Bi-Maxwell Muon / bench_report.json

Unverified

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "tabular",
  4  "model": "mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.005,
 10      "weight_decay": 0.0001
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 107.39756393432617
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "weight_decay": 0.0001
 24        },
 25        "mean": 107.39925003051758
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.002,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 44.976237297058105
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.002,
 37          "weight_decay": 0.0001
 38        },
 39        "mean": 44.978708267211914
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.003,
 44          "weight_decay": 0.0
 45        },
 46        "mean": 24.11869478225708
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.003,
 51          "weight_decay": 0.0001
 52        },
 53        "mean": 24.120020389556885
 54      },
 55      {
 56        "cfg": {
 57          "lr": 0.005,
 58          "weight_decay": 0.0
 59        },
 60        "mean": 21.456337451934814
 61      },
 62      {
 63        "cfg": {
 64          "lr": 0.005,
 65          "weight_decay": 0.0001
 66        },
 67        "mean": 21.456222534179688
 68      }
 69    ],
 70    "full": {
 71      "mean": 21.57731342315674,
 72      "std": 1.5687495825188427,
 73      "per_seed": [
 74        22.156478881835938,
 75        22.034189224243164,
 76        22.254423141479492,
 77        19.379798889160156,
 78        19.08241081237793,
 79        22.424152374267578,
 80        21.120243072509766,
 81        24.166810989379883
 82      ],
 83      "n": 8
 84    }
 85  },
 86  "idea": {
 87    "mean": 23.979673385620117,
 88    "std": 1.9050116210009282,
 89    "per_seed": [
 90      25.87602996826172,
 91      24.822738647460938,
 92      24.548673629760742,
 93      21.082088470458984,
 94      21.36577033996582,
 95      24.733022689819336,
 96      22.76095199584961,
 97      26.64811134338379
 98    ],
 99    "n": 8
100  },
101  "comparison": {
102    "delta_mean": 2.402359962463379,
103    "idea_wins": 0,
104    "n_pairs": 8,
105    "per_seed_diffs": [
106      3.7195510864257812,
107      2.7885494232177734,
108      2.29425048828125,
109      1.7022895812988281,
110      2.2833595275878906,
111      2.308870315551758,
112      1.6407089233398438,
113      2.4813003540039062
114    ],
115    "p_value": 0.0081,
116    "mde": 0.5469499653262384,
117    "mde_rel_pct": 2.534838117238695,
118    "verdict": "idea worse (significant)",
119    "system_worked": false
120  },
121  "mechanism_signature": {
122    "track_choice": "optimizer intervention matches tabular Friedman#1 track; identical mlp_tiny systems differ only in matrix momentum state",
123    "idea_config": {
124      "lr": 0.003,
125      "weight_decay": 0.0001,
126      "beta_fast": 0.9,
127      "beta_slow": 0.99,
128      "weight_fast": 0.5
129    },
130    "mechanism_signature": {
131      "prediction": "impulse response initialized by a measured trained-model gradient has decay slope between fast and slow log betas",
132      "predicted_interval": [
133        -0.10536051565782628,
134        -0.01005033585350145
135      ],
136      "observed_mean_slope": -0.09078844601504563,
137      "per_seed": [
138        {
139          "seed": 0,
140          "observed_mix_log_slope": -0.09078844601504561,
141          "predicted_fast_log_beta": -0.10536051565782628,
142          "predicted_slow_log_beta": -0.01005033585350145
143        },
144        {
145          "seed": 1,
146          "observed_mix_log_slope": -0.09078844601504557,
147          "predicted_fast_log_beta": -0.10536051565782628,
148          "predicted_slow_log_beta": -0.01005033585350145
149        },
150        {
151          "seed": 2,
152          "observed_mix_log_slope": -0.09078844601504571,
153          "predicted_fast_log_beta": -0.10536051565782628,
154          "predicted_slow_log_beta": -0.01005033585350145
155        },
156        {
157          "seed": 3,
158          "observed_mix_log_slope": -0.09078844601504561,
159          "predicted_fast_log_beta": -0.10536051565782628,
160          "predicted_slow_log_beta": -0.01005033585350145
161        }
162      ],
163      "confirmed": true
164    }
165  },
166  "custom_track": null
167}