Balanced design attention / 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": 0.001,
 10      "epochs": 12
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "epochs": 12
 17        },
 18        "mean": 0.3171548619866371
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "epochs": 12
 24        },
 25        "mean": 0.33575212955474854
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.006,
 30          "epochs": 12
 31        },
 32        "mean": 0.3609168827533722
 33      }
 34    ],
 35    "full": {
 36      "mean": 0.33212270587682724,
 37      "std": 0.052525560059948226,
 38      "per_seed": [
 39        0.34580686688423157,
 40        0.2720120847225189,
 41        0.27966615557670593,
 42        0.37113434076309204,
 43        0.28781864047050476,
 44        0.3107385039329529,
 45        0.4378826916217804,
 46        0.3519223630428314
 47      ],
 48      "n": 8
 49    },
 50    "union_grid": [
 51      {
 52        "lr": 0.001,
 53        "epochs": 12
 54      },
 55      {
 56        "lr": 0.003,
 57        "epochs": 12
 58      },
 59      {
 60        "lr": 0.006,
 61        "epochs": 12
 62      }
 63    ]
 64  },
 65  "idea": {
 66    "mean": 0.3073607422411442,
 67    "std": 0.05354461640119265,
 68    "per_seed": [
 69      0.3493795692920685,
 70      0.2742547392845154,
 71      0.2889987826347351,
 72      0.37771299481391907,
 73      0.25186100602149963,
 74      0.2312282919883728,
 75      0.38401901721954346,
 76      0.3014315366744995
 77    ],
 78    "n": 8,
 79    "selected_cfg": {
 80      "lr": 0.001,
 81      "epochs": 12
 82    }
 83  },
 84  "comparison": {
 85    "delta_mean": -0.02476196363568306,
 86    "idea_wins": 4,
 87    "n_pairs": 8,
 88    "per_seed_diffs": [
 89      0.003572702407836914,
 90      0.00224265456199646,
 91      0.009332627058029175,
 92      0.006578654050827026,
 93      -0.03595763444900513,
 94      -0.07951021194458008,
 95      -0.05386367440223694,
 96      -0.05049082636833191
 97    ],
 98    "p_value": 0.1218,
 99    "mde": 0.028806811988049977,
100    "mde_rel_pct": 8.673544891186518,
101    "verdict": "no significant win",
102    "system_worked": false
103  },
104  "mechanism_signature": {
105    "design": {
106      "predicted_pair_coverage": 1.0,
107      "observed_attention_pair_mean": 0.40865393000985345,
108      "observed_attention_pair_variance": 0.0008423293021837351,
109      "routing_pair_coverage_variance": 0.0,
110      "confirmed": true
111    },
112    "idea_sweep": [
113      {
114        "cfg": {
115          "lr": 0.001,
116          "epochs": 12
117        },
118        "mean": 0.3225979581475258
119      },
120      {
121        "cfg": {
122          "lr": 0.001,
123          "epochs": 12
124        },
125        "mean": 0.32260219752788544
126      },
127      {
128        "cfg": {
129          "lr": 0.006,
130          "epochs": 12
131        },
132        "mean": 0.324721522629261
133      }
134    ],
135    "selected_cfg": {
136      "lr": 0.001,
137      "epochs": 12
138    }
139  }
140}