`torchscan.modules`¶

The modules subpackage contains tools for inspecting modules. See Understanding results for counting conventions and Model and input support for the capability matrix.

FLOPs¶

Related to the number of floating-point operations performed during model inference.

module_flops(module: Module | Callable[..., Tensor], inputs: Tuple[Any, ...], out: Any) -> int

Estimate the number of floating point operations performed by the module

Parameters:

Name	Type	Description	Default
`module`	`Module \| Callable[..., Tensor]`	PyTorch module	required
`inputs`	`Tuple[Any, ...]`	input to the module	required
`out`	`Any`	output of the module	required

Returns: number of FLOPs

Related to the number of multiply-accumulate operations performed during model inference.

module_macs(module: Module, inp: Tensor, out: Tensor) -> int

Estimate the number of multiply-accumulation operations performed by the module

Parameters:

Name	Type	Description	Default
`module`	`Module`	PyTorch module	required
`inp`	`Tensor`	input to the module	required
`out`	`Tensor`	output of the module	required

Returns: int: number of MACs

Related to the number of direct memory accesses during model inference.

module_dmas(module: Module, inp: Tensor, out: Tensor) -> int

Estimate the number of direct memory accesses by the module. The implementation overhead is neglected.

Parameters:

Name	Type	Description	Default
`module`	`Module`	PyTorch module	required
`inp`	`Tensor`	input to the module	required
`out`	`Tensor`	output of the module	required

Returns: int: number of DMAs

Related to the effective receptive field of a layer.

module_rf(module: Module, inp: Tensor, out: Tensor) -> Tuple[float, float, float]

Estimate the spatial receptive field of the module

Parameters:

Name	Type	Description	Default
`module`	`Module`	PyTorch module	required
`inp`	`Tensor`	input to the module	required
`out`	`Tensor`	output of the module	required

Returns: receptive field effective stride effective padding