Called each time the simulation is stepped. Can usually be left.

Process loop executed by one worker thread or process. Parameters ---------- pipe_fd : int File descriptor opened by the parent (w

r""" Apply perturbed attention guidance to the noise prediction. Args: noise_pred (torch.Tensor): The noise prediction te

r""" Set the attention processor for the PAG layers.

Run batched inference over a batch of trajectories, and generate predictions either as a rollout or at specific time checkpoints.

r""" Set the the self-attention layers to apply PAG. Raise ValueError if the input is invalid. Args: pag_applied_layers (

r""" Performs Pipeline dtype and/or device conversion. A torch.dtype and torch.device are inferred from the arguments of `self.to(*arg

This does the final conversion step: take locally converted weights and apply a global renaming to them. It splits attention layers, and take

r""" Returns: `torch.device`: The torch device on which the pipeline is located.

Update the *latest row* of ``self.state_traj`` at column ``key``. mode : {'replace', 'extend'}, default 'replace' - repla

Downsamples the provided mask tensor to match the expected dimensions for scaled dot-product attention. If the aspect ratio of the ma

r""" Offloads all models to CPU using 🤗 Accelerate, significantly reducing memory usage. When called, the state dicts of all `torch.nn

Fetches all the values from the specified state column, and filter out empty values.

Create a single “user”-role OpenAI message with **one prompt** (may be empty) followed by N images.

Splits 'input_dict' into smaller batches and puts them into input_queue. Assumes all lists in 'input_dict' have the same length and

Optionally removes offloading in case the pipeline has been already sequentially offloaded to CPU. Args: _pipeline (`Dif

Applies the FreeU mechanism as introduced in https: //arxiv.org/abs/2309.11497. Adapted from the official code repository: https://github.com/Chen

Returns the current state observation.

Precompute the frequency tensor for complex exponentials (cis) with given dimensions. This function calculates a frequency tensor with compl

r""" Finds a rectangular region that contains all masked ares in an image, and expands region to match the aspect ratio of the origina

r""" Returns the height and width of the image, downscaled to the next integer multiple of `vae_scale_factor`. Args: imag

Compares the current Accelerate version to a given reference with an operation. Args: operation (`str`): A string repres

Reads a checkpoint file, returning properly formatted errors if they arise.

Forward process in flow-matching Args: sample (`torch.FloatTensor`): The input sample. times

Cleans up the history for the specified key by set all the values to pd none in the column.

Returns the action trajectory list.

Determine the metric file path based on whether we have a datum (episode-level file) or a parallel_ith scenario (global worker-level

Rotates the input tensor by performing a cyclic horizontal shift along its width dimension. - 0 or 360°: no shift. - ±22.5°:

returns a position within the specified boundary. Calculated using a weighted percentage. set ignore_collisions if you ar

Number of variations for the task. Can be determined dynamically. :return: Number of variations for this task.

Construct a layernorm module in the T5 style. No bias and no subtraction of mean.