Repeated module calls belong to TDHook

A TDHook Target selects a zero-based occurrence directly. XDRL does not install internal hooks or maintain call counters.

import torch
from tensordict import TensorDict
from tensordict.nn import TensorDictModule
from tdhook.latent import ActivationCaching
from tdhook.targets import Target
from tdhook.workflow import Workflow
from xdrl import interpret


class ReusedLayer(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.shared = torch.nn.Identity()

    def forward(self, value):
        return self.shared(value + 1) + self.shared(value + 2)


policy = TensorDictModule(ReusedLayer(), in_keys=["observation"], out_keys=["action"])
component = interpret(policy)
target = Target("module.shared", "activation", -1, (0,), occurrences=(1,))
workflow = Workflow(ActivationCaching(target, cache_key=("activations", "selected")))
data = TensorDict({"observation": torch.tensor([[1.0, 2.0]])}, batch_size=[1])
result = component.run(workflow, data)
torch.testing.assert_close(result.data["activations", "selected", "module.shared"], torch.tensor([[3.0]]))