WandaSparsifier

class torchao.sparsity.WandaSparsifier(sparsity_level: float = 0.5, semi_structured_block_size: Optional[int] = None)

Wanda 稀疏化器

Wanda (Pruning by Weights and activations)，发表于 https://arxiv.org/abs/2306.11695，是一种感知激活的剪枝方法。该稀疏化器根据输入激活范数与权重大小的乘积来移除权重。

此稀疏化器由三个变量控制：1. sparsity_level 定义了要归零的稀疏块的数量；

参数:

• sparsity_level – 目标的稀疏度级别；

• model – 要进行稀疏化的模型；

prepare(model: Module, config: List[Dict]) → None

准备模型，通过添加参数化。

注意

The model is modified inplace. If you need to preserve the original
model, use copy.deepcopy.

squash_mask(params_to_keep: Optional[Tuple[str, ...]] = None, params_to_keep_per_layer: Optional[Dict[str, Tuple[str, ...]]] = None, *args, **kwargs)

将稀疏掩码压缩到相应的张量中。

如果设置了 params_to_keep 或 params_to_keep_per_layer，则模块将附加一个 sparse_params 字典。

参数:

• params_to_keep – 要在模块中保存的键的列表，或表示将保存稀疏参数的模块和键的字典

• params_to_keep_per_layer – 用于指定要为特定层保存的参数的字典。字典中的键应为模块的 fqn，而值应为字符串列表，包含要在 sparse_params 中保存的变量名称

示例

>>> # xdoctest: +SKIP("locals are undefined")
>>> # Don't save any sparse params
>>> sparsifier.squash_mask()
>>> hasattr(model.submodule1, "sparse_params")
False

>>> # Keep sparse params per layer
>>> sparsifier.squash_mask(
...     params_to_keep_per_layer={
...         "submodule1.linear1": ("foo", "bar"),
...         "submodule2.linear42": ("baz",),
...     }
... )
>>> print(model.submodule1.linear1.sparse_params)
{'foo': 42, 'bar': 24}
>>> print(model.submodule2.linear42.sparse_params)
{'baz': 0.1}

>>> # Keep sparse params for all layers
>>> sparsifier.squash_mask(params_to_keep=("foo", "bar"))
>>> print(model.submodule1.linear1.sparse_params)
{'foo': 42, 'bar': 24}
>>> print(model.submodule2.linear42.sparse_params)
{'foo': 42, 'bar': 24}

>>> # Keep some sparse params for all layers, and specific ones for
>>> # some other layers
>>> sparsifier.squash_mask(
...     params_to_keep=("foo", "bar"),
...     params_to_keep_per_layer={"submodule2.linear42": ("baz",)},
... )
>>> print(model.submodule1.linear1.sparse_params)
{'foo': 42, 'bar': 24}
>>> print(model.submodule2.linear42.sparse_params)
{'foo': 42, 'bar': 24, 'baz': 0.1}

update_mask(module: Module, tensor_name: str, sparsity_level: float, **kwargs) → None

WandaSparsifier 的剪枝函数

首先在 act_per_input 变量中检索激活统计信息。然后计算 Wanda 剪枝指标。通过比较整个当前层的该指标来剪枝权重矩阵。