model:
  base_learning_rate: 2.0e-05
  target: ldm.models.diffusion.ddpm.LatentDiffusion
  params:
    linear_start: 0.0015
    linear_end: 0.0195
    num_timesteps_cond: 1
    log_every_t: 200
    timesteps: 1000
    first_stage_key: image
    cond_stage_key: action_
    #scheduler_sampling_rate: 0.0
    hybrid_key: c_concat
    image_size: 256
    channels: 3
    cond_stage_trainable: true
    conditioning_key: hybrid
    monitor: val/loss_simple_ema

    unet_config:
      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
      params:
        image_size: 256
        in_channels: 24
        out_channels: 3
        model_channels: 192
        attention_resolutions:
        - 32
        - 16
        - 8
        num_res_blocks: 2
        channel_mult:
        - 1
        - 2
        - 3
        - 5
        num_head_channels: 32
        use_spatial_transformer: true
        transformer_depth: 1
        context_dim: 768

    first_stage_config:
      target: ldm.models.autoencoder.IdentityFirstStage
      params:
        embed_dim: 3
        n_embed: 8192

    cond_stage_config:
      target: ldm.modules.encoders.modules.GPTEmbedder
      params:
        n_embed: 768
        n_layer: 12

data:
  target: data.data_processing.datasets.DataModule
  params:
    batch_size: 16
    num_workers: 1
    wrap: false
    shuffle: True
    drop_last: True
    pin_memory: True
    prefetch_factor: 2
    persistent_workers: True
    train:
      target: data.data_processing.datasets.ActionsData
      params:
        data_csv_path: train_dataset/train_dataset_14frames_firstframe.csv
    validation:
      target: data.data_processing.datasets.ActionsData
      params:
        data_csv_path: train_dataset/train_dataset_14frames_firstframe.csv

lightning:
  trainer:
    benchmark: False
    max_epochs: 4
    limit_val_batches: 0
    accelerator: gpu
    gpus: 1
    accumulate_grad_batches: 8
    gradient_clip_val: 1
    checkpoint_callback: True