Print message to the console. Include the parallelism rank information when verbose is set to True.

Sets the discrete timesteps used for the diffusion chain (to be run before inference). Args: num_inference_steps (`int`):

Update the loss record with new values. Args: loss: The loss value to add nan_mask_gradient_ratio: NaN ratio in gradi

Find the shape of input tensor for post-CP split, taking into account both temporal and spatial split, as well as patching values. The split

Called when checkpoint saving is fully finished, and succeeded. Not called if checkpoint failed. For synchronous checkpoint, it is ca

Sets the video conditioning frames for video-to-video generation. This method creates a conditioning mask for the input video frames

Start time measurements. Args: cuda_device (torch.device | None): CUDA device. Will use default CUDA device if not indic

Convert numpy array to tensor efficiently. Args: frames: numpy array of shape [T, H, W, C] in uint8 format Returns:

If the file checkpoint_path does not exist, raise an error. Args: checkpoint_path (str): full path to the checkpoint.

Track the file name of the latest saved checkpoint. Args: checkpoint_file (str): file name of the latest saved checkpoint.

Read bytes from a given ``filepath`` with 'rb' mode in range [offset, offset + size). Args: filepath (str or Path): Path to read

r"""Concatenate all file paths. Join one or more filepath components intelligently. The return value is the concatenation of filepath

Called when checkpoint loading is about to start, but after on_save_checkpoint_start(). FIXME - why do we need this callback, can't w

:param config: the instance of class `Config` (usually from `make_config`) :param overrides: list of overrides for config :return: the co

Run the safety check on the input.

Save a tensor as an image or video file based on shape Args: sample_C_T_H_W_in01 (Tensor): Input tensor with shape (C, T, H, W)

Performs a single optimization step. Arguments: closure (callable, optional): A closure that reevaluates the model

Performs a single optimization step. Arguments: closure (callable, optional): A closure that reevaluates the model

Get the file name of the latest saved checkpoint. If it doesn't exist, return None. Returns: checkpoint_file (str | None): file n

Broadcast a tensor and check if the shape is the same across CP/TP ranks. If not, create a new tensor matching rank 0 and broadcast it. Args:

Generates frame indices for video sampling with FPS control. This function determines valid stride lengths for sampling frames from a video,

Get world size. How many GPUs are available in this job. Returns: world_size (int): The total number of GPUs available in this job.

Given the camera intrinsics, calibrate input 3D points from image (pixel) frame to camera frame. Args: points (torch.Tensor/np.nd

r"""Function for obtaining size of the augmentation. When dealing with multi-aspect ratio dataloaders, we need to find the augmentation size f

Called when checkpoint saving is about to start.

Predict the sample from the previous timestep by reversing the SDE. This function propagates the sample with the multistep UniPC.

Checks whether the input tensor is a valid camera pose. Args: cam_pose (torch.Tensor [...,3,4]): Input camera pose in world2cam [

Checks whether the input tensor is a valid quaternion. Args: q (torch.Tensor [...,4]): Input quaternion(s) in (x, y, z, w) order.

Execute an operation with exponential backoff retry logic. Args: operation_func: Function to execute *args:

Broadcast the item from the minimum rank in the specified group(s).

Read bytes from a given ``filepath`` with 'rb' mode in range [offset, offset + size). Args: filepath (str or Path): Path to read

Convert (fx, fy, cx, cy) parameters to camera intrinsic matrix/matrices. Args: params (torch.Tensor/np.ndarray [...,4]): Intrinsi

We hanlde two types of data_batch. One comes from a joint_dataloader where "dataset_name" can be used to differenciate image_batch and video_batch.

r"""Concatenate all file paths. Join one or more filepath components intelligently. The return value is the concatenation of filepath

Scan a directory to find the interested directories or files in arbitrary order. Note: :meth:`list_dir_or_file` returns t

Loads data from S3 with optional caching. This function first attempts to load the data from a local cache file. If the cache file doesn't ex

Helper function for loading object from storage. Args: key (str): The key of the object. type (str): Specified for so

r"""Function for obtaining the image size from the data dict. Args: data_dict (dict): Input data dict input_keys (list): List of

Called when the synchronous part of checkpointing is finished, this function can be used along with on_save_checkpoint_start() to mea