Gauge-Free Spectral OT Layer / bench_report.json

Failed on benchmark

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "sequence",
  4  "model": "transformer_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.003,
 10      "delta": 0.0001
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "delta": 0.0001
 17        },
 18        "mean": 1.0469075739383698
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "delta": 0.0001
 24        },
 25        "mean": 1.004875659942627
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.01,
 30          "delta": 0.0001
 31        },
 32        "mean": 1.015213631093502
 33      }
 34    ],
 35    "full": {
 36      "mean": 1.004875659942627,
 37      "std": 0.08302304230055646,
 38      "per_seed": [
 39        1.0032966136932373,
 40        0.8694341778755188,
 41        1.1456023454666138,
 42        1.081549882888794,
 43        0.9330782294273376,
 44        0.9555575847625732,
 45        1.0619580745697021,
 46        0.9885283708572388
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "mean": 1.0048756524920464,
 53    "std": 0.0830230295635864,
 54    "per_seed": [
 55      1.0032966136932373,
 56      0.8694342374801636,
 57      1.1456023454666138,
 58      1.0815497636795044,
 59      0.9330782294273376,
 60      0.9555575847625732,
 61      1.0619581937789917,
 62      0.9885282516479492
 63    ],
 64    "n": 8
 65  },
 66  "comparison": {
 67    "delta_mean": -7.450580596923828e-09,
 68    "idea_wins": 2,
 69    "n_pairs": 8,
 70    "per_seed_diffs": [
 71      0.0,
 72      5.960464477539063e-08,
 73      0.0,
 74      -1.1920928955078125e-07,
 75      0.0,
 76      0.0,
 77      1.1920928955078125e-07,
 78      -1.1920928955078125e-07
 79    ],
 80    "p_value": 1.0,
 81    "mde": 6.759130857174038e-08,
 82    "mde_rel_pct": 6.726335532457765e-06,
 83    "verdict": "no measurable effect",
 84    "system_worked": false
 85  },
 86  "mechanism_signature": {
 87    "idea_config": {
 88      "lr": 0.003,
 89      "delta": 0.0001
 90    },
 91    "idea_sweep": [
 92      {
 93        "cfg": {
 94          "lr": 0.001,
 95          "delta": 0.0001
 96        },
 97        "result": {
 98          "mean": 1.046907588839531,
 99          "std": 0.07630641547099135,
100          "per_seed": [
101            1.033826470375061,
102            0.9115373492240906,
103            1.1899056434631348,
104            1.0792663097381592,
105            0.9867245554924011,
106            1.0289349555969238,
107            1.0983328819274902,
108            1.0467325448989868
109          ],
110          "n": 8
111        }
112      },
113      {
114        "cfg": {
115          "lr": 0.003,
116          "delta": 0.0001
117        },
118        "result": {
119          "mean": 1.0048756524920464,
120          "std": 0.0830230295635864,
121          "per_seed": [
122            1.0032966136932373,
123            0.8694342374801636,
124            1.1456023454666138,
125            1.0815497636795044,
126            0.9330782294273376,
127            0.9555575847625732,
128            1.0619581937789917,
129            0.9885282516479492
130          ],
131          "n": 8
132        }
133      },
134      {
135        "cfg": {
136          "lr": 0.01,
137          "delta": 0.0001
138        },
139        "result": {
140          "mean": 1.0152136087417603,
141          "std": 0.07053051427342504,
142          "per_seed": [
143            1.039136290550232,
144            0.8907526731491089,
145            1.1310786008834839,
146            1.0511094331741333,
147            1.0021077394485474,
148            0.9427624940872192,
149            1.071175456047058,
150            0.9935861825942993
151          ],
152          "n": 8
153        }
154      }
155    ],
156    "mechanism_signature": {
157      "prediction": "quotient attention suppresses row/column nuisance sensitivity while retaining interaction sensitivity",
158      "baseline": {
159        "kind": "baseline",
160        "rowcol_plan_rms": 4.0958703095839155e-10,
161        "interaction_plan_rms": 1.726159410964101e-07
162      },
163      "idea": {
164        "kind": "idea",
165        "rowcol_plan_rms": 2.545942379583721e-06,
166        "interaction_plan_rms": 5.0314388317929115e-06
167      },
168      "predicted_ratio_bound": 1.0,
169      "observed_ratio": 6200.737450574088,
170      "confirmed": false
171    }
172  }
173}