Ensure last call is logged

create video dataset from video links

Write stats to disk

Initialize the KeyPassThroughDecoder. :param *args: Positional arguments to be passed to the base Decoder class. :param pass

Cache $\cos$ and $\sin$ values

Returns a float_list from a float / double.

gets random reference

gets reference frame

gets resize size

https://github.com/pytorch/pytorch/blob/db2a237763eb8693a20788be94f8c192e762baa8/torch/nn/functional.py#L4726

Returns an int64_list from a bool / enum / int / uint.

Returns an list of int64_list, float_list, bytes_list.

Adapted from Mesh Tensorflow: https://github.com/tensorflow/mesh/blob/0cb87fe07da627bf0b7e60475d59f95ed6b5be3d/mesh_tensorflow/transf

Read the input file and save to arrow files in a temporary directory

Create and close a spark session if none exist

Compute binned relative position bias

Function to start an video downloading in one process

batches of dataloader contains a list of dict, keys: image, question, and instance_id

A simple multi device strategy, which distribute blocks in large language modles averagely into multi devices while keeping r

A simple multi device strategy, which distribute blocks in large language modles averagely into multi devices while keeping u

A simple multi device strategy, which distribute blocks in large language modles averagely into multi devices while keeping r

A simple multi device strategy, which distribute blocks in large language modles averagely into multi devices while keeping u

Function to start an video processing in one process

Resize tokenizer and embedding. Note: This is the unoptimized version that may make your embedding size not be divisible by 64.

Convert WebVTT to JSON, optionally removing duplicate lines

extract audio from video