sheaf-admm-icml2026-executor / SCIENCE-SPEC.yaml
Mindcraft's picture
Publish frozen reproduction executor image source
17d5066 verified
Raw
History Blame Contribute Delete
46.7 kB
{
"authorities": {
"approved_protocol_sha256": "f4904e05e8abd051281926abec6774c6fe39b84160367f0bfde806b0669e0c2b",
"base_image": "python:3.12-slim@sha256:cab2dbf575e971934a81e4622f5aba17aa7929719bd7e31033a3a83b97fd0464",
"canonical_claims_sha256": "26ca2fa3697061cb71a91a0f687ab5ce99e908b7b50c009ee2efcc645695d489",
"challenge_dataset_revision": "81166abbeb76e5f79ff87e51061b5a0306507203",
"challenge_space_revision": "5bbcad2e9a7e8a7479f3563ac1fc6c768d4bb050",
"deadline_utc": "2026-08-03T11:59:00Z",
"dependency_lock_sha256": "e9dd209b20905a9a553c30ab1bab0259d2c665ee7805942744d9c5f198610bd7",
"execution_bucket": "Mindcraft/sheaf-admm-icml2026-runs",
"execution_image_space": "Mindcraft/sheaf-admm-icml2026-executor",
"input_bucket": "Mindcraft/sheaf-admm-icml2026-inputs",
"paper_sha256": "95d6de2011cdeaf4eeba6c7cc320146a530183da7f045457227cab3165b83a70",
"poster_commit": "e503c399b5427ca6cb712ccb080a758e9c19cf23",
"space_id": "Mindcraft/repro-learning-multi-agent-coordination-via-sheaf-admm",
"sudoku_dataset_revision": "4d5aa527a9fb9aacca0b0d5b8b77d569fa9afcaa",
"trace_mode": "none",
"trackio_logbook_autonote": 0,
"trackio_version": "0.33.0",
"trackio_wheel_sha256": "277340507ac46c02c06900c1d680129bdb528223c8110b0b6bc9326bb9f0891d",
"upstream_commit": "1e2b5d648361802234348b0b1a7fb3a222128e7d"
},
"budget_limits": {
"billing_quantum": "ceil(rate*1e6*ceil(seconds/60)/60)",
"gpu_retry_reserve_micro_usd": 10000000,
"minimum_unspent_balance_micro_usd": 10000000,
"normal_cap_micro_usd": 75000000
},
"data_rules": {
"c5": {
"K": 100,
"exact_figure6_protocol": "unreleased",
"examples_per_size": 1000,
"generator_seed_formula": "21005000+n",
"min_path_length_formula": "3*(n-1)/2",
"n19_replacements": true,
"overlap_rejection": "canonical wall/start/goal/path against training",
"realized_minimum": [
27,
33,
39,
45,
51,
57
],
"sizes": [
19,
23,
27,
31,
35,
39
],
"test_augmentation": false
},
"maze": {
"builder": "released_deterministic_DFS",
"height": 19,
"min_path_length": 18,
"ood_sizes": true,
"test_size": 1000,
"train_size": 10000,
"width": 19
},
"mnist": {
"agent_count": 81,
"clean": true,
"drop_agents": 24,
"drop_effect": [
"zero_3x3_pixels",
"remove_incident_sheaf_edges",
"exclude_removed_agents_from_vote"
],
"drop_fraction_label": "30%",
"mask_derivation": [
"dataset_revision",
"example_id",
"condition",
"master_seed"
],
"master_seed": 21005300,
"padding_pixels": 16,
"tier": "B_target_informed_nonconfirmatory"
},
"sudoku": {
"revision": "4d5aa527a9fb9aacca0b0d5b8b77d569fa9afcaa",
"source": "Ritvik19/Sudoku-Dataset",
"test_rows": [
50000,
52000
],
"test_split": "test_hard",
"train_augmentation": "eight_way",
"train_rows": [
0,
50000
]
}
},
"evaluators": {
"C5-EVAL-2X-GENERALIZATION-A": {
"physical_jobs": 1,
"shard": "deterministic_lpt_A"
},
"C5-EVAL-2X-GENERALIZATION-B": {
"physical_jobs": 1,
"shard": "deterministic_lpt_B"
},
"MAZE-EVAL-C2-C4B": {
"physical_jobs": 1,
"units": [
"imported_default_3",
"mpnn84_3",
"quadratic_3"
]
},
"MNIST-EVAL-C3": {
"conditions": [
"clean",
"pad16",
"drop30"
],
"physical_jobs": 1,
"units": [
"imported_sheaf_3",
"cnn_3"
]
},
"SUD-EVAL-C1-C4A": {
"physical_jobs": 1,
"units": [
"imported_sheaf_3",
"mpnn225_3",
"identity_3"
]
}
},
"failure_classes": [
"INVALID_INPUT",
"INCOMPLETE_OUTPUT",
"HASH_MISMATCH",
"CONFIG_MISMATCH",
"CAPABILITY_VIOLATION",
"READINESS_TIMEOUT",
"HEARTBEAT_TIMEOUT",
"INFRASTRUCTURE",
"CODE_PARITY",
"DATA_ASSERTION",
"BUDGET_GATE",
"DEADLINE_GATE",
"PRIVACY_FINDING",
"AMBIGUOUS_SUBMISSION"
],
"fixture_rules": {
"disjoint_from_final_generators": true,
"id_prefix": "smoke-",
"maze_seed_formula": "91005200+n",
"mnist_seed": 91005301,
"purpose": [
"compile",
"memory",
"timing",
"serialization",
"one_step_gradient"
],
"sudoku_seed": 91005101,
"verdict_metrics_forbidden": true
},
"format": 1,
"hardware_routes": {
"maze_c5": {
"fallback": {
"flavor": "l40sx1",
"usd_per_hour": "1.80"
},
"primary": {
"flavor": "l4x1",
"usd_per_hour": "0.80"
}
},
"mnist": {
"fallback": {
"flavor": "l40sx1",
"usd_per_hour": "1.80"
},
"primary": {
"flavor": "l4x1",
"usd_per_hour": "0.80"
}
},
"sudoku": {
"fallback": {
"flavor": "h200",
"usd_per_hour": "5.00"
},
"primary": {
"flavor": "a100-large",
"usd_per_hour": "2.50"
}
}
},
"imports": [
{
"checkpoint_sha256": "4444ca900b84911f778ade7fb4c682853308fac6008748ff644f5193122951b8",
"config_sha256": "831675683b8ca586616be95d7377a2ed1da776a04cc9457b5b4f855c5598e74a",
"ema_decay": 0.999,
"final_epoch": 19,
"neutral_alias": "mnist-sheaf-seed-42",
"seed": 42,
"task": "mnist"
},
{
"checkpoint_sha256": "34b693aac81e1ab880324726eb6db80e35e0683f3c997a2d082acd5c9393a5f6",
"config_sha256": "c1f902d7509c5a2dbd0130757b55607b51016e2a9039392da4b09b79c694c978",
"ema_decay": 0.999,
"final_epoch": 19,
"neutral_alias": "mnist-sheaf-seed-123",
"seed": 123,
"task": "mnist"
},
{
"checkpoint_sha256": "f35206c24e79df75878e330ff01ff489ec29f82ae11eb3de04f401e2e4d2a3e5",
"config_sha256": "3a3c11e55551b9b820696abb7cf28ea94d4295e022c5da9525adfb802ee77d17",
"ema_decay": 0.999,
"final_epoch": 19,
"neutral_alias": "mnist-sheaf-seed-456",
"seed": 456,
"task": "mnist"
},
{
"checkpoint_sha256": "7a8295f5af4f52c2ec5963fa35e7e3906ebce5ea433551bfc9017e4a9825de9c",
"config_sha256": "6e93a94469a1cec98dd2dc3ff9df91ce613ad99d18a5cae0388b8a42a610595a",
"ema_decay": 0.999,
"final_epoch": 49,
"neutral_alias": "maze-sheaf-seed-42",
"seed": 42,
"task": "maze"
},
{
"checkpoint_sha256": "bd1369916039e3984911951217c7ae083efe6696f06f6cec9f2f5c009187a6f5",
"config_sha256": "8621e9687a39a5f28e5178561c4a6da9dbff09595dbc1340e1e97710c8af82d5",
"ema_decay": 0.999,
"final_epoch": 49,
"neutral_alias": "maze-sheaf-seed-123",
"seed": 123,
"task": "maze"
},
{
"checkpoint_sha256": "a61191e0a3ec14aa245acc4b566dc7dd6caafba69365f9aa347dc2f906e95f38",
"config_sha256": "0566b34b360220bcc8e32cea6236884a9b7a693c68ca1cdc56861ab95ec7da13",
"ema_decay": 0.999,
"final_epoch": 49,
"neutral_alias": "maze-sheaf-seed-456",
"seed": 456,
"task": "maze"
},
{
"checkpoint_sha256": "53fb0f4c287bb419247956a3ab4492acb42cde6a3c3335ed1c18694c705d163e",
"config_sha256": "66ac9154bb1a15483211e0e8458cea4250181fdabc864c1748f89561535621fb",
"ema_decay": 0.999,
"final_epoch": 9,
"neutral_alias": "sudoku-sheaf-seed-42",
"seed": 42,
"task": "sudoku"
},
{
"checkpoint_sha256": "a16e792b46aeca8ca6dc9d9323217328ecf43e0ce32022870c5626186997511f",
"config_sha256": "f080a4055a676a5094204fc78de5e2f57c07dd2f67af8c55cf7615dcb818d47d",
"ema_decay": 0.999,
"final_epoch": 9,
"neutral_alias": "sudoku-sheaf-seed-123",
"seed": 123,
"task": "sudoku"
},
{
"checkpoint_sha256": "7a6937eedf75f553965c21b5dda5c4da8164489125159f6fc8465e9159b2b3af",
"config_sha256": "1649e4d4e047bcceefb9b13ebee6fa830eeac668445714457a2ed258563a9625",
"ema_decay": 0.999,
"final_epoch": 9,
"neutral_alias": "sudoku-sheaf-seed-456",
"seed": 456,
"task": "sudoku"
}
],
"lifecycle": {
"CPU_CANARY": {
"science_freeze_sha256": "NOT_APPLICABLE",
"science_spec_sha256": "NOT_APPLICABLE"
},
"CPU_IMPORT": {
"science_freeze_sha256": "NOT_APPLICABLE",
"science_spec_sha256": "EXACT"
},
"GPU_SMOKE": {
"science_freeze_sha256": "NOT_APPLICABLE",
"science_spec_sha256": "EXACT"
},
"SCIENTIFIC_EVAL": {
"control_mount": "/repro-control read-only",
"science_freeze_sha256": "EXACT",
"science_spec_sha256": "EXACT"
},
"SCIENTIFIC_TRAIN": {
"control_mount": "/repro-control read-only",
"science_freeze_sha256": "EXACT",
"science_spec_sha256": "EXACT"
}
},
"metrics": {
"c1": "Sudoku exact puzzle accuracy and parameter count",
"c2": "Maze exact puzzle accuracy and per-vertex latent dimension ratio",
"c3": "MNIST classification accuracy by condition",
"c4": "Sudoku exact puzzle accuracy and Maze exact puzzle accuracy",
"c5": "Maze exact puzzle accuracy by size",
"replication_unit": "training_seed",
"report": [
"every_seed",
"mean",
"sample_standard_deviation",
"student_t_interval",
"paper_value",
"unrounded_delta"
],
"same_seed_comparisons": "paired",
"t_critical": {
"90_df2": 2.919985580355516,
"95_df2": 4.302652729911275
}
},
"outcomes": {},
"protocol_status": "frozen_pre_mutation",
"reducers": {
"c1": {
"negative": "Sheaf_mean<85 or MPNN_mean>20 or U95(paired_gap)<=60",
"parameter_match": "relative_count_mismatch<=0.05",
"support": "Sheaf_mean>=85 and MPNN_mean<=20 and L95(paired_gap)>60"
},
"c2": {
"negative": "either mean<90 or interval wholly beyond equivalence bounds",
"ratio": "84/10=8.4x per-vertex latent dimension",
"support": "both means>=95 and paired_90_interval within [-2,2]"
},
"c3": {
"adequacy": "each mean>=98.5 and each seed>=98.0",
"negative": "either upper interval<=threshold",
"required_suffix": "DROPOUT_SEMANTICS_TARGET_SELECTED",
"support": "L95(pad_gap)>20 and L95(drop_gap)>10"
},
"c4_maze": {
"falsifies_collapse": "default_mean>=90 and quadratic_mean>=90 and U95(default-quadratic)<20",
"required_suffix": "PROMPT_MISSTATES_PAPER_TABLE",
"supports_direction": "default_mean>=90 and quadratic_mean<=60 and L95(default-quadratic)>20"
},
"c4_sudoku": {
"negative": "learned_mean<85 or identity_mean>15 or U95(gap)<=60",
"support": "learned_mean>=85 and identity_mean<=15 and L95(gap)>60"
},
"c5": {
"negative": "any mean<95",
"required_suffix": "INCONCLUSIVE_EXACT_DENSE_FIGURE6_CONFIG_UNRELEASED",
"support": "all six three-seed means>=95"
},
"precedence": [
"invalid_or_incomplete",
"supported_or_negative",
"ambiguous_or_inconclusive"
]
},
"registered_configs": {
"C1-SUD-MPNN225-123": {
"canonical_sha256": "703fe3f833c3b2910a92783710faee4893ddb333d008b4784288171e78fa56f5",
"config": {
"data": {
"dir": "/data/train/sudoku_easy",
"loader": "puzzle",
"train_split": "train",
"val_splits": []
},
"dtype": "float32",
"model": {
"comm_norm_type": "layernorm",
"d_e": 32,
"d_v": 225,
"dec_hidden_dims": [
256
],
"decoder_arch": "sudoku",
"enc_d_model": 128,
"enc_num_blocks": 2,
"encoder_arch": "sudoku",
"mpnn_aggregation": "max",
"mpnn_edge_type_mode": "slot",
"mpnn_graph_readout": "per_node",
"mpnn_message_dim": 32,
"num_classes": 10,
"num_directions": 9
},
"model_type": "mpnn",
"task": "sudoku",
"task_cfg": {},
"training": {
"K_eval": 100,
"K_train": 40,
"batch_size": 128,
"ema_decay": 0.999,
"epochs": 10,
"exit_on_nan": true,
"grad_clip": 1.0,
"loss_window": 4,
"lr": 0.0017,
"mpnn_eval_rounds": 50,
"mpnn_train_rounds": 20,
"seed": 123,
"train_iters_dist": "fixed",
"train_iters_min": 15,
"val_interval": 1,
"warmup_steps": 200,
"weight_decay": 1e-07
},
"wandb": {
"entity": null,
"group": null,
"mode": "disabled",
"name": null,
"project": "sheaf-admm",
"tags": []
}
},
"file": "control/registered-configs/C1-SUD-MPNN225-123.json"
},
"C1-SUD-MPNN225-42": {
"canonical_sha256": "a0a0fad0c015647c5b03f31542c7f01fd2648ad04c19360b82d6dc3e060e3af2",
"config": {
"data": {
"dir": "/data/train/sudoku_easy",
"loader": "puzzle",
"train_split": "train",
"val_splits": []
},
"dtype": "float32",
"model": {
"comm_norm_type": "layernorm",
"d_e": 32,
"d_v": 225,
"dec_hidden_dims": [
256
],
"decoder_arch": "sudoku",
"enc_d_model": 128,
"enc_num_blocks": 2,
"encoder_arch": "sudoku",
"mpnn_aggregation": "max",
"mpnn_edge_type_mode": "slot",
"mpnn_graph_readout": "per_node",
"mpnn_message_dim": 32,
"num_classes": 10,
"num_directions": 9
},
"model_type": "mpnn",
"task": "sudoku",
"task_cfg": {},
"training": {
"K_eval": 100,
"K_train": 40,
"batch_size": 128,
"ema_decay": 0.999,
"epochs": 10,
"exit_on_nan": true,
"grad_clip": 1.0,
"loss_window": 4,
"lr": 0.0017,
"mpnn_eval_rounds": 50,
"mpnn_train_rounds": 20,
"seed": 42,
"train_iters_dist": "fixed",
"train_iters_min": 15,
"val_interval": 1,
"warmup_steps": 200,
"weight_decay": 1e-07
},
"wandb": {
"entity": null,
"group": null,
"mode": "disabled",
"name": null,
"project": "sheaf-admm",
"tags": []
}
},
"file": "control/registered-configs/C1-SUD-MPNN225-42.json"
},
"C1-SUD-MPNN225-456": {
"canonical_sha256": "b1e99a06aafa6fa0255a8c4c4141b5e7da37a72f21e0388fe10ab45528454a44",
"config": {
"data": {
"dir": "/data/train/sudoku_easy",
"loader": "puzzle",
"train_split": "train",
"val_splits": []
},
"dtype": "float32",
"model": {
"comm_norm_type": "layernorm",
"d_e": 32,
"d_v": 225,
"dec_hidden_dims": [
256
],
"decoder_arch": "sudoku",
"enc_d_model": 128,
"enc_num_blocks": 2,
"encoder_arch": "sudoku",
"mpnn_aggregation": "max",
"mpnn_edge_type_mode": "slot",
"mpnn_graph_readout": "per_node",
"mpnn_message_dim": 32,
"num_classes": 10,
"num_directions": 9
},
"model_type": "mpnn",
"task": "sudoku",
"task_cfg": {},
"training": {
"K_eval": 100,
"K_train": 40,
"batch_size": 128,
"ema_decay": 0.999,
"epochs": 10,
"exit_on_nan": true,
"grad_clip": 1.0,
"loss_window": 4,
"lr": 0.0017,
"mpnn_eval_rounds": 50,
"mpnn_train_rounds": 20,
"seed": 456,
"train_iters_dist": "fixed",
"train_iters_min": 15,
"val_interval": 1,
"warmup_steps": 200,
"weight_decay": 1e-07
},
"wandb": {
"entity": null,
"group": null,
"mode": "disabled",
"name": null,
"project": "sheaf-admm",
"tags": []
}
},
"file": "control/registered-configs/C1-SUD-MPNN225-456.json"
},
"C2-MAZE-MPNN84-123": {
"canonical_sha256": "545effae66a4ab9a6c802d32eaa28b06bf664a3618cac556aadd42a39285b40b",
"config": {
"data": {
"dir": "/data/train/maze_std3_19px_10k",
"loader": "puzzle",
"train_split": "train",
"val_splits": []
},
"dtype": "float32",
"model": {
"comm_norm_type": "layernorm",
"d_e": 42,
"d_v": 84,
"dec_hidden_dim": 256,
"decoder_arch": "mlp_concat_v2",
"enc_hidden_dim": 256,
"encoder_arch": "mlp_v2",
"mpnn_aggregation": "max",
"mpnn_edge_type_mode": "spatial",
"mpnn_graph_readout": "per_node",
"mpnn_message_dim": 42,
"num_classes": 6,
"num_directions": 8
},
"model_type": "mpnn",
"task": "maze",
"task_cfg": {
"connectivity": 8,
"num_classes": 6,
"patch_size": 3,
"stride": 2
},
"training": {
"K_eval": 100,
"K_train": 40,
"batch_size": 128,
"ema_decay": 0.999,
"epochs": 50,
"exit_on_nan": true,
"grad_clip": 1.0,
"loss_window": 4,
"lr": 0.0003,
"mpnn_eval_rounds": 100,
"mpnn_train_rounds": 40,
"seed": 123,
"train_iters_dist": "fixed",
"train_iters_min": 15,
"val_interval": 5,
"warmup_steps": 200,
"weight_decay": 1e-06
},
"wandb": {
"entity": null,
"group": null,
"mode": "disabled",
"name": null,
"project": "sheaf-admm",
"tags": []
}
},
"file": "control/registered-configs/C2-MAZE-MPNN84-123.json"
},
"C2-MAZE-MPNN84-42": {
"canonical_sha256": "f826d5b2c392645ceabd97ed219310e4c8dd8278dce3c935b893ef6ec67c598d",
"config": {
"data": {
"dir": "/data/train/maze_std3_19px_10k",
"loader": "puzzle",
"train_split": "train",
"val_splits": []
},
"dtype": "float32",
"model": {
"comm_norm_type": "layernorm",
"d_e": 42,
"d_v": 84,
"dec_hidden_dim": 256,
"decoder_arch": "mlp_concat_v2",
"enc_hidden_dim": 256,
"encoder_arch": "mlp_v2",
"mpnn_aggregation": "max",
"mpnn_edge_type_mode": "spatial",
"mpnn_graph_readout": "per_node",
"mpnn_message_dim": 42,
"num_classes": 6,
"num_directions": 8
},
"model_type": "mpnn",
"task": "maze",
"task_cfg": {
"connectivity": 8,
"num_classes": 6,
"patch_size": 3,
"stride": 2
},
"training": {
"K_eval": 100,
"K_train": 40,
"batch_size": 128,
"ema_decay": 0.999,
"epochs": 50,
"exit_on_nan": true,
"grad_clip": 1.0,
"loss_window": 4,
"lr": 0.0003,
"mpnn_eval_rounds": 100,
"mpnn_train_rounds": 40,
"seed": 42,
"train_iters_dist": "fixed",
"train_iters_min": 15,
"val_interval": 5,
"warmup_steps": 200,
"weight_decay": 1e-06
},
"wandb": {
"entity": null,
"group": null,
"mode": "disabled",
"name": null,
"project": "sheaf-admm",
"tags": []
}
},
"file": "control/registered-configs/C2-MAZE-MPNN84-42.json"
},
"C2-MAZE-MPNN84-456": {
"canonical_sha256": "2fb6db8a94a03033c21b698aeb700ebf2013abf803e4d34e23a778e3a5ab8d09",
"config": {
"data": {
"dir": "/data/train/maze_std3_19px_10k",
"loader": "puzzle",
"train_split": "train",
"val_splits": []
},
"dtype": "float32",
"model": {
"comm_norm_type": "layernorm",
"d_e": 42,
"d_v": 84,
"dec_hidden_dim": 256,
"decoder_arch": "mlp_concat_v2",
"enc_hidden_dim": 256,
"encoder_arch": "mlp_v2",
"mpnn_aggregation": "max",
"mpnn_edge_type_mode": "spatial",
"mpnn_graph_readout": "per_node",
"mpnn_message_dim": 42,
"num_classes": 6,
"num_directions": 8
},
"model_type": "mpnn",
"task": "maze",
"task_cfg": {
"connectivity": 8,
"num_classes": 6,
"patch_size": 3,
"stride": 2
},
"training": {
"K_eval": 100,
"K_train": 40,
"batch_size": 128,
"ema_decay": 0.999,
"epochs": 50,
"exit_on_nan": true,
"grad_clip": 1.0,
"loss_window": 4,
"lr": 0.0003,
"mpnn_eval_rounds": 100,
"mpnn_train_rounds": 40,
"seed": 456,
"train_iters_dist": "fixed",
"train_iters_min": 15,
"val_interval": 5,
"warmup_steps": 200,
"weight_decay": 1e-06
},
"wandb": {
"entity": null,
"group": null,
"mode": "disabled",
"name": null,
"project": "sheaf-admm",
"tags": []
}
},
"file": "control/registered-configs/C2-MAZE-MPNN84-456.json"
},
"C3-MNIST-CNN-123": {
"canonical_sha256": "528044762e54a8f02036c84c34836a857518c98b5aef2fec60d5352b04d89ef8",
"config": {
"data": {
"augmentation": false,
"dir": "/data/train/mnist",
"examples": 60000,
"input_range": [
0,
1
],
"loader": "image",
"normalization": false,
"train_split": "train",
"val_splits": []
},
"dtype": "float32",
"model": {
"layout": "NHWC/HWIO/NHWC",
"parameter_count": 65642,
"post_pool_sizes": [
14,
30
],
"pre_pool_sizes": [
28,
60
]
},
"model_type": "mnist_cnn_repro",
"task": "mnist",
"task_cfg": {},
"training": {
"batch_size": 128,
"betas": [
0.9,
0.999
],
"ema_decay": 0.999,
"epoch_permutation": "fold_in(PRNGKey(seed),epoch)",
"epochs": 20,
"epsilon": 1e-08,
"final_batch": "pad_to_128_mask_normalize",
"grad_clip": 1.0,
"lr": 0.001,
"schedule": "linear_to_task_lr_then_constant",
"seed": 123,
"selection": "fixed_final_epoch",
"warmup_steps": 200,
"weight_decay": 1e-07
},
"wandb": {
"mode": "disabled"
}
},
"file": "control/registered-configs/C3-MNIST-CNN-123.json"
},
"C3-MNIST-CNN-42": {
"canonical_sha256": "442cbeddd56b3eba9dc239e84403285d135bca4868d58a315fe67fcd97b1f4f6",
"config": {
"data": {
"augmentation": false,
"dir": "/data/train/mnist",
"examples": 60000,
"input_range": [
0,
1
],
"loader": "image",
"normalization": false,
"train_split": "train",
"val_splits": []
},
"dtype": "float32",
"model": {
"layout": "NHWC/HWIO/NHWC",
"parameter_count": 65642,
"post_pool_sizes": [
14,
30
],
"pre_pool_sizes": [
28,
60
]
},
"model_type": "mnist_cnn_repro",
"task": "mnist",
"task_cfg": {},
"training": {
"batch_size": 128,
"betas": [
0.9,
0.999
],
"ema_decay": 0.999,
"epoch_permutation": "fold_in(PRNGKey(seed),epoch)",
"epochs": 20,
"epsilon": 1e-08,
"final_batch": "pad_to_128_mask_normalize",
"grad_clip": 1.0,
"lr": 0.001,
"schedule": "linear_to_task_lr_then_constant",
"seed": 42,
"selection": "fixed_final_epoch",
"warmup_steps": 200,
"weight_decay": 1e-07
},
"wandb": {
"mode": "disabled"
}
},
"file": "control/registered-configs/C3-MNIST-CNN-42.json"
},
"C3-MNIST-CNN-456": {
"canonical_sha256": "2ba04eca4a4bec45d8cadb2c2363efe34b0ab731698ac12948f82c5ef288a099",
"config": {
"data": {
"augmentation": false,
"dir": "/data/train/mnist",
"examples": 60000,
"input_range": [
0,
1
],
"loader": "image",
"normalization": false,
"train_split": "train",
"val_splits": []
},
"dtype": "float32",
"model": {
"layout": "NHWC/HWIO/NHWC",
"parameter_count": 65642,
"post_pool_sizes": [
14,
30
],
"pre_pool_sizes": [
28,
60
]
},
"model_type": "mnist_cnn_repro",
"task": "mnist",
"task_cfg": {},
"training": {
"batch_size": 128,
"betas": [
0.9,
0.999
],
"ema_decay": 0.999,
"epoch_permutation": "fold_in(PRNGKey(seed),epoch)",
"epochs": 20,
"epsilon": 1e-08,
"final_batch": "pad_to_128_mask_normalize",
"grad_clip": 1.0,
"lr": 0.001,
"schedule": "linear_to_task_lr_then_constant",
"seed": 456,
"selection": "fixed_final_epoch",
"warmup_steps": 200,
"weight_decay": 1e-07
},
"wandb": {
"mode": "disabled"
}
},
"file": "control/registered-configs/C3-MNIST-CNN-456.json"
},
"C4-MAZE-QUADRATIC-123": {
"canonical_sha256": "5f7b04835bc32b012a5e2d5e12e2b1512c3c95343ac7d8ac97a77a46178f723e",
"config": {
"data": {
"dir": "/data/train/maze_std3_19px_10k",
"loader": "puzzle",
"train_split": "train",
"val_splits": []
},
"dtype": "float32",
"model": {
"cg_iters": 5,
"comm_norm_type": "layernorm",
"d_e": 5,
"d_v": 10,
"dec_hidden_dim": 256,
"decoder_arch": "mlp_concat_v2",
"enc_hidden_dim": 256,
"encoder_arch": "mlp_v2",
"gamma": 5.0,
"lora_init_style": "standard",
"lora_rank": 4,
"num_classes": 6,
"num_directions": 8,
"objective_mode": "quadratic",
"rho_init": 0.25,
"rm_init": "orthonormal",
"rm_mode": "context",
"rm_sharing": "directional",
"tikhonov_eps": 1e-05,
"x_solver": "diagonal_prox",
"z_mode": "prox",
"z_solver": "unrolled_cg"
},
"model_type": "sheaf",
"task": "maze",
"task_cfg": {
"connectivity": 8,
"num_classes": 6,
"patch_size": 3,
"stride": 2
},
"training": {
"K_eval": 100,
"K_train": 40,
"batch_size": 128,
"ema_decay": 0.999,
"epochs": 50,
"exit_on_nan": true,
"grad_clip": 1.0,
"loss_window": 4,
"lr": 0.0003,
"mpnn_eval_rounds": 100,
"mpnn_train_rounds": 40,
"seed": 123,
"train_iters_dist": "uniform",
"train_iters_min": 15,
"val_interval": 5,
"warmup_steps": 200,
"weight_decay": 1e-06
},
"wandb": {
"entity": null,
"group": null,
"mode": "disabled",
"name": null,
"project": "sheaf-admm",
"tags": []
}
},
"file": "control/registered-configs/C4-MAZE-QUADRATIC-123.json"
},
"C4-MAZE-QUADRATIC-42": {
"canonical_sha256": "87eb0690dbe06c6c70044866181311c9f2c0f7f724e75063464c501c3ef533db",
"config": {
"data": {
"dir": "/data/train/maze_std3_19px_10k",
"loader": "puzzle",
"train_split": "train",
"val_splits": []
},
"dtype": "float32",
"model": {
"cg_iters": 5,
"comm_norm_type": "layernorm",
"d_e": 5,
"d_v": 10,
"dec_hidden_dim": 256,
"decoder_arch": "mlp_concat_v2",
"enc_hidden_dim": 256,
"encoder_arch": "mlp_v2",
"gamma": 5.0,
"lora_init_style": "standard",
"lora_rank": 4,
"num_classes": 6,
"num_directions": 8,
"objective_mode": "quadratic",
"rho_init": 0.25,
"rm_init": "orthonormal",
"rm_mode": "context",
"rm_sharing": "directional",
"tikhonov_eps": 1e-05,
"x_solver": "diagonal_prox",
"z_mode": "prox",
"z_solver": "unrolled_cg"
},
"model_type": "sheaf",
"task": "maze",
"task_cfg": {
"connectivity": 8,
"num_classes": 6,
"patch_size": 3,
"stride": 2
},
"training": {
"K_eval": 100,
"K_train": 40,
"batch_size": 128,
"ema_decay": 0.999,
"epochs": 50,
"exit_on_nan": true,
"grad_clip": 1.0,
"loss_window": 4,
"lr": 0.0003,
"mpnn_eval_rounds": 100,
"mpnn_train_rounds": 40,
"seed": 42,
"train_iters_dist": "uniform",
"train_iters_min": 15,
"val_interval": 5,
"warmup_steps": 200,
"weight_decay": 1e-06
},
"wandb": {
"entity": null,
"group": null,
"mode": "disabled",
"name": null,
"project": "sheaf-admm",
"tags": []
}
},
"file": "control/registered-configs/C4-MAZE-QUADRATIC-42.json"
},
"C4-MAZE-QUADRATIC-456": {
"canonical_sha256": "14a74656730c7fcc84be1e1cbda241310e68a92cc34f714820b7a5f32f18997e",
"config": {
"data": {
"dir": "/data/train/maze_std3_19px_10k",
"loader": "puzzle",
"train_split": "train",
"val_splits": []
},
"dtype": "float32",
"model": {
"cg_iters": 5,
"comm_norm_type": "layernorm",
"d_e": 5,
"d_v": 10,
"dec_hidden_dim": 256,
"decoder_arch": "mlp_concat_v2",
"enc_hidden_dim": 256,
"encoder_arch": "mlp_v2",
"gamma": 5.0,
"lora_init_style": "standard",
"lora_rank": 4,
"num_classes": 6,
"num_directions": 8,
"objective_mode": "quadratic",
"rho_init": 0.25,
"rm_init": "orthonormal",
"rm_mode": "context",
"rm_sharing": "directional",
"tikhonov_eps": 1e-05,
"x_solver": "diagonal_prox",
"z_mode": "prox",
"z_solver": "unrolled_cg"
},
"model_type": "sheaf",
"task": "maze",
"task_cfg": {
"connectivity": 8,
"num_classes": 6,
"patch_size": 3,
"stride": 2
},
"training": {
"K_eval": 100,
"K_train": 40,
"batch_size": 128,
"ema_decay": 0.999,
"epochs": 50,
"exit_on_nan": true,
"grad_clip": 1.0,
"loss_window": 4,
"lr": 0.0003,
"mpnn_eval_rounds": 100,
"mpnn_train_rounds": 40,
"seed": 456,
"train_iters_dist": "uniform",
"train_iters_min": 15,
"val_interval": 5,
"warmup_steps": 200,
"weight_decay": 1e-06
},
"wandb": {
"entity": null,
"group": null,
"mode": "disabled",
"name": null,
"project": "sheaf-admm",
"tags": []
}
},
"file": "control/registered-configs/C4-MAZE-QUADRATIC-456.json"
},
"C4-SUD-IDENTITY-123": {
"canonical_sha256": "3bbb5dec4ca83cbab0379208b94d19b570eb97e2e411259208734c7467ef06ef",
"config": {
"data": {
"dir": "/data/train/sudoku_easy",
"loader": "puzzle",
"train_split": "train",
"val_splits": []
},
"dtype": "float32",
"model": {
"cg_iters": 5,
"comm_norm_type": "layernorm",
"d_e": 32,
"d_v": 288,
"dec_hidden_dims": [
256
],
"decoder_arch": "sudoku",
"enc_d_model": 128,
"enc_num_blocks": 2,
"encoder_arch": "sudoku",
"gamma": 2.0,
"num_classes": 10,
"num_directions": 9,
"objective_mode": "non_negative",
"rho_init": 0.25,
"rm_constant": true,
"rm_init": "identity",
"rm_mode": "fixed",
"rm_sharing": "sudoku",
"x_solver": "diagonal_prox",
"z_mode": "prox",
"z_solver": "unrolled_cg"
},
"model_type": "sheaf",
"task": "sudoku",
"task_cfg": {},
"training": {
"K_eval": 50,
"K_train": 20,
"batch_size": 128,
"ema_decay": 0.999,
"epochs": 10,
"exit_on_nan": true,
"grad_clip": 1.0,
"loss_window": 2,
"lr": 0.0017,
"mpnn_eval_rounds": 100,
"mpnn_train_rounds": 40,
"seed": 123,
"train_iters_dist": "fixed",
"train_iters_min": 15,
"val_interval": 1,
"warmup_steps": 200,
"weight_decay": 1e-07
},
"wandb": {
"entity": null,
"group": null,
"mode": "disabled",
"name": null,
"project": "sheaf-admm",
"tags": []
}
},
"file": "control/registered-configs/C4-SUD-IDENTITY-123.json"
},
"C4-SUD-IDENTITY-42": {
"canonical_sha256": "d6af26f0bfbcb23770bcd838ee12b937401c2dcc060dff601ff32de262e70f85",
"config": {
"data": {
"dir": "/data/train/sudoku_easy",
"loader": "puzzle",
"train_split": "train",
"val_splits": []
},
"dtype": "float32",
"model": {
"cg_iters": 5,
"comm_norm_type": "layernorm",
"d_e": 32,
"d_v": 288,
"dec_hidden_dims": [
256
],
"decoder_arch": "sudoku",
"enc_d_model": 128,
"enc_num_blocks": 2,
"encoder_arch": "sudoku",
"gamma": 2.0,
"num_classes": 10,
"num_directions": 9,
"objective_mode": "non_negative",
"rho_init": 0.25,
"rm_constant": true,
"rm_init": "identity",
"rm_mode": "fixed",
"rm_sharing": "sudoku",
"x_solver": "diagonal_prox",
"z_mode": "prox",
"z_solver": "unrolled_cg"
},
"model_type": "sheaf",
"task": "sudoku",
"task_cfg": {},
"training": {
"K_eval": 50,
"K_train": 20,
"batch_size": 128,
"ema_decay": 0.999,
"epochs": 10,
"exit_on_nan": true,
"grad_clip": 1.0,
"loss_window": 2,
"lr": 0.0017,
"mpnn_eval_rounds": 100,
"mpnn_train_rounds": 40,
"seed": 42,
"train_iters_dist": "fixed",
"train_iters_min": 15,
"val_interval": 1,
"warmup_steps": 200,
"weight_decay": 1e-07
},
"wandb": {
"entity": null,
"group": null,
"mode": "disabled",
"name": null,
"project": "sheaf-admm",
"tags": []
}
},
"file": "control/registered-configs/C4-SUD-IDENTITY-42.json"
},
"C4-SUD-IDENTITY-456": {
"canonical_sha256": "265179cbf9101405f1b04175bc3d247a719ceb8a415d11b84345a240365cbd49",
"config": {
"data": {
"dir": "/data/train/sudoku_easy",
"loader": "puzzle",
"train_split": "train",
"val_splits": []
},
"dtype": "float32",
"model": {
"cg_iters": 5,
"comm_norm_type": "layernorm",
"d_e": 32,
"d_v": 288,
"dec_hidden_dims": [
256
],
"decoder_arch": "sudoku",
"enc_d_model": 128,
"enc_num_blocks": 2,
"encoder_arch": "sudoku",
"gamma": 2.0,
"num_classes": 10,
"num_directions": 9,
"objective_mode": "non_negative",
"rho_init": 0.25,
"rm_constant": true,
"rm_init": "identity",
"rm_mode": "fixed",
"rm_sharing": "sudoku",
"x_solver": "diagonal_prox",
"z_mode": "prox",
"z_solver": "unrolled_cg"
},
"model_type": "sheaf",
"task": "sudoku",
"task_cfg": {},
"training": {
"K_eval": 50,
"K_train": 20,
"batch_size": 128,
"ema_decay": 0.999,
"epochs": 10,
"exit_on_nan": true,
"grad_clip": 1.0,
"loss_window": 2,
"lr": 0.0017,
"mpnn_eval_rounds": 100,
"mpnn_train_rounds": 40,
"seed": 456,
"train_iters_dist": "fixed",
"train_iters_min": 15,
"val_interval": 1,
"warmup_steps": 200,
"weight_decay": 1e-07
},
"wandb": {
"entity": null,
"group": null,
"mode": "disabled",
"name": null,
"project": "sheaf-admm",
"tags": []
}
},
"file": "control/registered-configs/C4-SUD-IDENTITY-456.json"
}
},
"run_identities": [
"C1-SUD-MPNN225-42",
"C1-SUD-MPNN225-123",
"C1-SUD-MPNN225-456",
"C2-MAZE-MPNN84-42",
"C2-MAZE-MPNN84-123",
"C2-MAZE-MPNN84-456",
"C3-MNIST-CNN-42",
"C3-MNIST-CNN-123",
"C3-MNIST-CNN-456",
"C4-SUD-IDENTITY-42",
"C4-SUD-IDENTITY-123",
"C4-SUD-IDENTITY-456",
"C4-MAZE-QUADRATIC-42",
"C4-MAZE-QUADRATIC-123",
"C4-MAZE-QUADRATIC-456"
],
"submission_limits": {
"all_returned_ids_max": 29,
"cpu_retry_ids": 1,
"cpu_returned_ids_max": 3,
"gpu_retry_ids": 3,
"gpu_returned_ids_max": 26,
"primary_cpu_ids": 2,
"primary_gpu_ids": 23,
"target_scientific_concurrency": 6
},
"title": "Learning Multi-Agent Coordination via Sheaf-ADMM",
"training_common": {
"batch_size": 128,
"dtype": "float32",
"early_stopping": false,
"ema_decay": 0.999,
"evaluation_parameters": "EMA",
"global_grad_clip": 1.0,
"jax_default_matmul_precision": "highest",
"optimizer": "AdamW",
"schedule": "linear_to_task_lr_then_constant",
"selection": "fixed_final_epoch",
"training_mount_content": [
"registered_task_train_split"
],
"validation_selection": false,
"validation_splits": [],
"warmup_steps": 200
},
"training_families": {
"C1-SUD-MPNN225": {
"base": "sudoku_mpnn",
"data": {
"dir": "datasets/sudoku_easy",
"loader": "puzzle",
"train_split": "train",
"val_splits": []
},
"model": {
"comm_norm_type": "layernorm",
"d_e": 32,
"d_v": 225,
"dec_hidden_dims": [
256
],
"decoder_arch": "sudoku",
"enc_d_model": 128,
"enc_num_blocks": 2,
"encoder_arch": "sudoku",
"mpnn_aggregation": "max",
"mpnn_edge_type_mode": "slot",
"mpnn_graph_readout": "per_node",
"mpnn_message_dim": 32,
"num_classes": 10,
"num_directions": 9
},
"model_type": "mpnn",
"paper_row_reconstruction": true,
"seeds": [
42,
123,
456
],
"task": "sudoku",
"task_cfg": {},
"training": {
"epochs": 10,
"lr": 0.0017,
"mpnn_eval_rounds": 50,
"mpnn_train_rounds": 20,
"weight_decay": 1e-07
}
},
"C2-MAZE-MPNN84": {
"base": "maze_mpnn",
"data": {
"dir": "datasets/maze_std3_19px_10k",
"loader": "puzzle",
"train_split": "train",
"val_splits": []
},
"model": {
"comm_norm_type": "layernorm",
"d_e": 42,
"d_v": 84,
"dec_hidden_dim": 256,
"decoder_arch": "mlp_concat_v2",
"enc_hidden_dim": 256,
"encoder_arch": "mlp_v2",
"mpnn_aggregation": "max",
"mpnn_edge_type_mode": "spatial",
"mpnn_graph_readout": "per_node",
"mpnn_message_dim": 42,
"num_classes": 6,
"num_directions": 8
},
"model_type": "mpnn",
"seeds": [
42,
123,
456
],
"task": "maze",
"task_cfg": {
"connectivity": 8,
"num_classes": 6,
"patch_size": 3,
"stride": 2
},
"training": {
"epochs": 50,
"lr": 0.0003,
"mpnn_eval_rounds": 100,
"mpnn_train_rounds": 40,
"train_round_distribution": "fixed",
"weight_decay": 1e-06
}
},
"C3-MNIST-CNN": {
"data": {
"augmentation": false,
"input_range": [
0,
1
],
"normalization": false,
"val_splits": []
},
"model": {
"bias_init": "zeros",
"kernel_init": "glorot_uniform",
"layers": [
"conv3x3-1-32-same-relu",
"conv3x3-32-32-same-relu",
"maxpool2x2-stride2-valid",
"conv3x3-32-64-same-relu",
"conv3x3-64-64-same-relu",
"global-spatial-mean",
"dense64-10"
],
"layout": "NHWC/HWIO/NHWC",
"parameter_count": 65642,
"post_pool_sizes": [
14,
30
],
"pre_pool_sizes": [
28,
60
]
},
"model_type": "mnist_cnn_repro",
"seeds": [
42,
123,
456
],
"task": "mnist",
"training": {
"betas": [
0.9,
0.999
],
"epoch_permutation": "fold_in(PRNGKey(seed),epoch)",
"epochs": 20,
"epsilon": 1e-08,
"examples": 60000,
"final_batch": "pad_to_128_mask_normalize",
"kernel_only_weight_decay": 1e-07,
"loss": "mean_sparse_categorical_cross_entropy",
"lr": 0.001
}
},
"C4-MAZE-QUADRATIC": {
"base": "maze_sheaf",
"data": {
"dir": "datasets/maze_std3_19px_10k",
"loader": "puzzle",
"train_split": "train",
"val_splits": []
},
"forbidden_heads": [
"l1",
"lower_bound",
"upper_bound"
],
"heads": [
"positive_q_diag",
"q"
],
"model": {
"cg_iters": 5,
"comm_norm_type": "layernorm",
"d_e": 5,
"d_v": 10,
"dec_hidden_dim": 256,
"decoder_arch": "mlp_concat_v2",
"enc_hidden_dim": 256,
"encoder_arch": "mlp_v2",
"gamma": 5,
"lora_init_style": "standard",
"lora_rank": 4,
"num_classes": 6,
"num_directions": 8,
"objective_mode": "quadratic",
"rho_init": 0.25,
"rm_init": "orthonormal",
"rm_mode": "context",
"rm_sharing": "directional",
"tikhonov_eps": 1e-05,
"x_solver": "diagonal_prox",
"z_mode": "prox",
"z_solver": "unrolled_cg"
},
"model_type": "sheaf",
"overrides": {
"data.val_splits": [],
"model.objective_mode": "quadratic"
},
"seeds": [
42,
123,
456
],
"task": "maze",
"task_cfg": {
"connectivity": 8,
"num_classes": 6,
"patch_size": 3,
"stride": 2
},
"training": {
"K_eval": 100,
"K_train": 40,
"epochs": 50,
"loss_window": 4,
"lr": 0.0003,
"train_K_distribution": "uniform_inclusive_15_40",
"train_iters_min": 15,
"weight_decay": 1e-06
}
},
"C4-SUD-IDENTITY": {
"base": "sudoku_sheaf",
"constant_map": "F=[I_32,0,...,0] for every slot and endpoint",
"data": {
"dir": "datasets/sudoku_easy",
"loader": "puzzle",
"train_split": "train",
"val_splits": []
},
"model": {
"cg_iters": 5,
"comm_norm_type": "layernorm",
"d_e": 32,
"d_v": 288,
"dec_hidden_dims": [
256
],
"decoder_arch": "sudoku",
"enc_d_model": 128,
"enc_num_blocks": 2,
"encoder_arch": "sudoku",
"gamma": 2.0,
"num_classes": 10,
"num_directions": 9,
"objective_mode": "non_negative",
"rho_init": 0.25,
"rm_constant": true,
"rm_init": "identity",
"rm_mode": "fixed",
"rm_sharing": "sudoku",
"x_solver": "diagonal_prox",
"z_mode": "prox",
"z_solver": "unrolled_cg"
},
"model_type": "sheaf",
"overrides": {
"data.val_splits": [],
"model.rm_constant": true,
"model.rm_init": "identity"
},
"pairing_requirement": "control and intervention common leaves and counters bit-identical before replacement",
"parameter_tree_requirement": "no restriction-map leaf",
"seeds": [
42,
123,
456
],
"task": "sudoku",
"task_cfg": {},
"training": {
"K_eval": 50,
"K_train": 20,
"epochs": 10,
"loss_window": 2,
"lr": 0.0017,
"train_iters_dist": "fixed",
"weight_decay": 1e-07
}
}
}
}