Prolate Energy-Preserving Bottleneck / 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      "rank": 32
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "rank": 32
 17        },
 18        "mean": 0.3797149360179901
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "rank": 32
 24        },
 25        "mean": 0.5440608412027359
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.005,
 30          "rank": 32
 31        },
 32        "mean": 0.6877395510673523
 33      }
 34    ],
 35    "full": {
 36      "mean": 0.3990439400076866,
 37      "std": 0.03555091686816573,
 38      "per_seed": [
 39        0.40484797954559326,
 40        0.35254013538360596,
 41        0.3537353277206421,
 42        0.40773630142211914,
 43        0.3955448567867279,
 44        0.39521750807762146,
 45        0.4743589758872986,
 46        0.4083704352378845
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "mean": 0.4859609939157963,
 53    "std": 0.030690392503688,
 54    "per_seed": [
 55      0.5398046374320984,
 56      0.4721372127532959,
 57      0.4599805176258087,
 58      0.4985331892967224,
 59      0.46909099817276,
 60      0.45167359709739685,
 61      0.5286514163017273,
 62      0.46781638264656067
 63    ],
 64    "n": 8
 65  },
 66  "comparison": {
 67    "delta_mean": 0.08691705390810966,
 68    "idea_wins": 0,
 69    "n_pairs": 8,
 70    "per_seed_diffs": [
 71      0.13495665788650513,
 72      0.11959707736968994,
 73      0.10624518990516663,
 74      0.09079688787460327,
 75      0.0735461413860321,
 76      0.05645608901977539,
 77      0.05429244041442871,
 78      0.05944594740867615
 79    ],
 80    "p_value": 0.0081,
 81    "mde": 0.025849569428390114,
 82    "mde_rel_pct": 6.477875450982211,
 83    "verdict": "idea worse (significant)",
 84    "system_worked": false
 85  },
 86  "idea_sweep": [
 87    {
 88      "cfg": {
 89        "lr": 0.001,
 90        "rank": 12
 91      },
 92      "result": {
 93        "mean": 0.8132649660110474,
 94        "std": 0.07039976968716767,
 95        "per_seed": [
 96          0.8669847846031189,
 97          0.6813386082649231,
 98          0.7922707200050354,
 99          0.7799481153488159,
100          0.8783856630325317,
101          0.7520660758018494,
102          0.8496934175491333,
103          0.9054323434829712
104        ],
105        "n": 8
106      }
107    },
108    {
109      "cfg": {
110        "lr": 0.001,
111        "rank": 16
112      },
113      "result": {
114        "mean": 0.6447810083627701,
115        "std": 0.0902293737866928,
116        "per_seed": [
117          0.8473581075668335,
118          0.532031774520874,
119          0.5603737235069275,
120          0.6733526587486267,
121          0.6476457715034485,
122          0.6398608684539795,
123          0.6668751835823059,
124          0.590749979019165
125        ],
126        "n": 8
127      }
128    },
129    {
130      "cfg": {
131        "lr": 0.001,
132        "rank": 20
133      },
134      "result": {
135        "mean": 0.4859609939157963,
136        "std": 0.030690392503688,
137        "per_seed": [
138          0.5398046374320984,
139          0.4721372127532959,
140          0.4599805176258087,
141          0.4985331892967224,
142          0.46909099817276,
143          0.45167359709739685,
144          0.5286514163017273,
145          0.46781638264656067
146        ],
147        "n": 8
148      }
149    },
150    {
151      "cfg": {
152        "lr": 0.003,
153        "rank": 12
154      },
155      "result": {
156        "mean": 0.8662505745887756,
157        "std": 0.08758038922827249,
158        "per_seed": [
159          0.9283100366592407,
160          0.69541335105896,
161          0.9051690101623535,
162          0.8301766514778137,
163          0.9451277852058411,
164          0.7725451588630676,
165          0.9667305946350098,
166          0.8865320086479187
167        ],
168        "n": 8
169      }
170    },
171    {
172      "cfg": {
173        "lr": 0.003,
174        "rank": 16
175      },
176      "result": {
177        "mean": 0.6834343373775482,
178        "std": 0.09683352283532777,
179        "per_seed": [
180          0.8036410808563232,
181          0.5426891446113586,
182          0.5894339084625244,
183          0.8408225774765015,
184          0.6754513382911682,
185          0.6724597215652466,
186          0.7292736768722534,
187          0.6137032508850098
188        ],
189        "n": 8
190      }
191    },
192    {
193      "cfg": {
194        "lr": 0.003,
195        "rank": 20
196      },
197      "result": {
198        "mean": 0.5512603707611561,
199        "std": 0.09140216767338745,
200        "per_seed": [
201          0.7226142287254333,
202          0.4266541600227356,
203          0.5001299381256104,
204          0.5239855051040649,
205          0.459748238325119,
206          0.5338931083679199,
207          0.6257912516593933,
208          0.6172665357589722
209        ],
210        "n": 8
211      }
212    },
213    {
214      "cfg": {
215        "lr": 0.005,
216        "rank": 12
217      },
218      "result": {
219        "mean": 0.9177583232522011,
220        "std": 0.11157952187308297,
221        "per_seed": [
222          1.0815995931625366,
223          0.7145638465881348,
224          0.9597082734107971,
225          0.8518447875976562,
226          1.0344592332839966,
227          0.8185186982154846,
228          0.9612360000610352,
229          0.9201361536979675
230        ],
231        "n": 8
232      }
233    },
234    {
235      "cfg": {
236        "lr": 0.005,
237        "rank": 16
238      },
239      "result": {
240        "mean": 0.768573522567749,
241        "std": 0.1272932016139664,
242        "per_seed": [
243          0.9035587906837463,
244          0.5587367415428162,
245          0.6324987411499023,
246          0.8873728513717651,
247          0.7445271611213684,
248          0.9305623769760132,
249          0.804309606552124,
250          0.6870219111442566
251        ],
252        "n": 8
253      }
254    },
255    {
256      "cfg": {
257        "lr": 0.005,
258        "rank": 20
259      },
260      "result": {
261        "mean": 0.5918549560010433,
262        "std": 0.11067653453007016,
263        "per_seed": [
264          0.8396977186203003,
265          0.4893060028553009,
266          0.5174769163131714,
267          0.6687707304954529,
268          0.5009192824363708,
269          0.5292778015136719,
270          0.6293551921844482,
271          0.5600360035896301
272        ],
273        "n": 8
274      }
275    }
276  ],
277  "audit": {
278    "structural_match": "multi-token sequence forecast",
279    "theorem_rank_asymptotic": 15.311856320206266,
280    "empirical_rank": 15,
281    "chosen_rank_grid": [
282      12,
283      16,
284      20
285    ],
286    "epochs": 8,
287    "n_train": 800,
288    "n_test": 300
289  },
290  "mechanism_signature": {
291    "prediction": "DPSS reconstruction error < Fourier at theorem rank",
292    "predicted_dpss_lt_fourier": true,
293    "observed_dpss_error_mean": 0.21955212997938756,
294    "observed_fourier_error_mean": 0.20034139348637556,
295    "observed_relative_improvement": -0.0958900013557031,
296    "attention_work_fraction": 0.25,
297    "trained_model_samples": [
298      {
299        "seed": 0,
300        "dpss_reconstruction_error": 0.22673050321714594,
301        "fourier_reconstruction_error": 0.20555503007570636,
302        "prediction_rms": 0.6146740913391113
303      },
304      {
305        "seed": 1,
306        "dpss_reconstruction_error": 0.221470346993945,
307        "fourier_reconstruction_error": 0.1998581531341289,
308        "prediction_rms": 0.6230881810188293
309      },
310      {
311        "seed": 2,
312        "dpss_reconstruction_error": 0.19929562370165232,
313        "fourier_reconstruction_error": 0.18154581981703843,
314        "prediction_rms": 0.6864827871322632
315      },
316      {
317        "seed": 3,
318        "dpss_reconstruction_error": 0.2255786421582673,
319        "fourier_reconstruction_error": 0.210578079503818,
320        "prediction_rms": 0.7499093413352966
321      },
322      {
323        "seed": 4,
324        "dpss_reconstruction_error": 0.2059191288393489,
325        "fourier_reconstruction_error": 0.1945450930599946,
326        "prediction_rms": 0.6713075637817383
327      },
328      {
329        "seed": 5,
330        "dpss_reconstruction_error": 0.21693499314985126,
331        "fourier_reconstruction_error": 0.19918181458416848,
332        "prediction_rms": 0.580341637134552
333      },
334      {
335        "seed": 6,
336        "dpss_reconstruction_error": 0.2512841631045155,
337        "fourier_reconstruction_error": 0.22277636588465038,
338        "prediction_rms": 0.654333770275116
339      },
340      {
341        "seed": 7,
342        "dpss_reconstruction_error": 0.2092036386703743,
343        "fourier_reconstruction_error": 0.1886907918314992,
344        "prediction_rms": 0.7088998556137085
345      }
346    ],
347    "confirmed": false
348  }
349}