KJr S r\"S S \ 55r\r\"\5S \R2R4S\4Sj5r"SS\ 5r\r\"\5S \R2R4S\S\R2R44Sj5rSS\R2R4S\ S\\\R2R4\/\ 4S\R2R44Sjjr!g)N) dataclass)CallableOptional)to_sparse_semi_structured) AOBaseConfig)WeightNormSparsifier) _is_linear_linear_extra_repr)_replace_with_custom_fn_if_matches_filter)_QUANTIZE_CONFIG_HANDLER register_quantize_module_handler)BlockSparseTensorc (URS[5n/nUR5H*upEU"XT5(dMURSUS305 M, [ SSSS9nUR X5 UR 5 UR5 g) zG This function simulates 2:4 sparsity on all linear layers in a model. filter_fn tensor_fqnz.weightg?))sparsity_levelsparse_block_shapezeros_per_blockN)popr named_modulesappendrpreparestep squash_mask)modelkwargsr sparse_confignamemod sparsifiers U/home/james-whalen/.local/lib/python3.13/site-packages/torchao/sparsity/sparse_api.pyapply_fake_sparsityr%s ; 3IM((*  S    ,4&0@!A B+&vqJu,OOc*\rSrSr%Sr\\S'SrSrg)BlockSparseWeightConfig1@ blocksizecB[RRS5 g)Nz(torchao.sparsity.BlockSparseWeightConfigtorch_C_log_api_usage_onceselfs r$ __post_init__%BlockSparseWeightConfig.__post_init__5s $$%OPr&N) __name__ __module__ __qualname____firstlineno__r+int__annotations__r3__static_attributes__r5r&r$r(r(1sIsQr&r(moduleconfigcURn[R"URU5n[R R USS9Ul[R"[U5Ul U$NF) requires_grad) r+r from_denseweightr.nn Parametertypes MethodTyper extra_repr)r=r>r+ new_weights r$_block_sparse_weight_transformrJ=s[   I"--fmmYGJHH&&z&GFM(();VDF Mr&c\rSrSrSrSrSrg)SemiSparseWeightConfigIz] Configuration for converting the weight of linear modules to semi-structured (2:4) sparsity cB[RRS5 g)Nz'torchao.sparsity.SemiSparseWeightConfigr-r1s r$r3$SemiSparseWeightConfig.__post_init__Ns $$%NOr&r5N)r6r7r8r9__doc__r3r<r5r&r$rLrLIs Pr&rLreturnc[UR5n[RR USS9Ul[ R "[U5UlU$r@) rrCr.rDrErFrGr rH)r=r>rIs r$_semi_sparse_weight_transformrSVsH +6==9JHH&&z&GFM(();VDF Mr&rrc[RRS5 [[ U5n[ UUUc[ OUU4S9 g)aConvert the weight of linear modules in the model with `apply_tensor_subclass`. This function is essentially the same as quantize, put for sparsity subclasses. Currently, we support three options for sparsity: - semi-structured (2:4) sparsity with `semi_sparse_weight` - int8 dynamic quantization + 2:4 sparsity with `layout=SemiSparseLayout` - int4 weight-only quantization + 2:4 sparsity with `layout=SparseMarlinLayout` Args: model (torch.nn.Module): input model config (AOBaseConfig): a workflow configuration object filter_fn (Optional[Callable[[torch.nn.Module, str], bool]]): function that takes a nn.Module instance and fully qualified name of the module, returns True if we want to apply the specified workflow to this module. **Example:** :: import torch import torch.nn as nn from torchao.sparsity import sparsify_ def filter_fn(module: nn.Module, fqn: str) -> bool: return isinstance(module, nn.Linear) m = nn.Sequential(nn.Linear(32, 1024), nn.Linear(1024, 32)) # for 2:4 sparsity from torchao.sparse_api import semi_sparse_weight m = sparsify_(m, semi_sparse_weight(), filter_fn) # for int8 dynamic quantization + 2:4 sparsity from torchao.dtypes import SemiSparseLayout m = quantize_(m, Int8DynamicActivationInt8WeightConfig(layout=SemiSparseLayout), filter_fn) ztorchao.sparsity.sparsify_N) extra_args)r.r/r0r typer r )rr>rhandlers r$ sparsify_rXasCJ HH  !=>&tF|4G- ' Y9 r&)N)"rF dataclassesrtypingrrr. torch.sparsertorchao.core.configrrfsE  !% 2, ;& QlQ Q.""9: HHOO #;P\P,""89 HHOO " XX__:CG, 88??, ,%((//3!7!=>?, XX__ ,r&