Synchronized timer to count the inference time of `nn.Module.forward`. Supports both context manager and decorator usage. Example a

We rename some layer names to align with flash attention

Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).

Fake file-like stream object that redirects writes to a logger instance.

The final layer of HunYuanDiT.

Output class for the scheduler's `step` function output. Args: prev_sample (`torch.FloatTensor` of shape `(batch_size, num_channels,

The sin/cosine positional embedding. Given an input tensor `x` of shape [n_batch, ..., c_dim], it converts each feature dimension of `x[..., i]` i

A DiT block with parallel linear layers as described in https://arxiv.org/abs/2302.05442 and adapted modulation interface.

Embeds scalar timesteps into vector representations.

The trick function of adding auxiliary (aux) loss, which includes the gradient of the aux loss during backpropagation.

NOTE: this is very similar to diffusers.FlowMatchEulerDiscreteScheduler. Except our timesteps are reversed Euler scheduler. This model

view order: front, front clockwise 90, back, front clockwise 270