torchscan.utils#

torchscan.utils.format_info(module_info: Dict[str, Any], wrap_mode: str = 'mid', receptive_field: bool = False, effective_rf_stats: bool = False) → str[source]#

Print module summary for an expected input tensor shape

Parameters:

  • module_info – dictionary output of crawl_module

  • wrap_mode – wrapping mode

  • receptive_field – whether to display receptive field

  • effective_rf_stats – if receptive_field is True, displays effective stride and padding

Returns:

formatted information

torchscan.utils.aggregate_info(info: Dict[str, Any], max_depth: int) → Dict[str, Any][source]#

Aggregate module information to a maximum depth

Parameters:

  • info – dictionary output of crawl_module

  • max_depth – depth at which parent node aggregates children information

Returns:

edited dictionary information