NefAI focused on stabilizing memory format handling in the pytorch/executorch repository, addressing a bug that caused runtime errors when cloning or copying tensors without specifying a memory format. Using Python and leveraging deep learning expertise with PyTorch, NefAI implemented a targeted fix in MemoryFormatOpsPass to ensure the input tensor’s dimension order is preserved by default and correctly derived when preserve_format is active. The solution included updating memory format logic, expanding unit tests to cover new behaviors, and validating the changes with a standalone reproduction script. This work provided a minimal, robust fix that improved reliability without altering existing tests.