torch.js has not been released yet.

torch.use_deterministic_algorithms

function use_deterministic_algorithms(mode: boolean, options?: DeterministicOptions): voidfunction use_deterministic_algorithms(mode: boolean, warn_only: boolean, options?: DeterministicOptions): void

Enables or disables deterministic algorithms for reproducibility.

When enabled, all operations use deterministic algorithms instead of potentially faster non-deterministic ones. This ensures exact reproducibility across runs but may reduce performance.

This is a no-op in torch.js. Use manual_seed to set seeds for reproducible random number generation. WebGPU operations are deterministic by design.

Parameters

modeboolean: If true, enable deterministic mode; if false, disable it
optionsDeterministicOptionsoptional

Returns

void

Examples

// Enable deterministic mode for reproducible results
torch.use_deterministic_algorithms(true);

// This would warn but not error if a non-deterministic op is used
torch.use_deterministic_algorithms(true, { warn_only: true });

// Disable deterministic mode for better performance
torch.use_deterministic_algorithms(false);

See Also

PyTorch torch.use_deterministic_algorithms()
manual_seed - Set random seed for reproducibility
get_rng_state - Capture and restore RNG state

ValidateBatchedSquareMatrix