{
  "format": "miowtion-veda-predictor-v1",
  "model_type": "veda_tile_score_predictor",
  "weights_file": "minimax_h3_r2va_veda_preview_fp8.safetensors",
  "base_model": "MiniMaxAI/MiniMax-H3",
  "task": "reference-to-audio-video",
  "variant": "Ref2VA",
  "num_layers": 50,
  "num_heads": 56,
  "head_dim": 128,
  "tile_size": 128,
  "dtype": "float8_e4m3fn",
  "target_budget_kind": "tiles",
  "target_budget_value": 32,
  "ref_budget_kind": "tiles",
  "ref_budget_value": 32,
  "tile_conditions": true,
  "num_inference_steps": 8,
  "geometries": {
    "16x9_t102": {
      "latent_grid": [
        102,
        24,
        42
      ],
      "tile_shapes": [
        "8x4x4",
        "2x8x8",
        "4x8x4",
        "8x8x2"
      ]
    },
    "16x9_t37": {
      "latent_grid": [
        37,
        24,
        42
      ],
      "tile_shapes": [
        "8x4x4",
        "2x8x8",
        "4x8x4",
        "8x8x2"
      ]
    },
    "16x9_t72": {
      "latent_grid": [
        72,
        24,
        42
      ],
      "tile_shapes": [
        "8x4x4",
        "2x8x8",
        "4x4x8",
        "4x8x4",
        "8x8x2"
      ]
    },
    "1x1_t102": {
      "latent_grid": [
        102,
        24,
        24
      ],
      "tile_shapes": [
        "8x4x4",
        "2x8x8",
        "4x4x8",
        "4x8x4",
        "8x8x2",
        "8x2x8"
      ]
    },
    "1x1_t37": {
      "latent_grid": [
        37,
        24,
        24
      ],
      "tile_shapes": [
        "8x4x4",
        "2x8x8",
        "4x4x8",
        "8x8x2",
        "8x2x8"
      ]
    },
    "1x1_t72": {
      "latent_grid": [
        72,
        24,
        24
      ],
      "tile_shapes": [
        "8x4x4",
        "2x8x8",
        "4x4x8",
        "4x8x4",
        "8x8x2",
        "8x2x8"
      ]
    },
    "4x3_t102": {
      "latent_grid": [
        102,
        24,
        32
      ],
      "tile_shapes": [
        "8x4x4",
        "2x8x8",
        "4x4x8",
        "4x8x4",
        "8x8x2",
        "8x2x8"
      ]
    },
    "4x3_t37": {
      "latent_grid": [
        37,
        24,
        32
      ],
      "tile_shapes": [
        "8x4x4",
        "2x8x8",
        "4x4x8",
        "4x8x4",
        "8x8x2",
        "8x2x8"
      ]
    },
    "4x3_t72": {
      "latent_grid": [
        72,
        24,
        32
      ],
      "tile_shapes": [
        "8x4x4",
        "2x8x8",
        "4x4x8",
        "4x8x4",
        "8x8x2",
        "8x2x8"
      ]
    },
    "9x16_t102": {
      "latent_grid": [
        102,
        42,
        24
      ],
      "tile_shapes": [
        "8x4x4",
        "2x8x8",
        "4x4x8",
        "8x2x8"
      ]
    },
    "9x16_t37": {
      "latent_grid": [
        37,
        42,
        24
      ],
      "tile_shapes": [
        "8x4x4",
        "2x8x8",
        "4x4x8",
        "8x2x8"
      ]
    },
    "9x16_t72": {
      "latent_grid": [
        72,
        42,
        24
      ],
      "tile_shapes": [
        "8x4x4",
        "2x8x8",
        "4x8x4",
        "4x4x8",
        "8x2x8"
      ]
    }
  },
  "scale_suffix": ".__scale",
  "scale_granularity": "per_head_amax",
  "load_dtype": "bfloat16",
  "compute_dtype": "float32",
  "few_step_lora": "trained with Turbo 8 NFE; other schedules need validation",
  "training": {
    "stage": "tile-score predictor distillation",
    "initialization": "T2VA predictor EMA",
    "updates_r2va_finetuning": 600,
    "training_conditions": 1000,
    "trained_geometries": [
      "16:9",
      "9:16",
      "4:3",
      "1:1"
    ],
    "trained_latent_t": [
      37
    ],
    "export": "live",
    "backbone_frozen": true
  }
}