{ "created_at": "2026-09-22T13:41:06.181366+00:00", "archive": "/mnt/data/limingleyang/checkpoints", "catalog_sha256": "c7b44f74e6eb166cdd7f78eb837d28eb5dcfbee16860fa1e21d350e697e0c65b", "units": "decimal bytes; 1 GB = 1000000000 bytes", "archive_bytes_before_this_report": 285845345439, "file_category_bytes": { "other": 794615716, ".safetensors": 280365751620, ".pt": 4684978103 }, "safetensors_dtype_bytes": { "BF16": 280357023542, "I64": 1136, "U8": 21470, "F32": 83328 }, "safetensors_stage_bytes": { "sft": 244629968772, "svae": 1057420600, "pretrain": 11101591426, "shared": 23576770822 }, "policy_original_bytes": 1078534785832, "policy_private_component_bytes": { "action_backbone.action_decoder": 6888000, "action_backbone.action_encoder": 6967296, "action_backbone.blocks": 59274313728, "action_backbone.text_embedding": 176332800, "action_backbone.time_embedding": 110272512, "action_backbone.time_projection": 528998400, "proprio_encoder.bias": 86016, "proprio_encoder.weight": 6881280, "video_backbone.dit": 172980217534, "vlm_gate_step": 224, "video_backbone.video_encoder": 3968, "vlm_expert.vlm": 22550560768, "vlm_dino.current_head": 39884800, "vlm_dino.current_queries": 1048576, "vlm_dino.future_head": 39884800, "vlm_dino.future_queries": 1048576 }, "pt_models": [ { "id": "svae/run11", "file": "svae/run11_2sys_scratch_dinov3-svae/private/svae.pt", "file_bytes": 134585325, "top_keys": [ "format_version", "model_config", "state_dict", "metrics", "raw_dim", "num_clips", "world_size", "wallclock_seconds" ], "components": { "format_version": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "model_config": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "state_dict": { "tensors": 82, "tensor_bytes": 132799872, "dtypes": { "torch.float32": 132799872 } }, "metrics": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "raw_dim": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "num_clips": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "world_size": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "wallclock_seconds": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} } } }, { "id": "svae/run23", "file": "svae/run23-2sys-pretrain-dinov3-patch2-pretrain-svae/private/svae.pt", "file_bytes": 132842413, "top_keys": [ "format_version", "model_config", "state_dict", "metrics", "raw_dim", "num_clips", "world_size", "wallclock_seconds", "sampling_fingerprint", "statistics_fingerprint", "validation", "training_steps" ], "components": { "format_version": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "model_config": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "state_dict": { "tensors": 82, "tensor_bytes": 132799872, "dtypes": { "torch.float32": 132799872 } }, "metrics": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "raw_dim": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "num_clips": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "world_size": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "wallclock_seconds": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "sampling_fingerprint": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "statistics_fingerprint": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "validation": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "training_steps": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} } } }, { "id": "svae/run24", "file": "svae/run24_3sys_frozen_scratch_vjepa_svae-svae/private/svae.pt", "file_bytes": 331718125, "top_keys": [ "format_version", "model_config", "state_dict", "metrics", "raw_dim", "num_clips", "world_size", "wallclock_seconds" ], "components": { "format_version": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "model_config": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "state_dict": { "tensors": 82, "tensor_bytes": 329932672, "dtypes": { "torch.float32": 329932672 } }, "metrics": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "raw_dim": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "num_clips": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "world_size": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "wallclock_seconds": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} } } }, { "id": "svae/run26", "file": "svae/run26-future-dit-dinov3-frozen-svae/private/svae.pt", "file_bytes": 134168941, "top_keys": [ "provenance", "validation", "training", "format_version", "model_config", "state_dict", "metrics", "raw_dim", "num_clips", "world_size", "wallclock_seconds", "sampling_fingerprint", "statistics_fingerprint", "training_steps" ], "components": { "provenance": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "validation": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "training": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "format_version": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "model_config": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "state_dict": { "tensors": 82, "tensor_bytes": 132799872, "dtypes": { "torch.float32": 132799872 } }, "metrics": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "raw_dim": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "num_clips": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "world_size": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "wallclock_seconds": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "sampling_fingerprint": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "statistics_fingerprint": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "training_steps": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} } } }, { "id": "svae/run30", "file": "svae/run30_3sys_vjepa_pairwise-svae/private/svae.pt", "file_bytes": 331720749, "top_keys": [ "feature_contract", "artifact_smoke", "format_version", "model_config", "state_dict", "metrics", "raw_dim", "num_clips", "world_size", "wallclock_seconds", "sampling_fingerprint", "statistics_fingerprint", "validation", "training_steps" ], "components": { "feature_contract": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "artifact_smoke": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "format_version": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "model_config": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "state_dict": { "tensors": 82, "tensor_bytes": 329932672, "dtypes": { "torch.float32": 329932672 } }, "metrics": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "raw_dim": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "num_clips": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "world_size": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "wallclock_seconds": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "sampling_fingerprint": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "statistics_fingerprint": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "validation": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "training_steps": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} } } }, { "id": "svae/run31", "file": "svae/run31_3sys_vjepa_future_only-svae/private/svae.pt", "file_bytes": 331720749, "top_keys": [ "feature_contract", "artifact_smoke", "format_version", "model_config", "state_dict", "metrics", "raw_dim", "num_clips", "world_size", "wallclock_seconds", "sampling_fingerprint", "statistics_fingerprint", "validation", "training_steps" ], "components": { "feature_contract": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "artifact_smoke": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "format_version": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "model_config": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "state_dict": { "tensors": 82, "tensor_bytes": 329932672, "dtypes": { "torch.float32": 329932672 } }, "metrics": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "raw_dim": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "num_clips": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "world_size": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "wallclock_seconds": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "sampling_fingerprint": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "statistics_fingerprint": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "validation": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "training_steps": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} } } }, { "id": "svae/run34", "file": "svae/run34_3sys_frozen_scratch_dinov3_svae_temporal-svae/private/svae.pt", "file_bytes": 66867571, "top_keys": [ "format_version", "temporal_contract", "artifact_smoke", "training", "model_config", "state_dict", "metrics", "val_metrics", "raw_dim", "temporal", "latent_affine_calibrated", "latent_calibration", "world_size", "wallclock_seconds" ], "components": { "format_version": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "temporal_contract": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "artifact_smoke": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "training": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "model_config": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "state_dict": { "tensors": 102, "tensor_bytes": 66494976, "dtypes": { "torch.float32": 66494976 } }, "metrics": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "val_metrics": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "raw_dim": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "temporal": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "latent_affine_calibrated": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "latent_calibration": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "world_size": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "wallclock_seconds": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} } } }, { "id": "svae/run35", "file": "svae/run35_2sys_scratch_dinov3_svae_temporal-svae/private/svae.pt", "file_bytes": 66867571, "top_keys": [ "format_version", "temporal_contract", "artifact_smoke", "training", "model_config", "state_dict", "metrics", "val_metrics", "raw_dim", "temporal", "latent_affine_calibrated", "latent_calibration", "world_size", "wallclock_seconds" ], "components": { "format_version": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "temporal_contract": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "artifact_smoke": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "training": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "model_config": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "state_dict": { "tensors": 102, "tensor_bytes": 66494976, "dtypes": { "torch.float32": 66494976 } }, "metrics": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "val_metrics": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "raw_dim": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "temporal": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "latent_affine_calibrated": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "latent_calibration": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "world_size": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "wallclock_seconds": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} } } }, { "id": "svae/run37", "file": "svae/run37-future-dit-vjepa-with-f0-svae/private/svae.pt", "file_bytes": 330581869, "top_keys": [ "provenance", "validation", "training", "format_version", "model_config", "state_dict", "metrics", "raw_dim", "num_clips", "world_size", "wallclock_seconds", "sampling_fingerprint", "statistics_fingerprint", "training_steps" ], "components": { "provenance": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "validation": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "training": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "format_version": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "model_config": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "state_dict": { "tensors": 82, "tensor_bytes": 329932672, "dtypes": { "torch.float32": 329932672 } }, "metrics": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "raw_dim": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "num_clips": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "world_size": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "wallclock_seconds": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "sampling_fingerprint": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "statistics_fingerprint": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "training_steps": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} } } }, { "id": "svae/run38", "file": "svae/run38-future-dit-vjepa-without-f0-svae/private/svae.pt", "file_bytes": 330581869, "top_keys": [ "provenance", "validation", "training", "format_version", "model_config", "state_dict", "metrics", "raw_dim", "num_clips", "world_size", "wallclock_seconds", "sampling_fingerprint", "statistics_fingerprint", "training_steps" ], "components": { "provenance": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "validation": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "training": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "format_version": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "model_config": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "state_dict": { "tensors": 82, "tensor_bytes": 329932672, "dtypes": { "torch.float32": 329932672 } }, "metrics": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "raw_dim": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "num_clips": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "world_size": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "wallclock_seconds": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "sampling_fingerprint": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "statistics_fingerprint": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "training_steps": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} } } }, { "id": "pretrain/run19-encoder", "file": "pretrain/run19_2sys_scratch_videorae_ft-encoder/private/encoder.pt", "file_bytes": 457947487, "top_keys": [ "format", "model", "provenance", "layout" ], "components": { "format": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "model": { "tensors": 240, "tensor_bytes": 457866496, "dtypes": { "torch.float32": 457866496 } }, "provenance": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "layout": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} } } }, { "id": "pretrain/run19-encoder", "file": "pretrain/run19_2sys_scratch_videorae_ft-encoder/private/last.pt", "file_bytes": 2035233015, "top_keys": [ "format", "fingerprint", "contract", "step", "model", "loss", "optimizer", "d_optimizer", "rng", "best", "baseline" ], "components": { "format": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "fingerprint": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "contract": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "step": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "model": { "tensors": 212, "tensor_bytes": 500115724, "dtypes": { "torch.float32": 500115724 } }, "loss": { "tensors": 180, "tensor_bytes": 218510116, "dtypes": { "torch.float32": 218510116 } }, "optimizer": { "tensors": 636, "tensor_bytes": 1000232296, "dtypes": { "torch.float32": 1000232296 } }, "d_optimizer": { "tensors": 417, "tensor_bytes": 315085364, "dtypes": { "torch.float32": 315085364 } }, "rng": { "tensors": 128, "tensor_bytes": 324608, "dtypes": { "torch.uint8": 324608 } }, "best": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} }, "baseline": { "tensors": 0, "tensor_bytes": 0, "dtypes": {} } } } ], "projections": { "all_models_excluding_run19_training_snapshot": { "removed_bytes": 2035233015, "remaining_bytes": 283810112424, "note": "Keeps all SFT, WAM pretrain, independent S-VAE and encoder.pt; no deletion performed." }, "sft_run01_run40": { "entry_count": 40, "referenced_file_count": 729, "referenced_bytes": 269911315653, "stage_bytes": { "sft": 245277124231, "svae": 1057420600, "shared": 23576770822 }, "note": "Existing index and directly referenced weight/asset files counted once. Excludes new catalog/tools/component metadata/docs (small), training initialization provenance packages. Size projection only, no exported package." } }, "extra_private_tensor_dedup_upper_bound": { "bytes": 3020227839, "largest_component_bytes": { "video_backbone.dit.crossattn_proj": 2466275328, "vlm_expert.vlm.language_model": 553843968 }, "basis": "Equal dtype, shape and stored tensor SHA-256 across private files; excludes new layout/header overhead; not implemented." }, "method": "Live file stat, safetensors headers, index tensor SHA-256 and torch.load CPU .pt structure inspection; no weight mutation, no quantization, no new inference or closed-loop evaluation. The full initial JSON transfer was interrupted; complete audit sections and prior successful independent size/dedup query results were combined locally." }