model:
  target: lvdm.models.ddpm3d_cond.T2VFintoneStyleAS
  params:
    linear_start: 0.00085
    linear_end: 0.012
    num_timesteps_cond: 1
    log_every_t: 200
    timesteps: 1000
    first_stage_key: video
    cond_stage_key: caption
    cond_stage_trainable: false
    conditioning_key: crossattn
    image_size: [64, 64]
    channels: 4
    #monitor: val/loss_simple
    scale_by_std: false
    scale_factor: 0.18215
    # training related
    use_ema: false
    uncond_prob: 0.0
    uncond_type: 'empty_seq'


    scheduler_config:
        target: utils.lr_scheduler.LambdaLRScheduler
        interval: 'step'
        frequency: 100
        params:
          start_step: 0
          final_decay_ratio: 0.01
          decay_steps: 20000

    # train_strategy: 'video_only'

    unet_config:
      target: lvdm.modules.networks.openaimodel3d.UNetModel
      params:
        in_channels: 4
        out_channels: 4
        model_channels: 320
        attention_resolutions: [4, 2, 1]
        num_res_blocks: 2
        channel_mult: [1, 2, 4, 4]
        #num_heads: 8
        num_head_channels: 64 # need to fix for flash-attn
        transformer_depth: 1
        context_dim: 1024
        use_linear: true
        use_checkpoint: true
        temporal_conv: false
        temporal_attention: true
        temporal_selfatt_only: true
        use_relative_position: true
        use_causal_attention: false
        temporal_length: 16
        addition_attention: true

    first_stage_config:
      target: lvdm.models.autoencoder.AutoencoderKL
      params:
        embed_dim: 4
        monitor: val/rec_loss
        ddconfig:
          double_z: true
          z_channels: 4
          resolution: 256
          in_channels: 3
          out_ch: 3
          ch: 128
          ch_mult: [1, 2, 4, 4]
          num_res_blocks: 2
          attn_resolutions: []
          dropout: 0.0
        lossconfig:
          target: torch.nn.Identity

    cond_stage_config:
      target: lvdm.modules.encoders.condition.FrozenOpenCLIPEmbedder
      params:
        # version: checkpoints/open_clip/CLIP-ViT-H-14-laion2B-s32B-b79K/open_clip_pytorch_model.bin
        freeze: true
        layer: "penultimate"

    style_stage_config:
      target: lvdm.modules.encoders.condition.FrozenOpenCLIPImageEmbedder
      params:
        # version: checkpoints/open_clip/CLIP-ViT-H-14-laion2B-s32B-b79K/open_clip_pytorch_model.bin
        freeze: true
        only_cls: false
        use_proj: false
        use_shuffle: false
        mask_ratio: 0.0

    adapter_config:
      target: lvdm.modules.encoders.adapter.StyleAdapterDualAttnAS
      cond_name: style
      trainable: true
      params:
        scale: 1.0
        use_norm: true
        image_context_config:
          target: lvdm.modules.encoders.adapter.StyleTransformer
          params:
            in_dim: 1280
            out_dim: 1024
            num_heads: 8
            num_tokens: 8
            n_layers: 3
        scale_predictor_config:
          target: lvdm.modules.encoders.adapter.ScaleEncoder
          params:
            in_dim: 1024
            out_dim: 1
            num_heads: 8
            num_tokens: 16
            n_layers: 2
          # target: lvdm.modules.encoders.adapter.ImageContext
          # params:
          #   width: 1024
          #   context_dim: 1024
          #   token_num: 4