{
  "format": "ttvidt-release-v1",
  "model": {
    "base_model_name": null,
    "motion_encoder_config": {
      "temporal_patch_size": 1,
      "temporal_patch_overlap": 0,
      "motion_layers_period": 1,
      "num_motion_tokens": 8,
      "motion_ffn_type": "gelu",
      "tt_mode": "3d",
      "tt_downsample": 4
    },
    "motion_decoder_config": {
      "image_dim": 4,
      "patch_size": 2,
      "decoder_style": "dit",
      "num_layers": 12,
      "hidden_size": 768,
      "intermediate_size": 3072,
      "num_heads": 12,
      "decode_mode": "diffusion",
      "qk_norm": true,
      "attn_bias": false,
      "use_final_norm": true,
      "encoder_hidden_size": 768,
      "temporal_patch_size": 1
    },
    "encoder_ae_mode": "video",
    "decoder_ae_mode": "image",
    "latent_mean": [
      -0.69,
      -0.48,
      -0.6,
      0.28
    ],
    "latent_std": [
      12.38,
      11.22,
      7.93,
      21.22
    ],
    "patch_size": null,
    "image_dim": null,
    "token_counts": [
      1
    ],
    "gradient_checkpointing": false,
    "train_mode": "regression",
    "mae_mask_ratio": 0.75,
    "ar_shift_range": null,
    "ar_shift_sampling": "gamma",
    "source_frame_count": 8,
    "use_ref": false,
    "unfreeze_backbone": true,
    "decoder_mask_ratio": 0.0,
    "mask_patch_size": 4,
    "diffusion_timestep_sampling": "uniform",
    "backbone_arch": "ttvidt",
    "backbone_config": {
      "transformers_version": "5.14.1",
      "architectures": [
        "DINOv3ViTModel"
      ],
      "output_hidden_states": false,
      "return_dict": true,
      "dtype": "float32",
      "chunk_size_feed_forward": 0,
      "is_encoder_decoder": false,
      "id2label": {
        "0": "LABEL_0",
        "1": "LABEL_1"
      },
      "label2id": {
        "LABEL_0": 0,
        "LABEL_1": 1
      },
      "problem_type": null,
      "patch_size": 16,
      "hidden_size": 768,
      "intermediate_size": 3072,
      "num_hidden_layers": 12,
      "num_attention_heads": 12,
      "hidden_act": "gelu",
      "attention_dropout": 0.0,
      "initializer_range": 0.02,
      "layer_norm_eps": 1e-05,
      "rope_theta": 100.0,
      "image_size": 224,
      "num_channels": 3,
      "query_bias": true,
      "key_bias": false,
      "value_bias": true,
      "proj_bias": true,
      "mlp_bias": true,
      "layerscale_value": 1.0,
      "drop_path_rate": 0.0,
      "use_gated_mlp": false,
      "num_register_tokens": 4,
      "pos_embed_shift": null,
      "pos_embed_jitter": null,
      "pos_embed_rescale": 2.0,
      "apply_layernorm": true,
      "reshape_hidden_states": true,
      "stage_names": [
        "stem",
        "stage1",
        "stage2",
        "stage3",
        "stage4",
        "stage5",
        "stage6",
        "stage7",
        "stage8",
        "stage9",
        "stage10",
        "stage11",
        "stage12"
      ],
      "_name_or_path": "facebook/dinov3-vitb16-pretrain-lvd1689m",
      "model_type": "dinov3_vit",
      "temporal_patch_size": 1,
      "temporal_patch_overlap": 0,
      "motion_layers_period": 1,
      "num_motion_tokens": 8,
      "motion_ffn_type": "gelu",
      "tt_mode": "3d",
      "tt_downsample": 4,
      "output_attentions": false,
      "out_features": [
        "stage12"
      ],
      "out_indices": [
        12
      ]
    },
    "decoder_lr_multiplier": 1.0,
    "fused_adam": false,
    "ema_foreach": false,
    "nan_guard_interval": 1,
    "name": "ttvidt-tt3d-diffcomp",
    "log_interval": 2500,
    "learning_rate": 0.0005,
    "base_dim": 256,
    "decoder_base_dim": null,
    "weight_decay": 0.01,
    "betas": [
      0.9,
      0.98
    ],
    "scheduler_config": {
      "lr": {
        "mode": "cosine",
        "end": 436705,
        "min_value": 0.01,
        "warmup": 10000
      }
    }
  },
  "encoder_only": true,
  "name": "ttvidt-tt3d",
  "model_type": "ttvidt",
  "architectures": [
    "TTVidTModel"
  ],
  "auto_map": {
    "AutoConfig": "modeling_ttvidt.TTVidTConfig",
    "AutoModel": "modeling_ttvidt.TTVidTModel"
  }
}