OT-Sufficient Bottleneck Flow Matching / bench_report.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "conditional_multitoken_diffusion",
  4  "model": "shared_encoder_mlp",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.01,
 10      "epochs": 12
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "epochs": 12
 17        },
 18        "mean": 0.390290267765522
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "epochs": 12
 24        },
 25        "mean": 0.3733091354370117
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.01,
 30          "epochs": 12
 31        },
 32        "mean": 0.36269813030958176
 33      }
 34    ],
 35    "full": {
 36      "mean": 0.3607390485703945,
 37      "std": 0.01654029145653189,
 38      "per_seed": [
 39        0.37062346935272217,
 40        0.32732465863227844,
 41        0.3899999260902405,
 42        0.36284446716308594,
 43        0.3566600978374481,
 44        0.3669852316379547,
 45        0.3537239730358124,
 46        0.3577505648136139
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "per_seed": [
 53      0.829505980014801,
 54      0.8032728433609009,
 55      0.6855372190475464,
 56      0.7373040318489075,
 57      0.7540731430053711,
 58      0.773855447769165,
 59      0.7264139652252197,
 60      0.7292614579200745
 61    ],
 62    "mean": 0.7549030110239983,
 63    "std": 0.046067620085947815
 64  },
 65  "comparison": {
 66    "delta_mean": 0.39416396245360374,
 67    "idea_wins": 0,
 68    "n_pairs": 8,
 69    "per_seed_diffs": [
 70      0.45888251066207886,
 71      0.47594818472862244,
 72      0.2955372929573059,
 73      0.37445956468582153,
 74      0.397413045167923,
 75      0.4068702161312103,
 76      0.37268999218940735,
 77      0.37151089310646057
 78    ],
 79    "p_value": 0.0081,
 80    "mde": 0.047023554672292786,
 81    "mde_rel_pct": 13.03533810898673,
 82    "verdict": "idea worse (significant)",
 83    "system_worked": false
 84  },
 85  "mechanism_signature": {
 86    "mechanism_signature": {
 87      "observed_token_std": 0.7545899152755737,
 88      "predicted_token_std": 0.7607559561729431,
 89      "observed_sample_mean_std": 0.14106108248233795,
 90      "predicted_sample_mean_std": 0.1528526395559311,
 91      "confirmed": true
 92    },
 93    "idea_lambda": 8.0,
 94    "idea_runs": [
 95      {
 96        "lambda": 0.0,
 97        "result": {
 98          "per_seed": [
 99            0.924494743347168,
100            1.0625691413879395,
101            0.7344903349876404,
102            0.912911593914032,
103            0.8331963419914246,
104            0.8818286657333374,
105            0.8657779693603516,
106            0.7827775478363037
107          ],
108          "mean": 0.8747557923197746,
109          "std": 0.09942397904692929
110        }
111      },
112      {
113        "lambda": 2.0,
114        "result": {
115          "per_seed": [
116            0.871182918548584,
117            0.8115776777267456,
118            0.6836869716644287,
119            0.7553004026412964,
120            0.7379289865493774,
121            0.782780110836029,
122            0.7276151776313782,
123            0.7247848510742188
124          ],
125          "mean": 0.7618571370840073,
126          "std": 0.058718477961378
127        }
128      },
129      {
130        "lambda": 8.0,
131        "result": {
132          "per_seed": [
133            0.829505980014801,
134            0.8032728433609009,
135            0.6855372190475464,
136            0.7373040318489075,
137            0.7540731430053711,
138            0.773855447769165,
139            0.7264139652252197,
140            0.7292614579200745
141          ],
142          "mean": 0.7549030110239983,
143          "std": 0.046067620085947815
144        }
145      }
146    ]
147  }
148}