o }oi@sBddlmZddlmZddlmZddZGdddejZdS))TupleN) functionalcCs|D]}tj|q|S)z Initializes all parameters of the given module to zero. Args: module (nn.Module): The module whose parameters will be initialized to zero. Returns: nn.Module: The same module with zero-initialized parameters. ) parametersnninitzeros_)modulepr l/home/ubuntu/.local/lib/python3.10/site-packages/nemo/collections/diffusion/models/flux_controlnet/layers.py zero_modules r c sDeZdZdZ  d dededeedfffdd Zd d ZZS) ControlNetConditioningEmbeddingu Quoting from https://arxiv.org/abs/2302.05543: "Stable Diffusion uses a pre-processing method similar to VQ-GAN [11] to convert the entire dataset of 512 × 512 images into smaller 64 × 64 “latent images” for stabilized training. This requires ControlNets to convert image-based conditions to 64 × 64 feature space to match the convolution size. We use a tiny network E(·) of four convolution layers with 4 × 4 kernels and 2 × 2 strides (activated by ReLU, channels are 16, 32, 64, 128, initialized with Gaussian weights, trained jointly with the full model) to encode image-space conditions ... into feature maps ..."  `conditioning_embedding_channelsconditioning_channelsblock_out_channels.c sttj||dddd|_tg|_tt|dD]'}||}||d}|j tj||ddd|j tj||ddddqt tj|d|ddd|_ dS) a Initializes the model with convolutional layers for processing conditioning inputs. Args: conditioning_embedding_channels (int): Number of output channels for the conditioning embedding. conditioning_channels (int): Number of input channels for the conditioning data. Default is 3. block_out_channels (Tuple[int, ...]): Tuple specifying the output channels for each block. Default is (16, 32, 96, 256). rr) kernel_sizepadding)rrstrideN) super__init__rConv2dconv_in ModuleListblocksrangelenappendr conv_out)selfrrri channel_in channel_out __class__r r r.s    z(ControlNetConditioningEmbedding.__init__cCs@||}t|}|jD] }||}t|}q ||}|S)a0 Passes the conditioning input through the model to produce an embedding. Args: conditioning (torch.Tensor): Input tensor representing conditioning data. Returns: torch.Tensor: The resulting embedding tensor after processing through the network. )r Fsilur"r&)r' conditioning embeddingblockr r r forwardOs    z'ControlNetConditioningEmbedding.forward)rr) __name__ __module__ __qualname____doc__intrrr2 __classcell__r r r+r r $s  !r ) typingrtorch.nnrrr-r Moduler r r r r s