Fourier Replay-Mode Stabilizer / bench_report.json

Failed on benchmark

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "dynamics",
  4  "model": "ring_rnn",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.006,
 10      "wd": 0.0001
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "wd": 0.0
 17        },
 18        "mean": 0.09482549037784338
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "wd": 0.0001
 24        },
 25        "mean": 0.0948077691718936
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.003,
 30          "wd": 0.0
 31        },
 32        "mean": 0.009329887572675943
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "wd": 0.0001
 38        },
 39        "mean": 0.009322796016931534
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.006,
 44          "wd": 0.0
 45        },
 46        "mean": 0.0037712849443778396
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.006,
 51          "wd": 0.0001
 52        },
 53        "mean": 0.003756003046873957
 54      }
 55    ],
 56    "full": {
 57      "mean": 0.004300541011616588,
 58      "std": 0.002254021150865667,
 59      "per_seed": [
 60        0.0024547018110752106,
 61        0.004727087914943695,
 62        0.0022790830116719007,
 63        0.005563139449805021,
 64        0.009408666752278805,
 65        0.0037723351269960403,
 66        0.002070292131975293,
 67        0.004129021894186735
 68      ],
 69      "n": 8
 70    }
 71  },
 72  "idea": {
 73    "mean": 0.004188583581708372,
 74    "std": 0.002221294410541217,
 75    "per_seed": [
 76      0.0025346376933157444,
 77      0.004659987520426512,
 78      0.0020646981429308653,
 79      0.004926756490021944,
 80      0.00941324234008789,
 81      0.0038586521986871958,
 82      0.0022045269142836332,
 83      0.003846167353913188
 84    ],
 85    "n": 8
 86  },
 87  "comparison": {
 88    "delta_mean": -0.000111957429908216,
 89    "idea_wins": 4,
 90    "n_pairs": 8,
 91    "per_seed_diffs": [
 92      7.993588224053383e-05,
 93      -6.71003945171833e-05,
 94      -0.00021438486874103546,
 95      -0.0006363829597830772,
 96      4.575587809085846e-06,
 97      8.631707169115543e-05,
 98      0.00013423478230834007,
 99      -0.00028285454027354717
100    ],
101    "p_value": 0.2995,
102    "mde": 0.0002159844454493105,
103    "mde_rel_pct": 5.022262195986389,
104    "verdict": "no significant win",
105    "system_worked": false
106  },
107  "mechanism_signature": {
108    "predicted_vs_observed_growth": {
109      "predicted_max_growth": -0.9413630873202358,
110      "observed_perturbation_growth": 0.001524896847113279,
111      "local_slope": 0.7495013698935509,
112      "growth_abs_error": 0.9428879841673491,
113      "confirmed": false
114    },
115    "stage1_sanity": {
116      "fft_error": 7.850462293418876e-17,
117      "predicted_growth": -0.8527323896911315,
118      "observed_growth": -0.8530675168113421,
119      "abs_error": 0.00033512712021055435,
120      "passed": true
121    },
122    "idea_sweep": [
123      {
124        "cfg": {
125          "lr": 0.006,
126          "wd": 0.0001,
127          "strength": 0.01
128        },
129        "mean": 0.0042766523256432265
130      },
131      {
132        "cfg": {
133          "lr": 0.006,
134          "wd": 0.0001,
135          "strength": 0.03
136        },
137        "mean": 0.0042337587219662964
138      },
139      {
140        "cfg": {
141          "lr": 0.006,
142          "wd": 0.0001,
143          "strength": 0.1
144        },
145        "mean": 0.004188583581708372
146      }
147    ]
148  },
149  "parameter_count": 97,
150  "device": "cuda"
151}