Summary: 39 instances, 33 unique

Text	Count
# TODO: Set these with configurable parameters	1
# FIXME: pytorch currently does not register `torch.cat` and	1
# TODO: Handle batch dimension here	1
# TODO: Implement conv2d gradient under following condition:	1
# TODO: Rename this to __copy__()?	3
# TODO: ONNX specification says the permutation should be	1
# TODO: fix replacement in global `torch` module - perhaps use __torch_function__	1
# TODO: Implement conv1d gradient under following condition:	1
# TODO: Add support for loading from correct device (kwarg: map_location=device)	1
# TODO: Async / parallelize this	1
# TODO: Automatically register all these functions in CrypTensor?	1
# TODO: Eliminate dependency on torch internal function by implementing in util	3
# TODO: Encrypt modules before returning them	1
# TODO: Add test coverage for this code path (next 4 lines)	1
# TODO: Add validation_mode / validate_correctness	1
# TODO (brianknott): Check whether this RNG contains the full range we want.	1
# TODO: Although paramiko.SSHClient.exec_command() can accept	1
# TODO: make this optional?	1
# TODO: Eliminate copy-pasta by implementing _Conv parent class	1
# TODO: Find a better solution for padding with max_pooling	1
# TODO: Implement DP properly to make correct DP guarantees	1
# TODO: Remove explicit broadcasts to allow smaller beaver triples	1
# TODO: Support addition with different encoder scales	1
# TODO: parallelize / async this	1
# TODO: Add support for saving to correct device (kwarg: map_location=device)	1
# TODO: Rename this to __deepcopy__()?	2
# TODO: Incorporate eta_xr	1
# TODO: Implement custom DP mechanism (split noise / magnitude)	1
return attr.s  # TODO: Sanitize string.	1
# TODO: Add check for whether ceil_mode would change size of output and allow ceil_mode when it wouldn't	1
# TODO: Compute size without executing computation	1
# TODO: Deal with any overwrite issues	2
# TODO: Replace this	1