{ "schema_version": 2, "identity_source": "saved inference_config", "model": { "model.arch": "rgb_dep_sem", "model.attention_layout": "rgb_auxiliary_causal", "model.action_conditioning": "predicted_depth_head_semantic", "model.world_action_visibility": "none", "model.input_paradigm": "t_concat", "model.view_layout": "robotwin_t", "model.depth.repr": "dvae", "model.depth.condition_source": "dataset", "model.depth.target_source": "dataset", "model.depth.condition_temporal_context": "current_frame", "model.depth_projection": "independent", "model.semantic_projection": "independent", "model.semantic.parametrization": "v_pred", "model.semantic.compression": "svae", "model.semantic.channels": 48, "model.semantic.compression.output_normalization": "layernorm" }, "historical_restoration_defaults": { "model.world_action_visibility": { "saved": null, "effective": "none", "reason": "historical checkpoint restoration rule" } }, "data": { "releases": [], "scene_filter": null, "scene_order": null, "episode_split": null, "train_episode_count": null, "train_episodes_sha256": null, "validation_fraction": null, "depth_codec": null, "action_window_frames": 33, "action_video_freq_ratio": 4, "global_sample_stride": null, "temporal_sampling": null, "normalization_statistics_fingerprint": "82b5328e093511ec35dd52ba17a622d5357fa385903c49dd0a544fceb7b5e243" }, "training": { "num_epochs": null, "learning_rate": null, "batch_size": null, "gradient_accumulation_steps": null, "mixed_precision": null, "seed": null, "total_optimizer_steps": null, "execution": {}, "world_size": null, "global_batch_size": null, "runtime_reference": { "global_step": 32220, "scheduler_type": null, "sched_total_train_steps": null, "sched_warmup_steps": null, "lr_conv_ratio": null, "enforce_grad_clipping": null, "effective_grad_clip": null, "video_moe_warmup": null } }, "augmentation": { "enabled": null, "patchwam": null, "conditioning_latent_noise": null, "color": { "brightness": null, "contrast": null, "saturation": null, "hue": null, "probability": null }, "geometry": {}, "coarse_dropout": null, "effective_geometry_enabled": null, "evaluation_random_augmentation": false }, "frozen_assets": { "semantic_compressor_state_sha256": "e8760f44c242bb79232f2d3740dda71956cac2001ba8c00ceaec77ea80e36961", "semantic_artifact_fingerprint": "d3383726600435ddc50852e7e12db67c393b1a2f54568803214753980bf3597d", "depth_vae_identity": { "type": "prism_wan22_depth", "weights_sha256": "c869ccb52c79b398e0f173f3054cfa229f620ef108ceeec82ad0670ca8bb1957", "latent_channels": 48, "spatial_factor": 16, "temporal_factor": 4, "input_channels": 1, "latent_mean": [ 0.4887385368347168, -2.0717055797576904, -0.09361350536346436, -1.9305212497711182, 0.6804881691932678, -0.18310534954071045, -0.017370181158185005, -0.05728627368807793, -0.7831002473831177, -1.8836477994918823, -0.2327776700258255, 1.9035433530807495, 0.46046552062034607, -0.6453059911727905, -1.4464999437332153, 0.21215873956680298, 1.3156590461730957, 0.007310659624636173, 0.745159924030304, -0.7342119216918945, -0.3538833260536194, 2.9265501499176025, 1.9496617317199707, 0.3119705617427826, -1.0393187999725342, -0.45994818210601807, -0.13466505706310272, -1.2324262857437134, 0.3329582214355469, -0.20562906563282013, 1.6063544750213623, -0.37199077010154724, 0.10520733147859573, -0.005950210150331259, -1.021066427230835, -0.16209781169891357, -0.8151541948318481, 2.7302606105804443, -0.5384726524353027, -3.4191009998321533, -0.4929030239582062, 2.472595691680908, 0.09065636992454529, 0.1566804200410843, 0.6077839732170105, 0.16436097025871277, 0.574763834476471, -1.018074631690979 ], "latent_std": [ 0.8759883046150208, 4.3842878341674805, 0.600670576095581, 4.464785575866699, 1.4707190990447998, 0.7163799405097961, 0.39080148935317993, 0.521407425403595, 2.459089994430542, 4.432198524475098, 0.9659329056739807, 4.131554126739502, 1.7189385890960693, 1.1850491762161255, 3.289055347442627, 0.6653669476509094, 2.821338176727295, 0.44802847504615784, 1.4034730195999146, 1.3087146282196045, 0.9007586240768433, 6.145054340362549, 3.761077404022217, 1.2898753881454468, 2.543177366256714, 1.234912633895874, 0.765889585018158, 2.0544753074645996, 0.9371235966682434, 0.6772623062133789, 3.1669251918792725, 0.5857013463973999, 0.5159927606582642, 0.5009284615516663, 2.351271152496338, 0.6470572352409363, 1.8693331480026245, 5.487253189086914, 1.4262751340866089, 7.51733922958374, 0.8003208637237549, 4.9757843017578125, 0.6112902760505676, 0.5305010676383972, 1.6170129776000977, 0.4477066993713379, 1.232177972793579, 2.2924752235412598 ], "preprocess": { "unit": "metres", "near_m": 0.0, "far_m": 1.5, "resize": "bilinear", "align_corners": false, "layout": "robotwin_t", "normalization": "linear_minus_one_one", "posterior": "mean" } } }, "saved_construction_and_objectives": { "model.video_dit_config": { "in_dim": 48, "out_dim": 48, "text_dim": 4096, "freq_dim": 256, "patch_size": [ 1, 2, 2 ], "num_layers": 30, "has_image_input": false, "has_image_pos_emb": false, "has_ref_conv": false, "add_control_adapter": false, "in_dim_control_adapter": 24, "seperated_timestep": true, "require_vae_embedding": false, "require_clip_embedding": false, "fuse_vae_embedding_in_latents": true, "action_conditioned": false, "action_dim": 14, "action_group_causal_mask_mode": "group_diagonal", "video_attention_mask_mode": "first_frame_causal", "moe": { "enabled": true, "routing": "per_sample", "num_experts": 4, "top_k": 0, "use_shared_expert": true, "moe_layers": [ 21, 24, 27 ], "router_cond": "modality", "token_modality_bias": true, "token_dispatch": "dense_masked", "loss_lambda_aux": 0.01, "loss_lambda_entropy": 0.0, "aux_warmup_steps": 1611, "aux_warmup_ratio": 0.05, "aux_warmup_unit": "optimizer_step", "init_noise_std": 0.02 }, "attn_head_dim": 128, "eps": 1e-06, "ffn_dim": 14336, "hidden_dim": 3072, "num_heads": 24 }, "model.action_dit_config": { "action_dim": 14, "text_dim": 4096, "freq_dim": 256, "num_layers": 30, "moe": null, "attn_head_dim": 128, "eps": 1e-06, "ffn_dim": 4096, "hidden_dim": 1024, "num_heads": 24 }, "model.loss": { "lambda_video": 1.0, "lambda_action": 1.0, "lambda_semantic": 1.0, "lambda_depth": 1.0 }, "model.semantic.velocity_loss_form": "direct_error", "model.video_scheduler": { "train_shift": 5.0, "infer_shift": 5.0, "num_train_timesteps": 1000 }, "model.action_scheduler": { "train_shift": 5.0, "infer_shift": 5.0, "num_train_timesteps": 1000 }, "model.semantic.backbone": "dinov3_vitb16", "model.semantic.feature_normalization": null, "model.semantic.temporal_pool": "mean", "model.depth.pointmap": { "norm_bounds": { "x": [ -0.6, 0.6 ], "y": [ -0.6, 0.6 ], "z": [ 0.0, 1.5 ] }, "max_depth_m": 2.0 }, "model.depth.intrinsics": { "source": "auto", "sim": { "high": { "fovy_deg": 37.0, "ref_h": 480, "ref_w": 640 }, "left": { "fovy_deg": 37.0, "ref_h": 480, "ref_w": 640 }, "right": { "fovy_deg": 37.0, "ref_h": 480, "ref_w": 640 }, "image": { "fovy_deg": 45.0, "ref_h": 224, "ref_w": 224 }, "wrist_image": { "fovy_deg": 75.0, "ref_h": 224, "ref_w": 224 } } } }, "archived_run_configuration": { "source": "original resolved run/config.yaml; never current recipe defaults", "missing_fields": "unknown/not recorded; null is not false", "inactive_fields": "recorded values may be inactive; use arch and enabled/mode gates", "choices": { "Architecture and Observation": { "model.model_family": "wam", "model.model_id": "Wan-AI/Wan2.2-TI2V-5B", "model.arch": "rgb_dep_sem", "model.attention_layout": "rgb_auxiliary_causal", "model.action_conditioning": "predicted_depth_head_semantic", "model.world_action_visibility": null, "model.input_paradigm": "t_concat", "model.view_layout": "robotwin_t", "model.depth_projection": "independent", "model.semantic_projection": "independent", "model.wrist": { "view_size": [ 128, 160 ], "frame": "first", "attends_video": false, "rope_offset": null, "cameras": [ "left", "right" ] }, "model.proprio_dim": 14, "model.video_dit_config.video_attention_mask_mode": "first_frame_causal", "model.video_dit_config.action_conditioned": false, "model.semantic.fusion_layers": [ 7, 23 ], "model.depth.fusion_layers": null }, "Representations and Frozen Encoders": { "model.semantic.backbone": "dinov3_vitb16", "model.semantic.feat_dim": 768, "model.semantic.feature_normalization": null, "model.semantic.temporal_pool": "mean", "model.semantic.clip_frames": null, "model.semantic.image_size": null, "model.semantic.compression.type": "svae", "model.semantic.compression.latent_dim": 48, "model.semantic.compression.output_normalization": "layernorm", "model.semantic.compression.model_config": { "input_dim": 768, "latent_dim": 48, "num_heads": 16, "num_layers": 3, "intermediate_size": 2048, "dropout": 0.0 }, "model.depth.repr": "dvae", "model.depth.condition_source": "dataset", "model.depth.target_source": "dataset", "model.depth.condition_temporal_context": "current_frame", "model.depth.estimator": "vda", "model.depth.vda": { "backbone": "vitb", "input_size": 518 }, "model.depth.unidepth": { "version": "v2", "backbone": "vitb14", "resolution_level": null }, "model.depth.pointmap": { "norm_bounds": { "x": [ -0.6, 0.6 ], "y": [ -0.6, 0.6 ], "z": [ 0.0, 1.5 ] }, "max_depth_m": 2.0 }, "model.depth.intrinsics": { "source": "auto", "sim": { "high": { "fovy_deg": 37.0, "ref_h": 480, "ref_w": 640 }, "left": { "fovy_deg": 37.0, "ref_h": 480, "ref_w": 640 }, "right": { "fovy_deg": 37.0, "ref_h": 480, "ref_w": 640 }, "image": { "fovy_deg": 45.0, "ref_h": 224, "ref_w": 224 }, "wrist_image": { "fovy_deg": 75.0, "ref_h": 224, "ref_w": 224 } } }, "model.depth.vae.identity.type": "prism_wan22_depth", "model.depth.vae.identity.latent_channels": 48, "model.depth.vae.identity.input_channels": 1, "model.depth.vae.identity.spatial_factor": 16, "model.depth.vae.identity.temporal_factor": 4, "model.depth.vae.identity.preprocess": { "unit": "metres", "near_m": 0.0, "far_m": 1.5, "resize": "bilinear", "align_corners": false, "layout": "robotwin_t", "normalization": "linear_minus_one_one", "posterior": "mean" }, "model.depth.colormap": { "name": "turbo", "near_m": 0.1, "far_m": 2.0 }, "model.depth.defm": { "backbone": "defm_vit_s14", "feat_dim": 384, "patch_size": 14 } }, "Backbone Capacity and MoE": { "model.video_dit_config": { "in_dim": 48, "out_dim": 48, "text_dim": 4096, "freq_dim": 256, "patch_size": [ 1, 2, 2 ], "num_layers": 30, "has_image_input": false, "has_image_pos_emb": false, "has_ref_conv": false, "add_control_adapter": false, "in_dim_control_adapter": 24, "seperated_timestep": true, "require_vae_embedding": false, "require_clip_embedding": false, "fuse_vae_embedding_in_latents": true, "action_conditioned": false, "action_dim": 14, "action_group_causal_mask_mode": "group_diagonal", "video_attention_mask_mode": "first_frame_causal", "moe": { "enabled": true, "routing": "per_sample", "num_experts": 4, "top_k": 0, "use_shared_expert": true, "moe_layers": [ 21, 24, 27 ], "router_cond": "modality", "token_modality_bias": true, "token_dispatch": "dense_masked", "loss_lambda_aux": 0.01, "loss_lambda_entropy": 0.0, "aux_warmup_steps": 1611, "aux_warmup_ratio": 0.05, "aux_warmup_unit": "optimizer_step", "init_noise_std": 0.02 }, "attn_head_dim": 128, "eps": 1e-06, "ffn_dim": 14336, "hidden_dim": 3072, "num_heads": 24 }, "model.action_dit_config": { "action_dim": 14, "text_dim": 4096, "freq_dim": 256, "num_layers": 30, "moe": null, "attn_head_dim": 128, "eps": 1e-06, "ffn_dim": 4096, "hidden_dim": 1024, "num_heads": 24 } }, "Dataset Scope and Split": { "data.train.expected_dataset_family": null, "data.train.scene_filter": null, "data.train.scene_order": null, "data.train.selected_tasks": null, "data.train.selected_episodes": null, "data.train.selected_episode_ranges": null, "data.train.episode_split": null, "data.train.val_set_proportion": null, "data.train.depth_codec": null, "data.train.shape_meta": null, "data.train.processor.action_output_dim": 14, "data.train.processor.proprio_output_dim": 14, "data.train.processor.use_stepwise_action_norm": false, "data.train.processor.norm_default_mode": "z-score", "data.train.processor.norm_exception_mode": null, "data.train.processor.action_state_transforms": null, "statistics_source": "null/not recorded in config; use saved statistics identity, not a guessed refit", "data.releases": null }, "Windows and Sample Selection": { "data.train.num_frames": 33, "data.train.action_video_freq_ratio": 4, "data.train.global_sample_stride": null, "data.train.video_size": null, "data.train.skip_padding_as_possible": null, "data.train.temporal_sampling": null, "data.val.temporal_sampling": null, "data.train.semantic_compute_mode": null, "data.train.depth_compute_mode": null, "model.semantic.compute_mode": null, "model.depth.compute_mode": null, "data.vae_cache.enabled": null, "data.vae_cache.rgb": null, "data.vae_cache.depth": null, "data.vae_cache.dtype": null }, "Image Augmentation": { "data_augmentation.enabled": null, "data_augmentation.brightness": null, "data_augmentation.contrast": null, "data_augmentation.saturation": null, "data_augmentation.hue": null, "data_augmentation.probability": null, "data_augmentation.depth_source": null, "data_augmentation.geometry": null, "data_augmentation.coarse_dropout": null, "data.train.color_jitter": null, "data.train.geometry_augment": null, "data.train.coarse_dropout": null, "data_augmentation.patchwam": null, "data.train.patchwam_augmentation": null, "conditioning_latent_noise": null, "data.train.augmented_depth_source": null, "data.val.color_jitter": null, "data.val.geometry_augment": null, "data.val.coarse_dropout": null }, "Objectives and Initialization": { "model.loss": { "lambda_video": 1.0, "lambda_action": 1.0, "lambda_semantic": 1.0, "lambda_depth": 1.0 }, "model.semantic.parametrization": "v_pred", "model.semantic.velocity_loss_form": "direct_error", "model.semantic.loss_lambda_semantic": 1.0, "model.depth.loss_lambda_depth": 1.0, "model.video_scheduler": { "train_shift": 5.0, "infer_shift": 5.0, "num_train_timesteps": 1000 }, "model.action_scheduler": { "train_shift": 5.0, "infer_shift": 5.0, "num_train_timesteps": 1000 }, "model.skip_dit_load_from_pretrain": null, "model.action_dit_initializer_filename": null }, "Optimization and Execution": { "num_epochs": null, "max_steps": null, "learning_rate": null, "lr_scheduler_type": null, "lr_conv_ratio": null, "weight_decay": null, "max_grad_norm": null, "enforce_grad_clipping": null, "batch_size": null, "gradient_accumulation_steps": null, "mixed_precision": null, "seed": null, "num_workers": null, "model.mot_checkpoint_mixed_attn": null, "compile_model": null, "compile_mode": null, "eval_every": null, "eval_num_inference_steps": null, "save_every": null, "max_checkpoint_history": null, "save_weights_only": null } }, "activation_gates": { "Depth estimator": "inactive: native GT current and target", "Pointmap bounds": "inactive", "DVAE preprocessing": "active", "Event resampling": "unknown", "CoarseDropout": null, "Video MoE": true, "Action MoE": null, "Separate wrist stream": "inactive" } }, "archived_attention_guide": { "source": "WorldModelSpec applied to archived config, not checkpoint verification", "world_streams": [ "rgb", "depth", "semantic" ], "world_cross_reads": [ { "query": "rgb", "reads": "depth", "key_frames": "none" }, { "query": "rgb", "reads": "semantic", "key_frames": "none" }, { "query": "depth", "reads": "rgb", "key_frames": "none" }, { "query": "depth", "reads": "semantic", "key_frames": "all" }, { "query": "semantic", "reads": "rgb", "key_frames": "none" }, { "query": "semantic", "reads": "depth", "key_frames": "all" } ], "action_reads": [ { "modality": "depth", "key_frames": "all", "spatial_scope": "all" }, { "modality": "semantic", "key_frames": "all", "spatial_scope": "head positions only (top 2/3 of T token grid)" } ], "current_conditions_protected": true, "within_stream_mask": "first_frame_causal", "world_reads_noisy_action": "none", "fusion_layers": "not used", "separate_wrist_condition": false, "wrist_reads_primary_condition": "not applicable", "note": "Direct reads are not total information access: encoder mixing and world paths can carry indirect information. Protected current conditions do not imply strictly autoregressive future frames. World Video blocks are shared; Action DiT parameters are separate. MoE top_k masks weights, not expert computation." } }