model:
  base_learning_rate: 0.0001
  params:
    channels: 4
    cond_stage_config:
      target: ldm.modules.encoders.modules.FrozenCLIPEmbedder
    cond_stage_key: txt
    cond_stage_trainable: false
    conditioning_key: crossattn
    first_stage_config:
      params:
        ddconfig:
          attn_resolutions: []
          ch: 128
          ch_mult:
          - 1
          - 2
          - 4
          - 4
          double_z: true
          dropout: 0.0
          in_channels: 3
          num_res_blocks: 2
          out_ch: 3
          resolution: 256
          z_channels: 4
        embed_dim: 4
        lossconfig:
          target: torch.nn.Identity
        monitor: val/rec_loss
      target: ldm.models.autoencoder.AutoencoderKL
    first_stage_key: jpg
    image_size: 64
    linear_end: 0.012
    linear_start: 0.00085
    log_every_t: 200
    monitor: val/loss_simple_ema
    num_timesteps_cond: 1
    scale_factor: 0.18215
    scheduler_config:
      params:
        cycle_lengths:
        - 10000000000000
        f_max:
        - 1.0
        f_min:
        - 1.0
        f_start:
        - 1.0e-06
        warm_up_steps:
        - 10000
      target: ldm.lr_scheduler.LambdaLinearScheduler
    timesteps: 1000
    unet_config:
      params:
        attention_resolutions:
        - 4
        - 2
        - 1
        channel_mult:
        - 1
        - 2
        - 4
        - 4
        context_dim: 768
        image_size: 32
        in_channels: 4
        legacy: false
        model_channels: 320
        num_heads: 8
        num_res_blocks: 2
        out_channels: 4
        transformer_depth: 1
        use_checkpoint: true
        use_spatial_transformer: true
      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
    use_ema: false
  target: ldm.models.diffusion.ddpm.LatentDiffusion