Upload add_sub_baseline_100K_2L1H128d/metrics.json with huggingface_hub

add_sub_baseline_100K_2L1H128d/metrics.json

@@ -2582,502 +2582,567 @@
       "K": null,
       "mode": "sft",
       "n_digits": 6,
-      "n_per_split": 100
+      "n_per_split": 50
     },
     "splits": {
       "add_S0": {
-        "full_accuracy": 0.93,
-        "n_examples": 100,
+        "full_accuracy": 0.92,
+        "digit_accuracy": 0.9885714285714285,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9884297520661157,
-            "count": 605
+            "accuracy": 0.9864406779661017,
+            "count": 295
           },
           "SS": {
             "accuracy": 1.0,
-            "count": 95
+            "count": 55
           }
         }
       },
       "add_S1": {
-        "full_accuracy": 0.89,
-        "n_examples": 100,
+        "full_accuracy": 0.94,
+        "digit_accuracy": 0.9914285714285714,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9803921568627451,
-            "count": 204
+            "accuracy": 1.0,
+            "count": 126
           },
           "SC": {
-            "accuracy": 0.9881656804733728,
-            "count": 169
+            "accuracy": 0.9746835443037974,
+            "count": 79
           },
           "SS": {
-            "accuracy": 1.0,
-            "count": 31
+            "accuracy": 0.9523809523809523,
+            "count": 21
           },
           "UC": {
-            "accuracy": 0.9831081081081081,
-            "count": 296
+            "accuracy": 1.0,
+            "count": 124
           }
         }
       },
       "add_S2": {
-        "full_accuracy": 0.43,
-        "n_examples": 100,
+        "full_accuracy": 0.38,
+        "digit_accuracy": 0.8942857142857142,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9815950920245399,
-            "count": 163
+            "accuracy": 0.9733333333333334,
+            "count": 75
           },
           "SC": {
-            "accuracy": 0.9538461538461539,
-            "count": 130
+            "accuracy": 0.9032258064516129,
+            "count": 62
           },
           "SS": {
-            "accuracy": 0.9655172413793104,
-            "count": 87
+            "accuracy": 0.9743589743589743,
+            "count": 39
           },
           "UC": {
-            "accuracy": 0.729064039408867,
-            "count": 203
+            "accuracy": 0.7477477477477478,
+            "count": 111
           },
           "US": {
             "accuracy": 1.0,
-            "count": 117
+            "count": 63
           }
         }
       },
       "add_S3": {
-        "full_accuracy": 0.44,
-        "n_examples": 100,
+        "full_accuracy": 0.34,
+        "digit_accuracy": 0.8828571428571429,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9917355371900827,
-            "count": 121
+            "accuracy": 0.9833333333333333,
+            "count": 60
           },
           "SC": {
-            "accuracy": 0.9669421487603306,
-            "count": 121
+            "accuracy": 0.9649122807017544,
+            "count": 57
           },
           "SS": {
-            "accuracy": 1.0,
-            "count": 49
+            "accuracy": 0.8947368421052632,
+            "count": 19
           },
           "UC": {
-            "accuracy": 0.7849462365591398,
-            "count": 186
+            "accuracy": 0.7692307692307693,
+            "count": 104
           },
           "US": {
-            "accuracy": 0.8609865470852018,
-            "count": 223
+            "accuracy": 0.8909090909090909,
+            "count": 110
           }
         }
       },
       "add_S4": {
-        "full_accuracy": 0.4,
-        "n_examples": 100,
+        "full_accuracy": 0.42,
+        "digit_accuracy": 0.8228571428571428,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 104
+            "count": 48
           },
           "SC": {
-            "accuracy": 0.9716981132075472,
-            "count": 106
+            "accuracy": 0.9807692307692307,
+            "count": 52
           },
           "SS": {
             "accuracy": 1.0,
-            "count": 23
+            "count": 7
           },
           "UC": {
-            "accuracy": 0.7375,
-            "count": 160
+            "accuracy": 0.7191011235955056,
+            "count": 89
           },
           "US": {
-            "accuracy": 0.7035830618892508,
-            "count": 307
+            "accuracy": 0.7662337662337663,
+            "count": 154
           }
         }
       },
       "add_S5": {
-        "full_accuracy": 0.15,
-        "n_examples": 100,
+        "full_accuracy": 0.3,
+        "digit_accuracy": 0.6485714285714286,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "SC": {
-            "accuracy": 0.99,
-            "count": 100
+            "accuracy": 1.0,
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.32,
-            "count": 100
+            "accuracy": 0.36,
+            "count": 50
           },
           "US": {
-            "accuracy": 0.465,
-            "count": 400
+            "accuracy": 0.545,
+            "count": 200
           }
         }
       },
       "add_S6": {
-        "full_accuracy": 0.55,
-        "n_examples": 100,
+        "full_accuracy": 0.48,
+        "digit_accuracy": 0.7257142857142858,
+        "n_examples": 50,
         "per_subtask": {
           "SC": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.68,
-            "count": 100
+            "accuracy": 0.72,
+            "count": 50
           },
           "US": {
-            "accuracy": 0.704,
-            "count": 500
+            "accuracy": 0.672,
+            "count": 250
           }
         }
       },
       "add_random": {
-        "full_accuracy": 0.765,
+        "full_accuracy": 0.835,
+        "digit_accuracy": 0.9728571428571429,
         "n_examples": 200,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9798657718120806,
-            "count": 447
+            "accuracy": 0.9907192575406032,
+            "count": 431
           },
           "SC": {
-            "accuracy": 0.975,
-            "count": 320
+            "accuracy": 0.9873417721518988,
+            "count": 316
           },
           "SS": {
             "accuracy": 1.0,
-            "count": 56
+            "count": 39
           },
           "UC": {
-            "accuracy": 0.943289224952741,
-            "count": 529
+            "accuracy": 0.9517857142857142,
+            "count": 560
           },
           "US": {
-            "accuracy": 0.9791666666666666,
-            "count": 48
+            "accuracy": 0.9444444444444444,
+            "count": 54
+          }
+        }
+      },
+      "add_C1": {
+        "full_accuracy": 0.96,
+        "digit_accuracy": 0.9942857142857143,
+        "n_examples": 50,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.996,
+            "count": 250
+          },
+          "SC": {
+            "accuracy": 1.0,
+            "count": 50
+          },
+          "UC": {
+            "accuracy": 0.98,
+            "count": 50
+          }
+        }
+      },
+      "add_C2": {
+        "full_accuracy": 0.72,
+        "digit_accuracy": 0.9571428571428572,
+        "n_examples": 50,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.99,
+            "count": 200
+          },
+          "SC": {
+            "accuracy": 0.98,
+            "count": 50
+          },
+          "UC": {
+            "accuracy": 0.8554216867469879,
+            "count": 83
+          },
+          "US": {
+            "accuracy": 1.0,
+            "count": 17
           }
         }
       },
       "add_C3": {
-        "full_accuracy": 0.44,
-        "n_examples": 100,
+        "full_accuracy": 0.28,
+        "digit_accuracy": 0.8742857142857143,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 300
+            "count": 150
           },
           "SC": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.7202072538860104,
-            "count": 193
+            "accuracy": 0.66,
+            "count": 100
           },
           "US": {
-            "accuracy": 0.822429906542056,
-            "count": 107
+            "accuracy": 0.8,
+            "count": 50
           }
         }
       },
       "add_C4": {
         "full_accuracy": 0.38,
-        "n_examples": 100,
+        "digit_accuracy": 0.8942857142857142,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 200
+            "count": 100
           },
           "SC": {
-            "accuracy": 0.98,
-            "count": 100
+            "accuracy": 1.0,
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.7734375,
-            "count": 256
+            "accuracy": 0.7575757575757576,
+            "count": 132
           },
           "US": {
-            "accuracy": 0.8611111111111112,
-            "count": 144
+            "accuracy": 0.9264705882352942,
+            "count": 68
           }
         }
       },
       "add_C5": {
-        "full_accuracy": 0.33,
-        "n_examples": 100,
+        "full_accuracy": 0.48,
+        "digit_accuracy": 0.9028571428571428,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "SC": {
-            "accuracy": 0.99,
-            "count": 100
+            "accuracy": 0.98,
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.7908496732026143,
-            "count": 306
+            "accuracy": 0.821917808219178,
+            "count": 146
           },
           "US": {
-            "accuracy": 0.7938144329896907,
-            "count": 194
+            "accuracy": 0.9326923076923077,
+            "count": 104
           }
         }
       },
       "add_C6": {
-        "full_accuracy": 0.39,
-        "n_examples": 100,
+        "full_accuracy": 0.32,
+        "digit_accuracy": 0.8542857142857143,
+        "n_examples": 50,
         "per_subtask": {
           "SC": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.819672131147541,
-            "count": 366
+            "accuracy": 0.798941798941799,
+            "count": 189
           },
           "US": {
-            "accuracy": 0.8803418803418803,
-            "count": 234
+            "accuracy": 0.8828828828828829,
+            "count": 111
           }
         }
       },
       "sub_M0": {
-        "full_accuracy": 0.9,
-        "n_examples": 100,
+        "full_accuracy": 0.92,
+        "digit_accuracy": 0.9885714285714285,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9850249584026622,
-            "count": 601
+            "accuracy": 0.9867986798679867,
+            "count": 303
           },
           "ME": {
-            "accuracy": 0.9494949494949495,
-            "count": 99
+            "accuracy": 1.0,
+            "count": 47
           }
         }
       },
       "sub_M1": {
-        "full_accuracy": 0.89,
-        "n_examples": 100,
+        "full_accuracy": 0.86,
+        "digit_accuracy": 0.9771428571428571,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.985663082437276,
-            "count": 279
+            "accuracy": 1.0,
+            "count": 141
           },
           "MB": {
-            "accuracy": 0.9655172413793104,
-            "count": 145
+            "accuracy": 0.9444444444444444,
+            "count": 72
           },
           "ME": {
-            "accuracy": 1.0,
-            "count": 24
+            "accuracy": 0.9444444444444444,
+            "count": 18
           },
           "UB": {
-            "accuracy": 0.9920634920634921,
-            "count": 252
+            "accuracy": 0.9747899159663865,
+            "count": 119
           }
         }
       },
       "sub_M2": {
-        "full_accuracy": 0.3,
-        "n_examples": 100,
+        "full_accuracy": 0.18,
+        "digit_accuracy": 0.8628571428571429,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9671361502347418,
-            "count": 213
+            "accuracy": 0.9732142857142857,
+            "count": 112
           },
           "MB": {
-            "accuracy": 0.9646017699115044,
-            "count": 113
+            "accuracy": 0.9433962264150944,
+            "count": 53
           },
           "ME": {
-            "accuracy": 0.9764705882352941,
-            "count": 85
+            "accuracy": 0.9574468085106383,
+            "count": 47
           },
           "UB": {
-            "accuracy": 0.6243093922651933,
-            "count": 181
+            "accuracy": 0.5294117647058824,
+            "count": 85
           },
           "UD": {
-            "accuracy": 0.9907407407407407,
-            "count": 108
+            "accuracy": 1.0,
+            "count": 53
           }
         }
       },
       "sub_M3": {
-        "full_accuracy": 0.15,
-        "n_examples": 100,
+        "full_accuracy": 0.1,
+        "digit_accuracy": 0.8028571428571428,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.994413407821229,
-            "count": 179
+            "accuracy": 0.9896907216494846,
+            "count": 97
           },
           "MB": {
-            "accuracy": 0.9805825242718447,
-            "count": 103
+            "accuracy": 1.0,
+            "count": 51
           },
           "ME": {
             "accuracy": 1.0,
-            "count": 56
+            "count": 27
           },
           "UB": {
-            "accuracy": 0.5369127516778524,
-            "count": 149
+            "accuracy": 0.5405405405405406,
+            "count": 74
           },
           "UD": {
-            "accuracy": 0.6995305164319249,
-            "count": 213
+            "accuracy": 0.6633663366336634,
+            "count": 101
           }
         }
       },
       "sub_M4": {
-        "full_accuracy": 0.11,
-        "n_examples": 100,
+        "full_accuracy": 0.1,
+        "digit_accuracy": 0.7657142857142857,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 200
+            "count": 100
           },
           "MB": {
-            "accuracy": 0.99,
-            "count": 100
+            "accuracy": 1.0,
+            "count": 50
           },
           "UB": {
-            "accuracy": 0.35,
-            "count": 100
+            "accuracy": 0.46,
+            "count": 50
           },
           "UD": {
-            "accuracy": 0.5366666666666666,
-            "count": 300
+            "accuracy": 0.6333333333333333,
+            "count": 150
           }
         }
       },
       "sub_M5": {
         "full_accuracy": 0.14,
-        "n_examples": 100,
+        "digit_accuracy": 0.6857142857142857,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UB": {
-            "accuracy": 0.44,
-            "count": 100
+            "accuracy": 0.38,
+            "count": 50
           },
           "UD": {
-            "accuracy": 0.5275,
-            "count": 400
+            "accuracy": 0.605,
+            "count": 200
           }
         }
       },
       "sub_random": {
         "full_accuracy": 0.815,
+        "digit_accuracy": 0.9735714285714285,
         "n_examples": 200,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9883333333333333,
-            "count": 600
+            "accuracy": 0.9947368421052631,
+            "count": 570
           },
           "MB": {
-            "accuracy": 0.9850187265917603,
-            "count": 267
+            "accuracy": 0.9638989169675091,
+            "count": 277
           },
           "ME": {
             "accuracy": 1.0,
             "count": 53
           },
           "UB": {
-            "accuracy": 0.9362186788154897,
-            "count": 439
+            "accuracy": 0.9490445859872612,
+            "count": 471
           },
           "UD": {
-            "accuracy": 0.975609756097561,
-            "count": 41
+            "accuracy": 1.0,
+            "count": 29
           }
         }
       },
       "sub_B3": {
-        "full_accuracy": 0.42,
-        "n_examples": 100,
+        "full_accuracy": 0.58,
+        "digit_accuracy": 0.9171428571428571,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 300
+            "count": 150
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UB": {
-            "accuracy": 0.7258883248730964,
-            "count": 197
+            "accuracy": 0.801980198019802,
+            "count": 101
           },
           "UD": {
-            "accuracy": 0.912621359223301,
-            "count": 103
+            "accuracy": 0.8163265306122449,
+            "count": 49
           }
         }
       },
       "sub_B4": {
         "full_accuracy": 0.34,
-        "n_examples": 100,
+        "digit_accuracy": 0.8771428571428571,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 200
+            "count": 100
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UB": {
-            "accuracy": 0.7611336032388664,
-            "count": 247
+            "accuracy": 0.7520661157024794,
+            "count": 121
           },
           "UD": {
-            "accuracy": 0.7581699346405228,
-            "count": 153
+            "accuracy": 0.8354430379746836,
+            "count": 79
           }
         }
       },
       "sub_B5": {
-        "full_accuracy": 0.32,
-        "n_examples": 100,
+        "full_accuracy": 0.16,
+        "digit_accuracy": 0.8257142857142857,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UB": {
-            "accuracy": 0.761744966442953,
-            "count": 298
+            "accuracy": 0.7171052631578947,
+            "count": 152
           },
           "UD": {
-            "accuracy": 0.8415841584158416,
-            "count": 202
+            "accuracy": 0.8163265306122449,
+            "count": 98
           }
         }
       }
     },
     "summary": {
-      "overall_accuracy": 0.5025,
-      "total_examples": 2400,
-      "n_splits": 22
+      "overall_accuracy": 0.5626666666666666,
+      "digit_accuracy": 0.8972380952380953,
+      "total_examples": 1500,
+      "n_splits": 24
     }
   }
 }