Mode-Aware Mask Schedule / report_bench_2026-08-31T195728.md
Failed on benchmark
Стенд-проверка (stage-2) · промт оператора:
(универсальный)
Ответ агента:
{ "worked": false, "confidence": 9, "verdict": "Built a local custom multi-token masked-sequence track with identical tiny Transformer systems and ran the full 8-seed paired protocol. The schedule probabilities matched Monte Carlo measurements, but the best idea configuration had higher test BCE than baseline (0.149931 vs 0.143731), paired delta +0.006200 and permutation p=0.63155; the trained-model empty-context mechanism signature was also not confirmed.", "metrics": { "baseline": "Test BCE mean 0.1437309254, std 0.0238559610; best lr=0.002, epochs=14.", "idea": "Test BCE mean 0.1499312371, std 0.0381410620; best lambda=0.10, rho=0.01, lr=0.004, epochs=14.", "comparison": "delta_mean=+0.0062003117, p_value=0.63155, idea_wins=5/8, verdict=no significant win.", "mechanism": "Empty-context mode probability baseline=0.8209748 and idea=0.8391162; target=0.8; absolute error worsened from 0.0209748 to 0.0391162; confirmed=false." }, "how_to_run": "/home/maxwelhelp/main/bin/python3 bench_mode_schedule.py", "files": [ "bench_mode_schedule.py", "custom_masked_track.py", "bench_report.json", "bench_run.log" ], "limitations": "Only the local custom multi-token sequence track was tested because built-in tracks lack masked multi-token prediction structure. No language corpus, unconditional generation, larger Transformer, longer training, or compute-normalized sweep was tested.", "bench_report": { "bench_version": 1, "track": "masked_multitoken_modes", "model": "transformer_tiny", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.002, "epochs": 14, "hide_count": 1 }, "sweep": [ { "cfg": { "lr": 0.002, "epochs": 14, "hide_count": 1 }, "mean": 0.1428266018629074 }, { "cfg": { "lr": 0.004, "epochs": 14, "hide_count": 1 }, "mean": 0.20178561471402645 }, { "cfg": { "lr": 0.008, "epochs": 14, "hide_count": 1 }, "mean": 0.14775850623846054 } ], "full": { "mean": 0.14373092539608479, "std": 0.02385596096955299, "per_seed": [ 0.1670890897512436, 0.12829364836215973, 0.13317705690860748, 0.14274661242961884, 0.16149330139160156, 0.09596626460552216, 0.14481675624847412, 0.1762646734714508 ], "n": 8 } }, "idea": { "mean": 0.14993123710155487, "std": 0.038141062047303144, "per_seed": [ 0.1759413331747055, 0.13947612047195435, 0.12913769483566284, 0.14234721660614014, 0.149277925491333, 0.09132890403270721, 0.13944129645824432, 0.2324994057416916 ], "n": 8, "best_cfg": { "lr": 0.004, "epochs": 14, "lam": 0.1, "rho": 0.01 } }, "comparison": { "delta_mean": 0.006200311705470085, "idea_wins": 5, "n_pairs": 8, "per_seed_diffs": [ 0.008852243423461914, 0.011182472109794617, -0.004039362072944641, -0.00039939582347869873, -0.012215375900268555, -0.004637360572814941, -0.005375459790229797, 0.056234732270240784 ], "p_value": 0.63155, "verdict": "no significant win", "system_worked": false }, "mechanism_signature": { "schedule_pi1_predicted_observed": [ [ 0.0, 0.0 ], [ 0.1, 0.1014000028 ], [ 0.11, 0.1115500033 ], [ 0.26, 0.2624300122 ] ], "full_atom_predicted_observed": [ [ 0.0, 0.0 ], [ 0.0, 0.0 ], [ 0.01, 0.00988 ], [ 0.01, 0.00988 ] ], "target_true_weight": 0.8, "observed_baseline_empty_mode_probability": 0.8209748045, "observed_idea_empty_mode_probability": 0.8391162083, "confirmed": false }, "custom_track": { "name": "masked_multitoken_modes", "file": "custom_masked_track.py", "domain": "sequence" } }, "system_verdict": "failed", "practical_verdict": "inconclusive", "mechanism_ok": 0, "system_judged": true }