FP32 matmul for MMWeight when autocast does not apply (e.g. AdaIN).

Returns the closest integer to 'number' that is divisible by 'factor'.

Run a subprocess command and handle errors.

Set or append vision KV cache for context frames.

Rescales the image so that the following conditions are met: 1. Both dimensions (height and width) are divisible by 'factor'. 2. The to

Generate state dictionary containing base attributes and LoRA attributes. Args: cls: Target class instance containing tensor attributes

Tokenize the input text. Args: text (str or list): Input text.

Move parameters to GPU device (supports cuda/intel xpu)

Create a new SDOps instance with the specified replacement added to the mapping.

Apply Projective Positional Encoding to Q, K, V. Args: q: Query tensor [B, L, H, D] or [B, H, L, D] k: Key t

Materialize one segment's complete conditioning package. This is the core of the adapter. It decides: - which absolute frames this se

Calculate the normalized distances between preds and target. Note: batch_size: N num_keypoints: K dimension of keypoints:

Concatenate a tensor with a padding tensor of the given value.

Avoid leaking ltx2_s2v audio conditioning into t2av/i2av runs on a reused runner.

Count the actual decodable frame count of an mp4 (PyAV demux).

Create (or recreate) cache pools with resolution-dependent sizes.

So RoCE gid_index matches the data-plane IP on each worker (multi-node).

Recursively search for files matching a glob pattern and return the parent directory of the first match.