RealArDor's picture
download
raw
8.16 kB
{
"input": "data/appo_basic_supervision_v1/unified/hard",
"init_checkpoint": "/tmp/nanojev_appo_sft_20260919/initial_checkpoint",
"output_dir": "runs/appo_basic_supervision_v1/hard_s17",
"stage": "sft",
"loss": "ce",
"balance": "task",
"policy_pool_weights": "runs/appo_basic_supervision_v1/policy_pool_weights.json",
"retention_fraction": 0.25,
"reward_samples": 32,
"steps": 300,
"head_steps": 0,
"batch_questions": 24,
"microbatch_questions": 8,
"max_microbatch_tokens": 32768,
"max_length": 8192,
"eval_every": 100,
"seed": 17,
"backbone_lr": 2e-05,
"head_lr": 0.0002,
"head_warmup_lr": 0.001,
"weight_decay": 0.01,
"precision": "bf16",
"gradient_checkpointing": true,
"disable_native_triton": true,
"td_episodes": null,
"td_n_step": 3,
"td_weight": 0.0,
"target_update_every": 25,
"validate_only": false,
"resolved_policy_pool_weights": {
"maze/policy": 0.3333333333333333,
"snake/policy": 0.3333333333333333,
"shooting/policy/basic": 0.1556701030927835,
"shooting/policy/predict_position": 0.17766323024054984
},
"schema_version": "nanojev-unified-games-v1",
"model": "Qwen/Qwen3-0.6B",
"set_head": "attention",
"resolved_model_revision": "c1899de289a04d12100db370d81485cdf75e47ca",
"initialization": "local DecisionModel warm start; fresh optimizer",
"init_weights_sha256": "c5b2f41459c9ccabe7417a2947489eaa80443935bc79d3cb528a886aab58d3d5",
"data_sha256": {
"data/appo_basic_supervision_v1/unified/hard/manifest.json": "d1c803c79b3f8f7888ee1b8488bdaa21df3c1d2db94c389e829b3e15e199f308",
"data/appo_basic_supervision_v1/unified/hard/train.jsonl": "f669f5cb2381343126658a74d2ff39bac84fbeecadfb7959f72b16e6b08a38c9",
"data/appo_basic_supervision_v1/unified/hard/dev.jsonl": "408ba5c9f59ef97e15bd43d839e62133e302dace64ea039856125f3b08f221f0",
"data/appo_basic_supervision_v1/unified/hard/calibration.jsonl": "51e15894cc92c0ec89c27c4c2ec7dd48cfdae815bd5c3c387f2c9cfcfa30bd61",
"data/appo_basic_supervision_v1/unified/hard/test.jsonl": "c29c0c8aea031acf2f2dd79ac60e2a85707644a6ce40347e1fa13ebacb2d4e59",
"data/appo_basic_supervision_v1/unified/hard/ood.jsonl": "15a24144cc8e2d342f151c2649bf200ff8efb2fdb6620f6c05fe759170579fd9"
},
"implementation_sha256": "a4aedc3b7ad25d875af67943bef6dbc017799f46c4ae0a845fc200e1866fa6cd",
"continuation_policy_id": null,
"population_weights": {
"maze/policy": 0.3333333333333333,
"snake/policy": 0.3333333333333333,
"shooting/policy/basic": 0.1556701030927835,
"shooting/policy/predict_position": 0.17766323024054984
},
"policy_pool_weights_sha256": "28886027e7180b4f2e33fbc14799894e5061b3a2b204862324a6c6314f331fe0",
"eligible_by_split_sampling_pool": {
"calibration/maze/policy": 188,
"calibration/shooting/policy/basic": 377,
"calibration/shooting/policy/predict_position": 126,
"calibration/snake/policy": 137,
"dev/maze/policy": 213,
"dev/shooting/policy/basic": 384,
"dev/shooting/policy/predict_position": 138,
"dev/snake/policy": 157,
"ood/maze/policy": 202,
"ood/shooting/policy/basic": 589,
"ood/shooting/policy/predict_position": 99,
"ood/snake/policy": 141,
"test/maze/policy": 211,
"test/shooting/policy/basic": 756,
"test/shooting/policy/predict_position": 139,
"test/snake/policy": 116,
"train/maze/policy": 651,
"train/shooting/policy/basic": 3054,
"train/shooting/policy/predict_position": 517,
"train/snake/policy": 400
},
"sampling": "stratified with replacement; per-cell exact population weights; train rows only",
"parameter_storage": "float32",
"forward_autocast": "bf16",
"objective": "Choice CE with explicit API or expert targets plus observed Boolean event loss",
"selection": "minimum fixed population-weighted dev CE including initial checkpoint; test only after selection",
"temperature": 1.0,
"temperature_fitted": false,
"deps": {
"torch": "2.14.0",
"transformers": "5.17.0",
"safetensors": "0.8.0"
},
"gpu": "NVIDIA A100-SXM4-80GB",
"schema_counts": {
"records": 8609,
"outcome_questions": 0,
"continuation_policy_id": null,
"questions_by_split_task_role": {
"calibration/maze/policy": 188,
"calibration/shooting/policy": 503,
"calibration/snake/policy": 138,
"dev/maze/policy": 215,
"dev/shooting/policy": 522,
"dev/snake/policy": 158,
"ood/maze/policy": 202,
"ood/shooting/policy": 689,
"ood/snake/policy": 143,
"test/maze/policy": 211,
"test/shooting/policy": 895,
"test/snake/policy": 116,
"train/maze/policy": 653,
"train/shooting/policy": 3573,
"train/snake/policy": 403
},
"eligible_by_split_task_role": {
"calibration/maze/policy": 188,
"calibration/shooting/policy": 503,
"calibration/snake/policy": 137,
"dev/maze/policy": 213,
"dev/shooting/policy": 522,
"dev/snake/policy": 157,
"ood/maze/policy": 202,
"ood/shooting/policy": 688,
"ood/snake/policy": 141,
"test/maze/policy": 211,
"test/shooting/policy": 895,
"test/snake/policy": 116,
"train/maze/policy": 651,
"train/shooting/policy": 3571,
"train/snake/policy": 400
},
"quarantined_policy_questions": 14,
"policy_questions_by_split_task_target_kind": {
"calibration/maze/api_policy_distribution": 188,
"calibration/shooting/api_policy_distribution": 126,
"calibration/shooting/expert_action": 377,
"calibration/snake/api_policy_distribution": 138,
"dev/maze/api_policy_distribution": 215,
"dev/shooting/api_policy_distribution": 138,
"dev/shooting/expert_action": 384,
"dev/snake/api_policy_distribution": 158,
"ood/maze/api_policy_distribution": 202,
"ood/shooting/api_policy_distribution": 100,
"ood/shooting/expert_action": 589,
"ood/snake/api_policy_distribution": 143,
"test/maze/api_policy_distribution": 211,
"test/shooting/api_policy_distribution": 139,
"test/shooting/expert_action": 756,
"test/snake/api_policy_distribution": 116,
"train/maze/api_policy_distribution": 653,
"train/shooting/api_policy_distribution": 519,
"train/shooting/expert_action": 3054,
"train/snake/api_policy_distribution": 403
},
"unique_episodes_by_split_task_role": {
"calibration/maze/policy": 10,
"calibration/shooting/policy": 70,
"calibration/snake/policy": 8,
"dev/maze/policy": 10,
"dev/shooting/policy": 70,
"dev/snake/policy": 8,
"ood/maze/policy": 10,
"ood/shooting/policy": 134,
"ood/snake/policy": 8,
"test/maze/policy": 10,
"test/shooting/policy": 134,
"test/snake/policy": 8,
"train/maze/policy": 32,
"train/shooting/policy": 536,
"train/snake/policy": 28
},
"rows_missing_episode_id_by_split_task_role": {},
"outcome_episode_counts_by_split_task": {},
"outcome_state_label_counts_by_split_task": {},
"outcome_question_label_counts_by_split_task": {},
"episode_audit_note": "Repeated states/questions are not independent episodes. Single-outcome cells are retained without forced balancing. Missing episode IDs are counted explicitly and excluded from episode totals."
},
"predictive_samples_are_physical_actions": false,
"frozen_continuation_policy_is_updated_by_this_script": false,
"td_enabled": false,
"td_audit": null,
"td_implementation_sha256": null,
"td_contract": {
"active": false,
"gamma": 1,
"n_step_unit": "recorded decision transitions",
"continuation_policy_id": null,
"bootstrap_policy": "recorded frozen behavior_probs; never online greedy",
"target_microbatch_questions": 4,
"max_microbatch_tokens": 32768,
"heldout_target": "observed terminal outcome only",
"outcome_loss": "(1-td_weight)*MC_Brier + td_weight*detached_soft_TD_Brier",
"initial_target": "deepcopy of initial online model; eval mode; no gradients",
"prediction_cache": "none; only bootstrap input tokens and indices are cached",
"bootstrap_cache": null
}
}

Xet Storage Details

Size:
8.16 kB
·
Xet hash:
3ad2be9a5957d237a93033eec73522933a0da8786e39205c9d13b547a38d7d3f

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.