Add files using upload-large-folder tool

README.md

@@ -0,0 +1,43 @@
+# Checkpoint assets
+
+This directory separates fixed upstream assets from checkpoints learned by
+PLOT. Large tensor files are intentionally ignored by Git. Fixed bundle assets
+are recorded in `MANIFEST.json`; operational resume selections are recorded in
+`SELECTED.json`, both with exact size and SHA-256.
+
+```text
+pixel_vae/model.safetensors       required frozen RGB/latent codec for M3 and M4
+m3_backbone/model.safetensors     recommended 2DAction-compatible M3 warm start
+m4_text/text_catalog.json         canonical 476-string catalog
+m4_text/text_cache.safetensors    frozen T5-v1.1-base embeddings for that catalog
+m1/                               selected M1 resume checkpoint and evidence
+m2/                               future selected formal M2 checkpoint
+m3/                               selected M3 resume/identity checkpoints
+m4/                               future selected formal M4 checkpoint
+```
+
+M1 and base M2 do not require external pretrained encoders. M3 can omit its
+backbone warm start, but it cannot train with the current entry point without
+the frozen Pixel VAE. M4 requires the selected formal M3 checkpoint and the same
+Pixel VAE. It reads the cached text tensors directly and does not load T5 during
+training.
+
+The M3 warm start is an initialization rather than a resume: only tensors whose
+names and shapes match the PLOT renderer core are transferred. New voxel,
+resident, appearance, and event-conditioning parameters remain trainable.
+
+## Selected operational resume points
+
+Large files below are ignored by Git. `SELECTED.json` records their immutable
+names, sizes, hashes, and lineage.
+
+- M1 resumes from
+  `m1/m1_flow_2nodes_b32_eval10k_v2/step_001476000.pt` on
+  `zhizhou-avgen-2` after compatibility smoke tests.
+- M3 identity supervision resumes from
+  `m3/m3_entity_reference_stage1/step_0026000.pt` on `js-public` and also
+  requires
+  `m3/player_identity_real_finetune/step_0003300.pt`.
+- M3 step 25,000 is a parent of step 26,000, not the selected current resume
+  point. The aborted identity run logged through step 26,420 but did not save a
+  checkpoint.

SELECTED.json

@@ -0,0 +1,67 @@
+{
+  "schema_version": "plot-selected-checkpoints-v1",
+  "updated_utc": "2026-09-13",
+  "checkpoints": [
+    {
+      "path": "m1/m1_flow_2nodes_b32_eval10k_v2/step_001476000.pt",
+      "role": "selected M1 operational resume checkpoint",
+      "bytes": 5491811243,
+      "sha256": "8f75b7a43f049a5b81b8133b8374749b5915cc1494d4c3b9d14574f9b2073048",
+      "source": "legacy Polis run m1_flow_2nodes_b32_eval10k_v2",
+      "step": 1476000,
+      "code": "legacy source hashes; compatible PLOT migration at 4b6ce14",
+      "target_worktree_commit": "639ee45869e45cc1b943f1e206ee5f74fc261eb6",
+      "validation": "torch.load with mmap succeeded; current MultiViewVoxelDiT strict-load on zhizhou-avgen-2 matched all 288 tensors; AdamW restored all 288 optimizer states at lr 1e-4"
+    },
+    {
+      "path": "m3/m3_entity_reference_stage1/step_0026000.pt",
+      "role": "selected M3 warm start for identity-supervision continuation",
+      "bytes": 1946223273,
+      "sha256": "12b4fc82f29a4c3d41dcc2ed29248115d7b96f29e045ff0b8b074bcaad63eab5",
+      "source": "historical H100 run m3_h100_8gpu_entity_reference_stage1_from25000_to27000_20260914",
+      "step": 26000,
+      "code": "launch provenance reconstructed as 60c83b9; no launch manifest",
+      "target_worktree_commit": "172f278a60a1cf38269c65d85c9a9671704cefbd",
+      "validation": "source and shared-copy SHA-256 match; current Renderer strict-load on js-public at the target worktree commit matched all 513 tensors"
+    },
+    {
+      "path": "m3/player_identity_real_finetune/step_0003300.pt",
+      "role": "frozen player-identity encoder for M3 identity supervision",
+      "bytes": 13350729,
+      "sha256": "6a2119fd6042de738fd729fa409645edbfa72b7729c35667c02fc502b30258b3",
+      "source": "historical H100 run player_identity_real_finetune_h100_5gpu_20260914",
+      "step": 3300,
+      "code": "33270b9",
+      "validation": "source and shared-copy SHA-256 match; current PlayerIdentityEncoder strict-load on js-public matched all 41 tensors"
+    }
+  ],
+  "dependencies": {
+    "m3_validated_train_index": {
+      "path": "derived/m3/validated/train_c65.pt",
+      "bytes": 344160441,
+      "sha256": "d0e483e85949bccdca3899c0c13c7d95e35f8a6b8dc3662aa8aa20d504521b7a"
+    },
+    "m3_validated_val_index": {
+      "path": "derived/m3/validated/val_id_c65.pt",
+      "bytes": 21900351,
+      "sha256": "6fdad62f55921ca5b53bc45c35d564753abfb28d646f4062b43c5dee43133399"
+    },
+    "block_vocabulary": {
+      "path": "derived/common/block_vocabulary.json",
+      "sha256": "ee8e3c89f810d2f6b49a1e07ba168ba187b85623eea168aaa324e8196060c6de",
+      "validation": "shared-cluster and historical-H100 copies match"
+    }
+  },
+  "dataset": {
+    "repository": "xixibuxixi/polis-v1",
+    "local_release": "polis_v1_20260909_360p",
+    "release_json_sha256": "d3a69316575d4e87d91a991a6fe1fadaec0baf7ab31d26f67b782b3f077ad8ba",
+    "train_jsonl_sha256": "47190499ec76ff24bbca5600ca4f1bd39fafec9b87618ac65039bf0109cb95f6",
+    "val_id_jsonl_sha256": "c2e2646dfa52a93cc8d15796e2d0019ffbf14901ead064520dc631c3e4c74b6b",
+    "validation": "shared-cluster and historical-H100 copies match for all three files"
+  },
+  "publishing": {
+    "hugging_face_repo": "xixibuxixi/plot-checkpoints",
+    "status": "pending Hugging Face authentication"
+  }
+}

m1/m1_flow_2nodes_b32_eval10k_v2/source_manifest.json

@@ -0,0 +1,24 @@
+{
+  "protocol": "all S01 memorization, supplied GT cameras, two players, frozen codecs",
+  "objective": "flow",
+  "time_sampling": "original",
+  "count": 46200,
+  "epochs": 17483,
+  "max_steps": 2000000,
+  "eval_every_steps": 10000,
+  "checkpoint_every_steps": 1000,
+  "start_step": 1127000,
+  "effective_batch": 512,
+  "initialize": "/public/0_DATA/2_Avatar/zhizhou_share/rcz/Polis/outputs/m1_multiview_persist_gtcam_fit8_v1/checkpoint_latest.pt",
+  "lr": [
+    0.0001
+  ],
+  "resume": "/public/0_DATA/2_Avatar/zhizhou_share/rcz/Polis/outputs/m1_flow_2nodes_b32_eval10k_v2_control/resume_snapshot.pt",
+  "start_epoch": 7889,
+  "seed": 20260910,
+  "config": {
+    "width": 1024,
+    "depth": 12,
+    "heads": 16
+  }
+}

m1/m1_flow_2nodes_b32_eval10k_v2/step_001476000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f75b7a43f049a5b81b8133b8374749b5915cc1494d4c3b9d14574f9b2073048
+size 5491811243

m3/m3_entity_reference_stage1/source_config.json

@@ -0,0 +1,499 @@
+{
+  "renderer": {
+    "num_block_classes": 402,
+    "item_vocab_size": 12,
+    "input_h": 36,
+    "input_w": 64,
+    "in_channels": 16,
+    "hidden_size": 1024,
+    "depth": 12,
+    "num_heads": 16,
+    "voxel_channels": 32,
+    "condition_dim": 256,
+    "actor_channels": 16,
+    "max_agents": 8,
+    "context_frames": 65,
+    "cache_frames": 64,
+    "block_frames": 8,
+    "gradient_checkpointing": true,
+    "gpu_rasterizer": true,
+    "deep_condition_reinjection": false,
+    "view_aware_appearance": true,
+    "detail_preserving_appearance": false,
+    "entity_reference_attention": true
+  },
+  "training": {
+    "dataset_root": "/data/huangyh/hxh/polis_v1_20260909_360p",
+    "vocabulary": "derived/common/block_vocabulary.json",
+    "pixel_vae": "checkpoints/pixel_vae/model.safetensors",
+    "backbone_checkpoint": null,
+    "warm_start": "/data/huangyh/hxh/Plot/outputs/m3_h100_8gpu_blockcausal8_cache64_full_from17000_to30000_20260913/step_0025000.pt",
+    "output_dir": "/data/huangyh/hxh/Plot/outputs/m3_h100_8gpu_entity_reference_stage1_from25000_to27000_20260914",
+    "window_index": "derived/m3/validated/train_c65.pt",
+    "val_window_index": "derived/m3/validated/val_id_c65.pt",
+    "health_focus_index": null,
+    "health_focus_oversample": 1,
+    "resume": null,
+    "deep_condition_reinjection": false,
+    "view_aware_appearance": true,
+    "detail_preserving_appearance": false,
+    "entity_reference_attention": true,
+    "freeze_base_for_appearance": false,
+    "freeze_base_for_reference": true,
+    "appearance_unfreeze_last_spatial_blocks": 0,
+    "context_frames": 65,
+    "cache_frames": 64,
+    "block_frames": 8,
+    "hidden_size": 1024,
+    "depth": 12,
+    "heads": 16,
+    "voxel_channels": 32,
+    "condition_dim": 256,
+    "actor_channels": 16,
+    "target_views_per_window": 2,
+    "batch_size": 1,
+    "workers": 2,
+    "steps": 27000,
+    "save_every": 500,
+    "validate_every": 500,
+    "visualize_every": 500,
+    "visualization_denoising_steps": 20,
+    "val_batches": 16,
+    "log_every": 20,
+    "gradient_accumulation": 1,
+    "precision": "bf16",
+    "latent_entity_region_upweight": 4.0,
+    "pixel_loss_frames": 4,
+    "entity_pixel_l1_weight": 0.5,
+    "entity_pixel_edge_weight": 0.2,
+    "player_pixel_l1_weight": 1.0,
+    "player_pixel_edge_weight": 0.5,
+    "health_pixel_l1_weight": 1.0,
+    "damaged_health_upweight": 4.0,
+    "lr": 0.0001,
+    "seed": 0,
+    "device": "cuda",
+    "wandb_project": "plot-m3",
+    "wandb_name": "m3-h100-8gpu-entity-reference-stage1-from25000-to27000",
+    "wandb_mode": "online",
+    "checkpoint_staging_dir": "/data/huangyh/hxh/Plot/.checkpoint_staging",
+    "checkpoint_errors": "raise"
+  },
+  "item_vocabulary": {
+    "mcl_core:brick_block": 1,
+    "mcl_core:cobble": 2,
+    "mcl_core:dirt": 3,
+    "mcl_core:glass": 4,
+    "mcl_core:stonebrick": 5,
+    "mcl_core:tree": 6,
+    "mcl_core:wood": 7,
+    "mcl_ocean:sea_lantern": 8,
+    "mcl_tools:pick_iron": 9,
+    "textagent_task:npc_axe": 10,
+    "textagent_task:npc_sword": 11
+  },
+  "class_to_raw": [
+    1,
+    9,
+    13,
+    15,
+    25,
+    26,
+    31,
+    51,
+    52,
+    54,
+    56,
+    57,
+    59,
+    60,
+    61,
+    63,
+    64,
+    65,
+    87,
+    88,
+    89,
+    91,
+    92,
+    93,
+    95,
+    96,
+    97,
+    120,
+    126,
+    156,
+    157,
+    158,
+    160,
+    161,
+    162,
+    164,
+    165,
+    166,
+    188,
+    191,
+    192,
+    193,
+    195,
+    201,
+    203,
+    216,
+    217,
+    218,
+    219,
+    220,
+    223,
+    226,
+    227,
+    228,
+    229,
+    231,
+    232,
+    233,
+    235,
+    237,
+    238,
+    239,
+    240,
+    242,
+    244,
+    245,
+    246,
+    248,
+    249,
+    250,
+    251,
+    252,
+    253,
+    254,
+    258,
+    259,
+    261,
+    262,
+    266,
+    267,
+    270,
+    271,
+    272,
+    273,
+    278,
+    286,
+    287,
+    288,
+    289,
+    290,
+    291,
+    292,
+    293,
+    295,
+    297,
+    299,
+    301,
+    303,
+    305,
+    309,
+    313,
+    314,
+    317,
+    318,
+    319,
+    320,
+    322,
+    329,
+    330,
+    331,
+    332,
+    333,
+    334,
+    335,
+    336,
+    337,
+    338,
+    339,
+    341,
+    358,
+    359,
+    360,
+    362,
+    363,
+    382,
+    384,
+    394,
+    418,
+    421,
+    423,
+    424,
+    429,
+    434,
+    435,
+    437,
+    438,
+    450,
+    453,
+    456,
+    462,
+    464,
+    468,
+    474,
+    476,
+    477,
+    486,
+    487,
+    488,
+    489,
+    490,
+    491,
+    493,
+    494,
+    495,
+    496,
+    497,
+    499,
+    500,
+    501,
+    502,
+    503,
+    505,
+    506,
+    507,
+    508,
+    509,
+    511,
+    512,
+    513,
+    514,
+    515,
+    517,
+    518,
+    519,
+    520,
+    521,
+    523,
+    524,
+    525,
+    526,
+    528,
+    530,
+    532,
+    533,
+    534,
+    536,
+    566,
+    568,
+    571,
+    574,
+    577,
+    717,
+    720,
+    723,
+    726,
+    729,
+    755,
+    758,
+    759,
+    850,
+    852,
+    853,
+    854,
+    868,
+    869,
+    871,
+    872,
+    878,
+    880,
+    883,
+    884,
+    885,
+    892,
+    893,
+    894,
+    913,
+    914,
+    915,
+    916,
+    919,
+    931,
+    932,
+    934,
+    955,
+    956,
+    973,
+    974,
+    976,
+    1003,
+    1004,
+    1012,
+    1039,
+    1045,
+    1116,
+    1117,
+    1118,
+    1122,
+    1129,
+    1154,
+    1161,
+    1170,
+    1171,
+    1181,
+    1191,
+    1196,
+    1221,
+    1229,
+    1231,
+    1281,
+    1285,
+    1296,
+    1305,
+    1409,
+    1411,
+    1412,
+    1414,
+    1431,
+    1432,
+    1464,
+    1465,
+    1467,
+    1471,
+    1472,
+    1473,
+    1474,
+    1495,
+    1496,
+    1513,
+    1514,
+    1527,
+    1528,
+    1536,
+    1537,
+    1540,
+    1543,
+    1546,
+    1549,
+    1551,
+    1553,
+    1556,
+    1557,
+    1558,
+    1560,
+    1601,
+    1607,
+    1619,
+    1764,
+    1765,
+    1770,
+    1778,
+    1796,
+    1802,
+    1803,
+    1844,
+    1850,
+    1859,
+    1915,
+    1942,
+    1946,
+    1947,
+    1948,
+    1950,
+    1951,
+    1995,
+    2030,
+    2031,
+    2032,
+    2034,
+    2035,
+    2037,
+    2038,
+    2040,
+    2041,
+    2045,
+    2053,
+    2062,
+    2066,
+    2077,
+    2081,
+    2090,
+    2098,
+    2106,
+    2107,
+    2108,
+    2119,
+    2122,
+    2126,
+    2128,
+    2134,
+    2136,
+    2157,
+    2158,
+    2159,
+    2160,
+    2161,
+    2162,
+    2163,
+    2164,
+    2165,
+    2166,
+    2167,
+    2168,
+    2169,
+    2170,
+    2171,
+    2172,
+    2173,
+    2174,
+    2175,
+    2176,
+    2177,
+    2178,
+    2179,
+    2180,
+    2181,
+    2182,
+    2183,
+    2185,
+    2187,
+    2188,
+    2189,
+    2190,
+    2191,
+    2192,
+    2193,
+    2194,
+    2197,
+    2199,
+    2201,
+    2209,
+    2212,
+    2213,
+    2214,
+    2221,
+    2273,
+    2274,
+    2280,
+    2281,
+    2282,
+    2283,
+    2284,
+    2285,
+    2286,
+    2287,
+    2288,
+    2289,
+    2290,
+    2291,
+    2292,
+    2293,
+    2294,
+    2328,
+    2329,
+    2346,
+    2350,
+    2416,
+    2420,
+    2424,
+    2428,
+    2432,
+    2436,
+    2456
+  ]
+}

m3/m3_entity_reference_stage1/step_0026000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:12b4fc82f29a4c3d41dcc2ed29248115d7b96f29e045ff0b8b074bcaad63eab5
+size 1946223273

m3/m3_entity_reference_stage1/validation.jsonl

@@ -0,0 +1,2 @@
+{"step": 25500, "total_loss": 0.3414022624492645, "flow_loss": 0.29731646180152893, "auxiliary_loss": 0.0440857969224453, "entity_pixel_l1": 0.0094954464584589, "entity_pixel_edge": 0.00711657851934433, "player_pixel_l1": 0.011743898503482342, "player_pixel_edge": 0.008640612475574017, "health_pixel_l1": 0.021850550547242165}
+{"step": 26000, "total_loss": 0.3411499857902527, "flow_loss": 0.2971159815788269, "auxiliary_loss": 0.04403400421142578, "entity_pixel_l1": 0.009468615055084229, "entity_pixel_edge": 0.00711360527202487, "player_pixel_l1": 0.011712109670042992, "player_pixel_edge": 0.008632497861981392, "health_pixel_l1": 0.021848618984222412}

m3/m3_entity_reference_stage1/wandb_run.json

@@ -0,0 +1,5 @@
+{
+  "id": "7kbesehq",
+  "url": "https://wandb.ai/ckx23-tsinghua-university/plot-m3/runs/7kbesehq",
+  "project": "plot-m3"
+}

m3/player_identity_real_finetune/real_rendered_eval_step3300.json

@@ -0,0 +1,7 @@
+{
+  "checkpoint": "outputs/player_identity_real_finetune_h100_5gpu_20260914/step_0003300.pt",
+  "samples": 160,
+  "retrieval_at_1": 0.731249988079071,
+  "positive_similarity": 0.6846246123313904,
+  "hard_negative_similarity": 0.6141237616539001
+}

m3/player_identity_real_finetune/source_config.json

@@ -0,0 +1,22 @@
+{
+  "dataset_root": "/data/huangyh/hxh/polis_v1_20260909_360p",
+  "window_index": "derived/m3/validated/train_c65.pt",
+  "val_window_index": null,
+  "output_dir": "/data/huangyh/hxh/Plot/outputs/player_identity_real_finetune_h100_5gpu_20260914",
+  "steps": 3300,
+  "batch_size": 2,
+  "workers": 2,
+  "lr": 0.0001,
+  "temperature": 0.07,
+  "min_player_pixels": 96,
+  "canonical_only": false,
+  "save_every": 100,
+  "validate_every": 100,
+  "val_batches": 50,
+  "log_every": 10,
+  "resume": "/data/huangyh/hxh/Plot/outputs/player_identity_canonical_h100_5gpu_20260914/step_0003000.pt",
+  "seed": 0,
+  "wandb_project": "plot-m3",
+  "wandb_name": "player-identity-real-finetune-h100-5gpu-20260914",
+  "wandb_mode": "online"
+}

m3/player_identity_real_finetune/step_0003300.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a2119fd6042de738fd729fa409645edbfa72b7729c35667c02fc502b30258b3
+size 13350729

m3/player_identity_real_finetune/wandb_run.json

@@ -0,0 +1,4 @@
+{
+  "id": "b9dsjly9",
+  "url": "https://wandb.ai/ckx23-tsinghua-university/plot-m3/runs/b9dsjly9"
+}