torch.utils.module_tracker¶
This utility can be used to track the current position inside an torch.nn.Module
hierarchy.
It can be used within other tracking tools to be able to easily associate measured quantities to user-friendly names. This is used in particular in the FlopCounterMode today.
- class torch.utils.module_tracker.ModuleTracker[source][source]¶
ModuleTracker
is a context manager that tracks the nn.Module hierarchy during execution so that other system can query which Module is currently being executed (or its backward is being executed).You can access the
parents
attribute on this context manager to get the set of all the Modules currently being executed via their fqn (fully qualified name, also used as the key within the state_dict). You can access theis_bw
attribute to know if you are currently running in backward or not.Note that
parents
is never empty and always contains the “Global” key. Theis_bw
flag will remainTrue
after the forward until another Module is executed. If you need it to be more accurate, please submit an issue requesting this. Adding a map from fqn to the module instance is possible but not done yet, please submit an issue requesting this if you need it.Example usage
mod = torch.nn.Linear(2, 2) with ModuleTracker() as tracker: # Access anything during the forward pass def my_linear(m1, m2, bias): print(f"Current modules: {tracker.parents}") return torch.mm(m1, m2.t()) + bias torch.nn.functional.linear = my_linear mod(torch.rand(2, 2))