torch.RegisterBackwardOptions

export interface RegisterBackwardOptions {
  /**
   * Optional dtype restriction.
   * If specified, this autograd only applies to the given dtype.
   * Use '*' for wildcard to match any dtype.
   */
  readonly dtype?: AutogradDType;

  /**
   * Device restriction.
   * Use '*' for wildcard to match any device.
   * This is useful for composed backwards that use tensor operations
   * which auto-dispatch to the correct kernel.
   */
  readonly device: AutogradDevice;
}

readonlydtype(AutogradDType)optional: – Optional dtype restriction. If specified, this autograd only applies to the given dtype. Use '*' for wildcard to match any dtype.
readonlydevice(AutogradDevice): – Device restriction. Use '*' for wildcard to match any device. This is useful for composed backwards that use tensor operations which auto-dispatch to the correct kernel.

Options for register_backward.

torch.RegisterBackwardOptions

export interface RegisterBackwardOptions {
  /**
   * Optional dtype restriction.
   * If specified, this autograd only applies to the given dtype.
   * Use '*' for wildcard to match any dtype.
   */
  readonly dtype?: AutogradDType;

  /**
   * Device restriction.
   * Use '*' for wildcard to match any device.
   * This is useful for composed backwards that use tensor operations
   * which auto-dispatch to the correct kernel.
   */
  readonly device: AutogradDevice;
}

readonlydtype(AutogradDType)optional: – Optional dtype restriction. If specified, this autograd only applies to the given dtype. Use '*' for wildcard to match any dtype.
readonlydevice(AutogradDevice): – Device restriction. Use '*' for wildcard to match any device. This is useful for composed backwards that use tensor operations which auto-dispatch to the correct kernel.

Options for register_backward.