A noise sampler backed by a torchsde.BrownianTree. Args: x (Tensor): The tensor whose shape, device and dtype to use to generate

Model training loop phase. Each model's training loop iteration could be separated into (at least) two phases: training and validation.

Transformer block for image-like data. First, project the input (aka embedding) and reshape to b, t, d. Then apply standard transform

DPM-Solver. See https://arxiv.org/abs/2206.00927.

Untuned linear warmup schedule for Adam. This warmup scheme is described in `On the adequacy of untuned warmup for adaptive optimization

A wrapper around torchsde.BrownianTree that enables batches of entropy.

Groups together several callbacks and delegates training loop notifications to the encapsulated objects.

* \brief Stores a set of parameters controlling the way matrices are printed * * List of available parameters: * - \b precision number of digit

Calculates the natural cubic spline approximation to the batch of controls given. Also calculates its derivative. Example: t = torch.lins

A PID controller for ODE adaptive step size control.

* \brief Specifies the modifications on the two input arguments and the output argument. */

* \brief Flags that specify how the data in a MatrixBase-expression can be accessed. */

* \brief Specifies the assignment mode in \c Assignment::assign() . * This is the difference between regular assignment (operator==, \c AssignmentMode

* \brief Automatic algorithm selection. * Chooses the algorithm during runtime based on the matrix sizes. */

* \brief The axis over which reductions are performed. */

Base class for all warmup schedules Arguments: optimizer (Optimizer): an instance of a subclass of Optimizer warmup_params (list)