D1:
 x2v_interleave_local:
  dataset_type: text2image_general
  raw_bytes_input: true
  is_image: true
  target_modality: "image"
  text_template: true
  task_type: "t2v"
  dataset_names:
    local_100:
      data_dir:
        - datasets/text2image/local_256.parquet
  debug_parquet_repeat: 256  # debug option for quick smoke tests on small local datasets

  video_transform_args:
    resolution: 768
    mode: "bucket"
    divisible_crop_size: 16
    aspect_ratios: ["21:9", "16:9", "4:3", "1:1", "3:4", "9:16"]
    mean: 0.5
    std: 0.5

  is_mandatory: false
  weight: 1.0
  caption_key: "caption"

D2:
 x2v_interleave_local:
  dataset_type: text2video_general
  raw_bytes_input: true
  is_image: false
  target_modality: "video"
  text_template: true
  task_type: ["t2v", "ff2v"]
  task_type_rate: [0.8, 0.2]
  debug_parquet_repeat: 256  # debug option for quick smoke tests on small local datasets
  dataset_names:
    t2v_local_128:
      data_dir:
        - datasets/text2video/local_128.parquet

  video_transform_args:
    resolution: 640
    mode: "bucket"
    divisible_crop_size: 16
    stride_spatial: 16
    stride_temporal: 4
    aspect_ratios: ["21:9", "16:9", "4:3", "1:1", "3:4", "9:16"]
    mean: 0.5
    std: 0.5

  video_frame_sampler_args:
    type: "multi_clips"
    res_dump: ["12fps_360p"]
    params:
      temporal: 4
      sample_fps: 12
      max_duration: 6
      assert_seconds: false
      truncate: false

  is_mandatory: false
  weight: 1
  caption_key: "caption"


D3:
 x2v_interleave_local:
  dataset_type: image2image
  raw_bytes_input: true
  is_image: true
  target_modality: "image"
  task_type: "tv2v"
  sample_task: "edit"
  vision_cond_type: ["vit", "vae_Nsplit"]
  text_template: true
  dataset_names:
    i2i_local_256:
      data_dir:
        - datasets/image2image/local_256.parquet
  debug_parquet_repeat: 256  # debug option for quick smoke tests on small local datasets

  video_transform_args:
    resolution: 768
    mode: "bucket"
    divisible_crop_size: 16
    stride_spatial: 16
    stride_temporal: 4
    aspect_ratios: ["21:9", "16:9", "4:3", "1:1", "3:4", "9:16"]
    mean: 0.5
    std: 0.5

  vit_video_transform_args:
    resolution: 672
    mode: "bucket"
    divisible_crop_size: 28
    aspect_ratios: ["21:9", "16:9", "4:3", "1:1", "3:4", "9:16"]
    mean: [0.48145466, 0.4578275, 0.40821073]
    std: [0.26862954, 0.26130258, 0.27577711]

  is_mandatory: false
  weight: 1.0


D4:
 x2v_interleave_local:
  dataset_type: video2video
  raw_bytes_input: true
  is_image: false
  target_modality: "video"
  task_type: "tv2v"
  sample_task: "edit"
  vision_cond_type: ["vit", "vae_Nsplit"]
  text_template: true
  dataset_names:
    v2v_local_64:
      data_dir:
        - datasets/video2video/local_64.parquet
  debug_parquet_repeat: 256  # debug option for quick smoke tests on small local datasets

  video_transform_args:
    resolution: 640
    mode: "bucket"
    divisible_crop_size: 16
    stride_spatial: 16
    stride_temporal: 4
    aspect_ratios: ["21:9", "16:9", "4:3", "1:1", "3:4", "9:16"]
    mean: 0.5
    std: 0.5

  vit_video_transform_args:
    resolution: 616
    mode: "bucket"
    divisible_crop_size: 28
    aspect_ratios: ["21:9", "16:9", "4:3", "1:1", "3:4", "9:16"]
    mean: [0.48145466, 0.4578275, 0.40821073]
    std: [0.26862954, 0.26130258, 0.27577711]

  video_frame_sampler_args:
    type: "multi_clips"
    res_dump: ["image_480p", "12fps_480p"]
    params:
      temporal: 4
      sample_fps: 12
      max_duration: 6
      assert_seconds: false
      truncate: false

  is_mandatory: false
  weight: 1.0