o ̳i @srddlmZddlmZmZmZddddgdddd d f d ed ed ededeedededededefddZdS))List)FluxAutoencoder FluxDecoder FluxEncoder)r r gxz,C?g=U? resolutionch_inch_outch_basech_multsch_zn_layers_per_resample_block scale_factor shift_factorreturnc sZfdd|D} t||| |||d} t||tt| |d||d} t|||f| | dS)a The image autoencoder for all current Flux diffusion models: - FLUX.1-dev - FLUX.1-schnell - FLUX.1-Canny-dev - FLUX.1-Depth-dev - FLUX.1-Fill-dev ch = number of channels (size of the channel dimension) Args: resolution (int): The height/width of the square input image. ch_in (int): The number of channels of the input image. ch_out (int): The number of channels of the output image. ch_base (int): The base number of channels. This gets multiplied by `ch_mult` values to get the number of inner channels during downsampling/upsampling. ch_mults (List[int]): The channel multiple per downsample/upsample block. This gets multiplied by `ch_base` to get the number of inner channels during downsampling/upsampling. ch_z (int): The number of latent channels (dimension of the latent vector `z`). n_layers_per_resample_block (int): Number of resnet layers per downsample/upsample block. scale_factor (float): Constant for scaling `z`. shift_factor (float): Constant for shifting `z`. Returns: FluxAutoencoder csg|]}|qSr).0multrrY/home/ubuntu/.local/lib/python3.10/site-packages/torchtune/models/flux/_model_builders.py 0sz&flux_1_autoencoder..)rrchannelsn_layers_per_down_blockrrr )rrrn_layers_per_up_blockrr) img_shapeencoderdecoder)rrlistreversedr) r rrrrrrrrrr!r"rrrflux_1_autoencoder s,%   r%N) typingr"torchtune.models.flux._autoencoderrrrintfloatr%rrrrs@