Slow-Mode-Canceling Optimizer Packet / 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": 0.006,
 10      "weight_decay": 0.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 17.79831099510193
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "weight_decay": 0.0
 24        },
 25        "mean": 13.153547525405884
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.006,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 9.905518293380737
 33      }
 34    ],
 35    "full": {
 36      "mean": 10.041840195655823,
 37      "std": 1.3433768404487412,
 38      "per_seed": [
 39        10.178184509277344,
 40        11.6652193069458,
 41        8.723758697509766,
 42        9.054910659790039,
 43        7.620545387268066,
 44        10.61941909790039,
 45        11.13137149810791,
 46        11.341312408447266
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "mean": 7.694764673709869,
 53    "std": 0.9183754203462169,
 54    "per_seed": [
 55      7.17905855178833,
 56      8.86394214630127,
 57      7.2537841796875,
 58      6.873910903930664,
 59      6.19173526763916,
 60      8.16868782043457,
 61      8.054282188415527,
 62      8.972716331481934
 63    ],
 64    "n": 8
 65  },
 66  "comparison": {
 67    "delta_mean": -2.3470755219459534,
 68    "idea_wins": 8,
 69    "n_pairs": 8,
 70    "per_seed_diffs": [
 71      -2.9991259574890137,
 72      -2.8012771606445312,
 73      -1.4699745178222656,
 74      -2.180999755859375,
 75      -1.4288101196289062,
 76      -2.4507312774658203,
 77      -3.077089309692383,
 78      -2.368596076965332
 79    ],
 80    "p_value": 0.0081,
 81    "mde": 0.5303101801167621,
 82    "mde_rel_pct": 5.281005968867921,
 83    "verdict": "idea better (significant)",
 84    "system_worked": true
 85  },
 86  "mechanism_signature": {
 87    "predicted": "packet preparation makes common slow-direction projection near zero",
 88    "observed": {
 89      "initial_cancellation_ratio": 3.7172168874545825e-10,
 90      "observed_c_ratio_epoch1": 0.21678468534656278,
 91      "observed_c_ratio_final": 1.042262186528141,
 92      "v_norm": 1.0
 93    },
 94    "confirmed": true,
 95    "idea_sweep": [
 96      {
 97        "cfg": {
 98          "lr": 0.001,
 99          "kappa": 0.01
100        },
101        "mean": 20.009164333343506,
102        "result": {
103          "mean": 20.009164333343506,
104          "std": 1.230693031020796,
105          "per_seed": [
106            22.104595184326172,
107            19.132362365722656,
108            19.679386138916016,
109            19.12031364440918
110          ],
111          "n": 4
112        }
113      },
114      {
115        "cfg": {
116          "lr": 0.001,
117          "kappa": 0.03
118        },
119        "mean": 19.979770183563232,
120        "result": {
121          "mean": 19.979770183563232,
122          "std": 1.2766088175400236,
123          "per_seed": [
124            22.140043258666992,
125            18.95973014831543,
126            19.69390106201172,
127            19.12540626525879
128          ],
129          "n": 4
130        }
131      },
132      {
133        "cfg": {
134          "lr": 0.001,
135          "kappa": 0.06
136        },
137        "mean": 19.88680124282837,
138        "result": {
139          "mean": 19.88680124282837,
140          "std": 1.3482576598501705,
141          "per_seed": [
142            22.143774032592773,
143            18.91352653503418,
144            19.69554328918457,
145            18.794361114501953
146          ],
147          "n": 4
148        }
149      },
150      {
151        "cfg": {
152          "lr": 0.003,
153          "kappa": 0.01
154        },
155        "mean": 13.156998634338379,
156        "result": {
157          "mean": 13.156998634338379,
158          "std": 0.7376469596332139,
159          "per_seed": [
160            13.269169807434082,
161            13.964943885803223,
162            13.43444538116455,
163            11.95943546295166
164          ],
165          "n": 4
166        }
167      },
168      {
169        "cfg": {
170          "lr": 0.003,
171          "kappa": 0.03
172        },
173        "mean": 13.096021890640259,
174        "result": {
175          "mean": 13.096021890640259,
176          "std": 0.7681500483182941,
177          "per_seed": [
178            13.330429077148438,
179            13.886496543884277,
180            13.34359073638916,
181            11.82357120513916
182          ],
183          "n": 4
184        }
185      },
186      {
187        "cfg": {
188          "lr": 0.003,
189          "kappa": 0.06
190        },
191        "mean": 13.155548334121704,
192        "result": {
193          "mean": 13.155548334121704,
194          "std": 0.7771499510902515,
195          "per_seed": [
196            13.34993839263916,
197            13.988134384155273,
198            13.402823448181152,
199            11.88129711151123
200          ],
201          "n": 4
202        }
203      },
204      {
205        "cfg": {
206          "lr": 0.006,
207          "kappa": 0.01
208        },
209        "mean": 7.5606971979141235,
210        "result": {
211          "mean": 7.5606971979141235,
212          "std": 0.7941976381676852,
213          "per_seed": [
214            7.124575614929199,
215            8.919793128967285,
216            7.272003173828125,
217            6.926416873931885
218          ],
219          "n": 4
220        }
221      },
222      {
223        "cfg": {
224          "lr": 0.006,
225          "kappa": 0.03
226        },
227        "mean": 7.542673945426941,
228        "result": {
229          "mean": 7.542673945426941,
230          "std": 0.7759940585392692,
231          "per_seed": [
232            7.17905855178833,
233            8.86394214630127,
234            7.2537841796875,
235            6.873910903930664
236          ],
237          "n": 4
238        }
239      },
240      {
241        "cfg": {
242          "lr": 0.006,
243          "kappa": 0.06
244        },
245        "mean": 7.578917741775513,
246        "result": {
247          "mean": 7.578917741775513,
248          "std": 0.7888582333579148,
249          "per_seed": [
250            7.209147930145264,
251            8.925769805908203,
252            7.265168190002441,
253            6.915585041046143
254          ],
255          "n": 4
256        }
257      }
258    ]
259  }
260}