{ "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "sweep": [ { "config": { "lr": 0.0015, "weight_decay": 0.0 }, "per_seed": [ 0.0008893852354958653, 0.0011461004614830017, 0.0009290041634812951, 0.0013412678381428123 ], "mean": 0.0010764394246507436, "std": 0.00018146154865546845, "n": 4, "details": [ { "seed": 0, "metric": 0.0008893852354958653, "params": 13313, "last_loss": 0.0009362917765974999 }, { "seed": 1, "metric": 0.0011461004614830017, "params": 13313, "last_loss": 0.0009171083651017398 }, { "seed": 2, "metric": 0.0009290041634812951, "params": 13313, "last_loss": 0.0008064234256744385 }, { "seed": 3, "metric": 0.0013412678381428123, "params": 13313, "last_loss": 0.000837117291521281 } ] }, { "config": { "lr": 0.003, "weight_decay": 0.0 }, "per_seed": [ 0.0006203416851349175, 0.0007658391259610653, 0.0006099292659200728, 0.0009500968153588474 ], "mean": 0.0007365517230937257, "std": 0.0001378380190480065, "n": 4, "details": [ { "seed": 0, "metric": 0.0006203416851349175, "params": 13313, "last_loss": 0.000665924958884716 }, { "seed": 1, "metric": 0.0007658391259610653, "params": 13313, "last_loss": 0.0007195674593094737 }, { "seed": 2, "metric": 0.0006099292659200728, "params": 13313, "last_loss": 0.0005001857841853052 }, { "seed": 3, "metric": 0.0009500968153588474, "params": 13313, "last_loss": 0.00065102310734801 } ] }, { "config": { "lr": 0.006, "weight_decay": 0.0 }, "per_seed": [ 0.00043351209023967385, 0.0006403940496966243, 0.0003332794294692576, 0.0005523009458556771 ], "mean": 0.0004898716288153082, "std": 0.00011646022207023158, "n": 4, "details": [ { "seed": 0, "metric": 0.00043351209023967385, "params": 13313, "last_loss": 0.0004344877786934376 }, { "seed": 1, "metric": 0.0006403940496966243, "params": 13313, "last_loss": 0.000438494966365397 }, { "seed": 2, "metric": 0.0003332794294692576, "params": 13313, "last_loss": 0.00030466696247458456 }, { "seed": 3, "metric": 0.0005523009458556771, "params": 13313, "last_loss": 0.0003513497015228495 } ] }, { "config": { "lr": 0.003, "weight_decay": 0.0001 }, "per_seed": [ 0.0006000761059112847, 0.0007683035801164806, 0.0006060203886590898, 0.0009272810420952737 ], "mean": 0.0007254202791955322, "std": 0.00013467947201853758, "n": 4, "details": [ { "seed": 0, "metric": 0.0006000761059112847, "params": 13313, "last_loss": 0.0006504375394433737 }, { "seed": 1, "metric": 0.0007683035801164806, "params": 13313, "last_loss": 0.0007221100921742618 }, { "seed": 2, "metric": 0.0006060203886590898, "params": 13313, "last_loss": 0.0004942747787572443 }, { "seed": 3, "metric": 0.0009272810420952737, "params": 13313, "last_loss": 0.000634368343744427 } ] } ], "best_config": { "lr": 0.006, "weight_decay": 0.0 }, "full": { "config": { "lr": 0.006, "weight_decay": 0.0 }, "per_seed": [ 0.00043351209023967385, 0.0006403940496966243, 0.0003332794294692576, 0.0005523009458556771, 0.00037599660572595894, 0.0005211277748458087, 0.00046043944894336164, 0.0004712840309366584 ], "mean": 0.00047354179696412757, "std": 9.170253921476272e-05, "n": 8, "details": [ { "seed": 0, "metric": 0.00043351209023967385, "params": 13313, "last_loss": 0.0004344877786934376 }, { "seed": 1, "metric": 0.0006403940496966243, "params": 13313, "last_loss": 0.000438494966365397 }, { "seed": 2, "metric": 0.0003332794294692576, "params": 13313, "last_loss": 0.00030466696247458456 }, { "seed": 3, "metric": 0.0005523009458556771, "params": 13313, "last_loss": 0.0003513497015228495 }, { "seed": 4, "metric": 0.00037599660572595894, "params": 13313, "last_loss": 0.0005308800761122256 }, { "seed": 5, "metric": 0.0005211277748458087, "params": 13313, "last_loss": 0.0005809032125398516 }, { "seed": 6, "metric": 0.00046043944894336164, "params": 13313, "last_loss": 0.00040943760424852374 }, { "seed": 7, "metric": 0.0004712840309366584, "params": 13313, "last_loss": 0.00035525997867807746 } ] } }, "idea": { "config": { "lr": 0.006, "weight_decay": 0.0 }, "per_seed": [ 0.00012686560512520373, 0.0004387780209071934, 0.0002368502609897405, 0.00015413027722388506, 0.00017462033429183066, 0.00011361205542925745, 0.0002228845260106027, 0.0003435298858676106 ], "mean": 0.0002264088707306655, "std": 0.00010575777904307542, "n": 8, "details": [ { "seed": 0, "metric": 0.00012686560512520373, "params": 45857, "last_loss": 0.00018608432030305265 }, { "seed": 1, "metric": 0.0004387780209071934, "params": 45857, "last_loss": 0.00024508837144821883 }, { "seed": 2, "metric": 0.0002368502609897405, "params": 45857, "last_loss": 0.0001869376830291003 }, { "seed": 3, "metric": 0.00015413027722388506, "params": 45857, "last_loss": 0.00015381897304905578 }, { "seed": 4, "metric": 0.00017462033429183066, "params": 45857, "last_loss": 0.00021269463730277493 }, { "seed": 5, "metric": 0.00011361205542925745, "params": 45857, "last_loss": 0.0001970260846428573 }, { "seed": 6, "metric": 0.0002228845260106027, "params": 45857, "last_loss": 0.00033824800048023464 }, { "seed": 7, "metric": 0.0003435298858676106, "params": 45857, "last_loss": 0.0003234831592999399 } ] }, "comparison": { "delta_mean": -0.00024713292623346206, "idea_wins": 8, "n_pairs": 8, "per_seed_diffs": [ -0.0003066464851144701, -0.00020161602878943086, -9.64291684795171e-05, -0.00039817066863179207, -0.00020137627143412828, -0.0004075157194165513, -0.00023755492293275893, -0.0001277541450690478 ], "p_value": 0.0081, "mde": 9.65319348661709e-05, "mde_rel_pct": 20.385092822858788, "verdict": "idea better (significant)", "system_worked": true }, "mechanism_signature": { "task": "trained dynamics test input sensitivity", "predicted": "recursive feedback should give measurable sensitivity to non-adjacent tokens and bounded local sensitivity", "observed": { "nonlocal_input_grad_norm": 0.26336461305618286, "final_token_grad_norm": 0.570110559463501, "input_jacobian_norm": 0.628002405166626 }, "confirmed": true }, "idea_sweep": [ { "config": { "lr": 0.0015, "weight_decay": 0.0 }, "per_seed": [ 0.0006614502635784447, 0.0005860325763933361, 0.0005585338221862912, 0.001089978963136673, 0.0005720907356590033, 0.00043726706644520164, 0.000829867203719914, 0.0007765801274217665 ], "mean": 0.0006889750948175788, "std": 0.00019161090766784528, "n": 8, "details": [ { "seed": 0, "metric": 0.0006614502635784447, "params": 45857, "last_loss": 0.0008753743558190763 }, { "seed": 1, "metric": 0.0005860325763933361, "params": 45857, "last_loss": 0.0004818156384862959 }, { "seed": 2, "metric": 0.0005585338221862912, "params": 45857, "last_loss": 0.0004933090682607144 }, { "seed": 3, "metric": 0.001089978963136673, "params": 45857, "last_loss": 0.0005436877929605544 }, { "seed": 4, "metric": 0.0005720907356590033, "params": 45857, "last_loss": 0.0005770999426022172 }, { "seed": 5, "metric": 0.00043726706644520164, "params": 45857, "last_loss": 0.00048569669714197515 }, { "seed": 6, "metric": 0.000829867203719914, "params": 45857, "last_loss": 0.0008137347456067801 }, { "seed": 7, "metric": 0.0007765801274217665, "params": 45857, "last_loss": 0.0005952932010404766 } ] }, { "config": { "lr": 0.003, "weight_decay": 0.0 }, "per_seed": [ 0.0006935568526387215, 0.0004945265245623887, 0.00023093193885870278, 0.0005860979435965419, 0.00023517308000009507, 0.00020430154108908027, 0.000541141489520669, 0.0005798843340016901 ], "mean": 0.00044570171303348616, "std": 0.0001800566802955679, "n": 8, "details": [ { "seed": 0, "metric": 0.0006935568526387215, "params": 45857, "last_loss": 0.0006174481904599816 }, { "seed": 1, "metric": 0.0004945265245623887, "params": 45857, "last_loss": 0.0004121658857911825 }, { "seed": 2, "metric": 0.00023093193885870278, "params": 45857, "last_loss": 0.0002016581519274041 }, { "seed": 3, "metric": 0.0005860979435965419, "params": 45857, "last_loss": 0.00034360544756054876 }, { "seed": 4, "metric": 0.00023517308000009507, "params": 45857, "last_loss": 0.00022848578926641495 }, { "seed": 5, "metric": 0.00020430154108908027, "params": 45857, "last_loss": 0.00022899049799889325 }, { "seed": 6, "metric": 0.000541141489520669, "params": 45857, "last_loss": 0.0004576991230715066 }, { "seed": 7, "metric": 0.0005798843340016901, "params": 45857, "last_loss": 0.00044673100113868714 } ] }, { "config": { "lr": 0.006, "weight_decay": 0.0 }, "per_seed": [ 0.00012686560512520373, 0.0004387780209071934, 0.0002368502609897405, 0.00015413027722388506, 0.00017462033429183066, 0.00011361205542925745, 0.0002228845260106027, 0.0003435298858676106 ], "mean": 0.0002264088707306655, "std": 0.00010575777904307542, "n": 8, "details": [ { "seed": 0, "metric": 0.00012686560512520373, "params": 45857, "last_loss": 0.00018608432030305265 }, { "seed": 1, "metric": 0.0004387780209071934, "params": 45857, "last_loss": 0.00024508837144821883 }, { "seed": 2, "metric": 0.0002368502609897405, "params": 45857, "last_loss": 0.0001869376830291003 }, { "seed": 3, "metric": 0.00015413027722388506, "params": 45857, "last_loss": 0.00015381897304905578 }, { "seed": 4, "metric": 0.00017462033429183066, "params": 45857, "last_loss": 0.00021269463730277493 }, { "seed": 5, "metric": 0.00011361205542925745, "params": 45857, "last_loss": 0.0001970260846428573 }, { "seed": 6, "metric": 0.0002228845260106027, "params": 45857, "last_loss": 0.00033824800048023464 }, { "seed": 7, "metric": 0.0003435298858676106, "params": 45857, "last_loss": 0.0003234831592999399 } ] }, { "config": { "lr": 0.003, "weight_decay": 0.0001 }, "per_seed": [ 0.0007010238477960229, 0.0004941863589920104, 0.0002418020012555644, 0.0006359018152579665, 0.0002609499788377434, 0.0002295856538694352, 0.0005036439397372305, 0.0005770205752924085 ], "mean": 0.0004555142713797977, "std": 0.0001753433538338588, "n": 8, "details": [ { "seed": 0, "metric": 0.0007010238477960229, "params": 45857, "last_loss": 0.0006500197574496269 }, { "seed": 1, "metric": 0.0004941863589920104, "params": 45857, "last_loss": 0.0004117381514515728 }, { "seed": 2, "metric": 0.0002418020012555644, "params": 45857, "last_loss": 0.00021009275340475142 }, { "seed": 3, "metric": 0.0006359018152579665, "params": 45857, "last_loss": 0.00037261329242028297 }, { "seed": 4, "metric": 0.0002609499788377434, "params": 45857, "last_loss": 0.00026486166985705494 }, { "seed": 5, "metric": 0.0002295856538694352, "params": 45857, "last_loss": 0.0002658980851992965 }, { "seed": 6, "metric": 0.0005036439397372305, "params": 45857, "last_loss": 0.00045993857202120125 }, { "seed": 7, "metric": 0.0005770205752924085, "params": 45857, "last_loss": 0.00045114275999367235 } ] } ], "parameter_counts": { "baseline": 13313, "idea": 45857 } }