model:
  target: ldm.modules.latent_diffusion.ddpm.LatentDiffusion
  params:
    first_stage_config:
      target: ldm.modules.latent_encoder.autoencoder.AutoencoderKL
      params:
        reload_from_ckpt: ./models/vae/vae.ckpt
        vocoder_path: ./models/vocoder
        embed_dim: 16
        ddconfig:
          double_z: true
          mel_bins: 128
          z_channels: 16
          resolution: 128
          downsample_time: false
          in_channels: 1
          out_ch: 1
          ch: 128
          ch_mult:
          - 1
          - 1
          - 2
          - 2
          - 4
          num_res_blocks: 2
          attn_resolutions:
          - 16
          dropout: 0.0
    
    sampling_rate: 16000
    linear_start: 0.0015
    linear_end: 0.0195
    num_timesteps_cond: 1
    timesteps: 1000
    parameterization: eps
    first_stage_key: fbank
    latent_t_size: 112
    latent_f_size: 16
    channels: 16
    conditioning_key: crossattn
    scale_by_std: true
    model_config:
      target: ldm.modules.diffusionmodules.PixArt.PixArt_MDT
      params:
        input_size:
        - 112
        - 16
        patch_size:
        - 2
        - 2
        in_channels: 16
        hidden_size: 1472
        depth: 28
        num_heads: 16
        mlp_ratio: 4.0
        class_dropout_prob: 0.1
        pred_sigma: true
        drop_path: 0.0
        window_size: 0
        window_block_indexes: None
        use_rel_pos: false
        cond_dim: 1024
        lewei_scale: 1.0
        overlap:
        - 0
        - 0
        use_cfg: true
        mask_ratio: 0.3
        decode_layer: 8
    cond_stage_config:
      crossattn:
        cond_stage_key: text
        conditioning_key: crossattn
        target: ldm.utils.util.TextEncoder
        params: 
          ckpt_path: ./models/flan-t5-large