continue
if not hasattr(container, "gradient_checkpointing"):
continue
for child_name, child in container.named_children():
blocks = list(child.named_children())
if not blocks:
continue
child_fqn = f"{container_fqn}.{child_name}" if container_fqn else child_name
for block_name, block in blocks:
if id(block) in wrapped_module_ids:
continue
wrapped_module_ids.add(id(block))
wrapped_module = _get_checkpoint_wrapped_module(block)