Causal Conv3d with configurable padding for temporal axis.

Causal 3d convolution.

Causal 3d convolution.

Causal 3d convolusion.

A lockable/unlockable dictionary. After locking, any in-place modifications will raise TypeError. By default auto_wrap=True, which recursivel

Ring buffer backed by RDMA-accessible memory. Role model: - server: producer side, owns and registers memory regions. - client: consumer

Defines a single LoRA format pattern and how to extract its components.

ResNet-style block for 3D video tensors.

Immutable class representing state dict key operations.

Root Mean Square Layer Normalization for Channel-First or Last

Drop-in replacement for ``nn.Linear`` that routes through an MMWeight.

Container for decoded audio samples and metadata. Attributes: waveform: Audio waveform tensor. sampling_rate: Sampling rate (

Shape of audio in VAE latent space: (batch, channels, frames, mel_bins). mel_bins is the number of frequency bins from the mel-spectrogram en

Perform Conv3d on patches with memory-efficient symmetric padding.

Per-pixel (per-location) RMS normalization layer. For each element along the chosen dimension, this layer normalizes the tensor by the ro

Immutable class representing state dict key operations.

Z-Image single stream transformer block. Based on ZImageTransformerBlock with attention_norm1, attention_norm2, ffn_norm1, ffn_norm2.

Causal self-attention with a single head.

Causal self-attention with a single head.

# DPT Head for dense prediction tasks. # This module implements the DPT (Dense Prediction Transformer) head as proposed in # "Vision Tra

N-dimensional pixel shuffle operation for upsampling tensors. Args: dims (int): Number of dimensions to apply pixel shuffle to.

Loads weights and configuration metadata from safetensors model files. Unlike SafetensorsStateDictLoader, this loader can read model configur