Tempered-Stable Volatility Clock for Sequence Diffusion / 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": 0.0015,
 10      "weight_decay": 0.0001
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.0015,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 0.7337494194507599
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.0015,
 23          "weight_decay": 0.0001
 24        },
 25        "mean": 0.732865959405899
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.003,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 1.0512687861919403
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "weight_decay": 0.0001
 38        },
 39        "mean": 1.0608762055635452
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.006,
 44          "weight_decay": 0.0
 45        },
 46        "mean": 1.399800419807434
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.006,
 51          "weight_decay": 0.0001
 52        },
 53        "mean": 1.404996395111084
 54      }
 55    ],
 56    "full": {
 57      "mean": 0.7382896468043327,
 58      "std": 0.05677141706069671,
 59      "per_seed": [
 60        0.8005661368370056,
 61        0.7084230780601501,
 62        0.7421380281448364,
 63        0.680336594581604,
 64        0.6933529376983643,
 65        0.7014015913009644,
 66        0.8567653298377991,
 67        0.723333477973938
 68      ],
 69      "n": 8
 70    }
 71  },
 72  "idea": {
 73    "mean": 0.7525501698255539,
 74    "std": 0.053059163212739215,
 75    "per_seed": [
 76      0.8100458979606628,
 77      0.7514329552650452,
 78      0.7886123657226562,
 79      0.6894006133079529,
 80      0.7282218933105469,
 81      0.7097805738449097,
 82      0.844907820224762,
 83      0.6979992389678955
 84    ],
 85    "n": 8
 86  },
 87  "comparison": {
 88    "delta_mean": 0.014260523021221161,
 89    "idea_wins": 2,
 90    "n_pairs": 8,
 91    "per_seed_diffs": [
 92      0.009479761123657227,
 93      0.04300987720489502,
 94      0.046474337577819824,
 95      0.009064018726348877,
 96      0.03486895561218262,
 97      0.008378982543945312,
 98      -0.01185750961303711,
 99      -0.02533423900604248
100    ],
101    "p_value": 0.1657,
102    "mde": 0.021488876180993773,
103    "mde_rel_pct": 2.9106294899309244,
104    "verdict": "no significant win",
105    "system_worked": false
106  },
107  "mechanism_signature": {
108    "predicted_clock_excess_kurtosis": 0.7720588235294117,
109    "observed_clock_excess_kurtosis": 0.7388218939304352,
110    "predicted_clock_lag1_squared_acf": 0.0649867374005305,
111    "observed_clock_lag1_squared_acf": 0.4886988948518994,
112    "trained_model_response": [
113      {
114        "seed": 0,
115        "prediction_response_excess_kurtosis": 1.5980710983276367,
116        "response_sq_corr": 0.1433327302520934
117      },
118      {
119        "seed": 1,
120        "prediction_response_excess_kurtosis": 3.6790480613708496,
121        "response_sq_corr": 0.11140172653680885
122      },
123      {
124        "seed": 2,
125        "prediction_response_excess_kurtosis": 1.276343822479248,
126        "response_sq_corr": 0.3330819493478339
127      },
128      {
129        "seed": 3,
130        "prediction_response_excess_kurtosis": 0.395693302154541,
131        "response_sq_corr": 0.12126492127093162
132      },
133      {
134        "seed": 4,
135        "prediction_response_excess_kurtosis": 0.2646524906158447,
136        "response_sq_corr": 0.2837461795274339
137      },
138      {
139        "seed": 5,
140        "prediction_response_excess_kurtosis": 1.4415860176086426,
141        "response_sq_corr": -0.003209303770727274
142      },
143      {
144        "seed": 6,
145        "prediction_response_excess_kurtosis": 1.0070834159851074,
146        "response_sq_corr": 0.1870329129215599
147      },
148      {
149        "seed": 7,
150        "prediction_response_excess_kurtosis": 0.09569764137268066,
151        "response_sq_corr": 0.01287339296745192
152      }
153    ],
154    "response_mean_excess_kurtosis": 1.2197719812393188,
155    "confirmed": false
156  },
157  "idea_sweep": [
158    {
159      "cfg": {
160        "lr": 0.0015,
161        "weight_decay": 0.0001
162      },
163      "result": {
164        "mean": 0.7525501698255539,
165        "std": 0.053059163212739215,
166        "per_seed": [
167          0.8100458979606628,
168          0.7514329552650452,
169          0.7886123657226562,
170          0.6894006133079529,
171          0.7282218933105469,
172          0.7097805738449097,
173          0.844907820224762,
174          0.6979992389678955
175        ],
176        "n": 8
177      }
178    },
179    {
180      "cfg": {
181        "lr": 0.0015,
182        "weight_decay": 0.0001
183      },
184      "result": {
185        "mean": 0.7525501698255539,
186        "std": 0.053059163212739215,
187        "per_seed": [
188          0.8100458979606628,
189          0.7514329552650452,
190          0.7886123657226562,
191          0.6894006133079529,
192          0.7282218933105469,
193          0.7097805738449097,
194          0.844907820224762,
195          0.6979992389678955
196        ],
197        "n": 8
198      }
199    },
200    {
201      "cfg": {
202        "lr": 0.006,
203        "weight_decay": 0.0001
204      },
205      "result": {
206        "mean": 1.2776761278510094,
207        "std": 0.5370590006921943,
208        "per_seed": [
209          1.019742488861084,
210          2.5213093757629395,
211          0.7612278461456299,
212          1.4226016998291016,
213          1.0408846139907837,
214          0.8120333552360535,
215          1.0670547485351562,
216          1.5765548944473267
217        ],
218        "n": 8
219      }
220    }
221  ],
222  "protocol_notes": {
223    "structural_match": "sequence windows have multi-position correlations; blind clock augmentation is the only intervention",
224    "epochs": 5,
225    "n_train": 400,
226    "n_test": 400,
227    "clock": {
228      "alpha": 0.65,
229      "theta": 0.8,
230      "phi": 0.7
231    }
232  }
233}