Orbit-Consistent Equivariant Distillation / results.json

Mechanism failed

Raw ⬇ ZIP
  1{
  2  "device": "cuda",
  3  "seed": 2854,
  4  "core_math": {
  5    "target_equivariance_max_abs": 0.0,
  6    "target_value_invariance_max_abs": 0.0
  7  },
  8  "predictions": {
  9    "P1": "exact target residuals should be zero",
 10    "P2": "eq_error decreases with lambda_eq",
 11    "P3": "value_invariance decreases with lambda_v"
 12  },
 13  "equivariance_sweep": [
 14    {
 15      "lambda_eq": 0.0,
 16      "eq_error": 0.015910927206277847,
 17      "value_invariance": 0.00923738069832325,
 18      "canonical_mse": 0.00015656309551559389,
 19      "transformed_mse": 0.0001599600655026734
 20    },
 21    {
 22      "lambda_eq": 0.03,
 23      "eq_error": 0.013291101902723312,
 24      "value_invariance": 0.008584478870034218,
 25      "canonical_mse": 0.00010319300781702623,
 26      "transformed_mse": 0.00010051406570710242
 27    },
 28    {
 29      "lambda_eq": 0.1,
 30      "eq_error": 0.011657794937491417,
 31      "value_invariance": 0.007666226476430893,
 32      "canonical_mse": 0.0001495519099989906,
 33      "transformed_mse": 0.00015336080105043948
 34    },
 35    {
 36      "lambda_eq": 0.3,
 37      "eq_error": 0.006351675372570753,
 38      "value_invariance": 0.010713014751672745,
 39      "canonical_mse": 0.00012270035222172737,
 40      "transformed_mse": 0.0001229585031978786
 41    },
 42    {
 43      "lambda_eq": 1.0,
 44      "eq_error": 0.022295624017715454,
 45      "value_invariance": 0.012789005413651466,
 46      "canonical_mse": 0.00019733374938368797,
 47      "transformed_mse": 0.00020098048844374716
 48    }
 49  ],
 50  "value_sweep": [
 51    {
 52      "lambda_v": 0.0,
 53      "eq_error": 0.005819602403789759,
 54      "value_invariance": 0.007328704930841923,
 55      "canonical_mse": 0.00013409263920038939,
 56      "transformed_mse": 0.00013548567949328572
 57    },
 58    {
 59      "lambda_v": 0.03,
 60      "eq_error": 0.015905272215604782,
 61      "value_invariance": 0.010465560480952263,
 62      "canonical_mse": 0.00017110300541389734,
 63      "transformed_mse": 0.0001707408227957785
 64    },
 65    {
 66      "lambda_v": 0.1,
 67      "eq_error": 0.016019191592931747,
 68      "value_invariance": 0.007124156691133976,
 69      "canonical_mse": 0.00010748900240287185,
 70      "transformed_mse": 0.00010836163710337132
 71    },
 72    {
 73      "lambda_v": 0.3,
 74      "eq_error": 0.01917118951678276,
 75      "value_invariance": 0.003987547941505909,
 76      "canonical_mse": 0.00020443039829842746,
 77      "transformed_mse": 0.00020101213885936886
 78    }
 79  ],
 80  "checks": {
 81    "p1_pass": true,
 82    "p2_monotone_pass": false,
 83    "p2_fold_reduction": 0.7136345317644165,
 84    "p2_expected_at_least_5x": false,
 85    "p3_monotone_pass": false
 86  },
 87  "baseline": {
 88    "lambda_eq": 0.0,
 89    "eq_error": 0.015910927206277847,
 90    "value_invariance": 0.00923738069832325,
 91    "canonical_mse": 0.00015656309551559389,
 92    "transformed_mse": 0.0001599600655026734
 93  },
 94  "idea": {
 95    "lambda_eq": 1.0,
 96    "eq_error": 0.022295624017715454,
 97    "value_invariance": 0.012789005413651466,
 98    "canonical_mse": 0.00019733374938368797,
 99    "transformed_mse": 0.00020098048844374716
100  }
101}