1torch.utils.deterministic 2========================= 3.. py:module:: torch.utils.deterministic 4.. currentmodule:: torch.utils.deterministic 5 6.. attribute:: fill_uninitialized_memory 7 8 A :class:`bool` that, if True, causes uninitialized memory to be filled with 9 a known value when :meth:`torch.use_deterministic_algorithms()` is set to 10 ``True``. Floating point and complex values are set to NaN, and integer 11 values are set to the maximum value. 12 13 Default: ``True`` 14 15 Filling uninitialized memory is detrimental to performance. So if your 16 program is valid and does not use uninitialized memory as the input to an 17 operation, then this setting can be turned off for better performance and 18 still be deterministic. 19 20 The following operations will fill uninitialized memory when this setting is 21 turned on: 22 23 * :func:`torch.Tensor.resize_` when called with a tensor that is not 24 quantized 25 * :func:`torch.empty` 26 * :func:`torch.empty_strided` 27 * :func:`torch.empty_permuted` 28 * :func:`torch.empty_like`