set_seed

torchtune.training.set_seed(seed: Optional[int] = None, debug_mode: Optional[Union[int, str]] = None) → int

Function that sets seed for pseudo-random number generators across commonly used libraries.

This seeds PyTorch, NumPy, and the python.random module. For distributed jobs, each local process sets its own seed, computed seed + rank. For more details, see https://pytorch.org/docs/stable/notes/randomness.html.

Parameters:

• seed (Optional[int]) – the integer value seed. If None, a random seed will be generated and set.

• debug_mode (Optional[Union[str, int]]) –

  Controls debug_mode settings for deterministic operations within PyTorch.

  • If None, don’t set any PyTorch global values.

  • If “default” or 0, don’t error or warn on nondeterministic operations and additionally enable PyTorch CuDNN benchmark.

  • If “warn” or 1, warn on nondeterministic operations and disable PyTorch CuDNN benchmark.

  • If “error” or 2, error on nondeterministic operations and disable PyTorch CuDNN benchmark.

  • For more details, see torch.set_deterministic_debug_mode() and https://pytorch.org/docs/stable/notes/randomness.html#avoiding-nondeterministic-algorithms.

Returns:

the current seed

Return type:

int

Raises:

ValueError – If the input seed value is outside the required range.