{
  "model": "DiTGS camera-fixed Residual CA",
  "variant": "residual",
  "format": "ditgs-raw-full-model-v1",
  "file": "model.pt",
  "bytes": 22054988097,
  "sha256": "d20e90297953151a90ac53c9022ddf8530eaa7f6359e3e74a7bec8654e2c48c9",
  "weight_selection": "raw, not EMA",
  "storage": "original dtypes, no conversion or quantization",
  "training": {
    "warmup_accepted_updates": 10000,
    "joint_new_accepted_updates": 20000,
    "original_world_size": 16,
    "camera_fixed": true
  },
  "model_state": {
    "tensors": 1542,
    "bytes": 22044815088,
    "components": {
      "empty_text": {
        "tensors": 1,
        "bytes": 8388608,
        "elements": 2097152
      },
      "dit": {
        "tensors": 826,
        "bytes": 20001805056,
        "elements": 5000451264
      },
      "vae": {
        "tensors": 198,
        "bytes": 1031441904,
        "elements": 257860476
      },
      "adapter": {
        "tensors": 10,
        "bytes": 18382848,
        "elements": 4595712
      },
      "gs": {
        "tensors": 507,
        "bytes": 984796672,
        "elements": 246199168
      }
    },
    "dtypes": {
      "torch.float32": 1542
    }
  },
  "all_buffers": {
    "tensors": 5,
    "bytes": 9437568
  },
  "model_contract": {
    "protocol": "ditgs-flashworld-seqca-layerscale-camera-fixed-full8-v7",
    "model_repo": "Wan-AI/Wan2.2-TI2V-5B-Diffusers",
    "model_revision": "b8fff7315c768468a5333511427288870b2e9635",
    "dense_views": 8,
    "feature_dim": 1024,
    "scene_scale": 0.15,
    "image_size": 256,
    "image_width": 256,
    "vae_temporal_policy": "crop-causal-history-t1",
    "feature_source": "vae-decoder-final-norm-activation",
    "adapter_patch_size": 4,
    "ray_encoding": "tokengs-moment-direction",
    "conditioning_projection": "split-pretrained-flashworld-native-condition",
    "gaussian_policy": "local-repair-v1",
    "view_fusion": "separate-2view-gs-8view-dit-memories",
    "sparse_encoder_policy": "frozen-official-base1024-joint-two-view",
    "source_attention": "sequential-residual-ca-independent-source-kv",
    "data_recovery_policy": "unbounded-real-scene-resampling",
    "kv_sharing_policy": "independent-sources-shared-across-decoder-layers",
    "gs_memory_views": 2,
    "dit_memory_views": 8,
    "dit_residual_scaling": "native-trainable-per-channel-layerscale",
    "dit_layerscale_initialization": "copy-native-gs-ca-gamma-independent",
    "warmup_trainability": "adapter-ca-kv-layerscale-and-native-gs-decoder-head-wan-frozen",
    "gs_render_views": 8,
    "gs_render_policy": "all-eight-views-both-observed-and-six-missing",
    "backbone_source": "nightkiller/FlashWorldPretrain",
    "flashworld_revision": "896f3964301821d0bdd92ccb9560fbf51b855599",
    "flashworld_source_sha256": "5da5eb36fb5193cd057b04239e3576680512e998888eb64bfc195617df4e243c",
    "latent_normalization": "flashworld-pretrain-native-bf16-rounded",
    "dit_camera_policy": "flashworld-opengl-firstcam-maxradius-v1",
    "dit_camera_normalization_eps": 0.01,
    "dit_output_initialization": "copy-native-ca-unfolded",
    "visibility_policy": "unculled-frustum-log-distance-v1",
    "visibility_distance_scale": 0.15,
    "training_policy": "camera-fixed-adapter-gs-warmup10k-then-joint20k",
    "residual_post_norm": "none",
    "dit_kv_initialization": "copy-native-kv-and-keynorm-independent",
    "adapter_norm_initialization": "copy-native-encoder-final-ln",
    "camera_preprocessing": "dl3dv-actual-raster-rectified-pinhole-integer-crop-v1"
  },
  "code_repo": "https://github.com/Qingzhongyun/DiTGS",
  "code_commit": "c3d7e98679770fb407571016383095b803e27b44",
  "validation": {
    "cpu_round_trip_all_tensors_equal": true,
    "config_contract_identical": true,
    "private_server_paths_removed": true,
    "training_or_inference_rerun": false
  },
  "license": "cc-by-nc-sa-4.0",
  "upstream_notices": "UPSTREAM_NOTICES.md",
  "created_at": "2026-11-11T12:16:31.324554+08:00"
}