Upload add_sub_baseline_50K_1L3H510d/metrics.json with huggingface_hub

add_sub_baseline_50K_1L3H510d/metrics.json

@@ -1330,502 +1330,567 @@
       "K": null,
       "mode": "sft",
       "n_digits": 6,
-      "n_per_split": 100
+      "n_per_split": 50
     },
     "splits": {
       "add_S0": {
-        "full_accuracy": 0.94,
-        "n_examples": 100,
+        "full_accuracy": 0.92,
+        "digit_accuracy": 0.9885714285714285,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9900826446280991,
-            "count": 605
+            "accuracy": 0.9864406779661017,
+            "count": 295
           },
           "SS": {
             "accuracy": 1.0,
-            "count": 95
+            "count": 55
           }
         }
       },
       "add_S1": {
-        "full_accuracy": 0.94,
-        "n_examples": 100,
+        "full_accuracy": 0.88,
+        "digit_accuracy": 0.9828571428571429,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9901960784313726,
-            "count": 204
+            "accuracy": 1.0,
+            "count": 126
           },
           "SC": {
-            "accuracy": 0.9940828402366864,
-            "count": 169
+            "accuracy": 0.9746835443037974,
+            "count": 79
           },
           "SS": {
-            "accuracy": 1.0,
-            "count": 31
+            "accuracy": 0.9523809523809523,
+            "count": 21
           },
           "UC": {
-            "accuracy": 0.9898648648648649,
-            "count": 296
+            "accuracy": 0.9758064516129032,
+            "count": 124
           }
         }
       },
       "add_S2": {
         "full_accuracy": 0.64,
-        "n_examples": 100,
+        "digit_accuracy": 0.9428571428571428,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9877300613496932,
-            "count": 163
+            "accuracy": 0.9866666666666667,
+            "count": 75
           },
           "SC": {
-            "accuracy": 0.9384615384615385,
-            "count": 130
+            "accuracy": 0.9516129032258065,
+            "count": 62
           },
           "SS": {
-            "accuracy": 0.9540229885057471,
-            "count": 87
+            "accuracy": 0.9743589743589743,
+            "count": 39
           },
           "UC": {
-            "accuracy": 0.8669950738916257,
-            "count": 203
+            "accuracy": 0.8648648648648649,
+            "count": 111
           },
           "US": {
             "accuracy": 1.0,
-            "count": 117
+            "count": 63
           }
         }
       },
       "add_S3": {
-        "full_accuracy": 0.3,
-        "n_examples": 100,
+        "full_accuracy": 0.44,
+        "digit_accuracy": 0.8771428571428571,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9917355371900827,
-            "count": 121
+            "accuracy": 1.0,
+            "count": 60
           },
           "SC": {
-            "accuracy": 0.9421487603305785,
-            "count": 121
+            "accuracy": 1.0,
+            "count": 57
           },
           "SS": {
-            "accuracy": 0.9795918367346939,
-            "count": 49
+            "accuracy": 1.0,
+            "count": 19
           },
           "UC": {
-            "accuracy": 0.6935483870967742,
-            "count": 186
+            "accuracy": 0.7692307692307693,
+            "count": 104
           },
           "US": {
-            "accuracy": 0.820627802690583,
-            "count": 223
+            "accuracy": 0.8272727272727273,
+            "count": 110
           }
         }
       },
       "add_S4": {
-        "full_accuracy": 0.31,
-        "n_examples": 100,
+        "full_accuracy": 0.22,
+        "digit_accuracy": 0.74,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 104
+            "count": 48
           },
           "SC": {
-            "accuracy": 1.0,
-            "count": 106
+            "accuracy": 0.9807692307692307,
+            "count": 52
           },
           "SS": {
             "accuracy": 1.0,
-            "count": 23
+            "count": 7
           },
           "UC": {
-            "accuracy": 0.65625,
-            "count": 160
+            "accuracy": 0.6404494382022472,
+            "count": 89
           },
           "US": {
-            "accuracy": 0.6351791530944625,
-            "count": 307
+            "accuracy": 0.6233766233766234,
+            "count": 154
           }
         }
       },
       "add_S5": {
-        "full_accuracy": 0.26,
-        "n_examples": 100,
+        "full_accuracy": 0.18,
+        "digit_accuracy": 0.5971428571428572,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "SC": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UC": {
             "accuracy": 0.42,
-            "count": 100
+            "count": 50
           },
           "US": {
-            "accuracy": 0.525,
-            "count": 400
+            "accuracy": 0.44,
+            "count": 200
           }
         }
       },
       "add_S6": {
-        "full_accuracy": 0.45,
-        "n_examples": 100,
+        "full_accuracy": 0.5,
+        "digit_accuracy": 0.6485714285714286,
+        "n_examples": 50,
         "per_subtask": {
           "SC": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.58,
-            "count": 100
+            "accuracy": 0.54,
+            "count": 50
           },
           "US": {
-            "accuracy": 0.594,
-            "count": 500
+            "accuracy": 0.6,
+            "count": 250
           }
         }
       },
       "add_random": {
-        "full_accuracy": 0.88,
+        "full_accuracy": 0.89,
+        "digit_accuracy": 0.9814285714285714,
         "n_examples": 200,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9910514541387024,
-            "count": 447
+            "accuracy": 1.0,
+            "count": 431
           },
           "SC": {
-            "accuracy": 0.990625,
-            "count": 320
+            "accuracy": 1.0,
+            "count": 316
           },
           "SS": {
-            "accuracy": 0.9464285714285714,
-            "count": 56
+            "accuracy": 1.0,
+            "count": 39
           },
           "UC": {
-            "accuracy": 0.9716446124763705,
-            "count": 529
+            "accuracy": 0.9607142857142857,
+            "count": 560
           },
           "US": {
-            "accuracy": 0.9791666666666666,
-            "count": 48
+            "accuracy": 0.9259259259259259,
+            "count": 54
+          }
+        }
+      },
+      "add_C1": {
+        "full_accuracy": 0.94,
+        "digit_accuracy": 0.9914285714285714,
+        "n_examples": 50,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.996,
+            "count": 250
+          },
+          "SC": {
+            "accuracy": 1.0,
+            "count": 50
+          },
+          "UC": {
+            "accuracy": 0.96,
+            "count": 50
+          }
+        }
+      },
+      "add_C2": {
+        "full_accuracy": 0.84,
+        "digit_accuracy": 0.9742857142857143,
+        "n_examples": 50,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 1.0,
+            "count": 200
+          },
+          "SC": {
+            "accuracy": 1.0,
+            "count": 50
+          },
+          "UC": {
+            "accuracy": 0.9036144578313253,
+            "count": 83
+          },
+          "US": {
+            "accuracy": 0.9411764705882353,
+            "count": 17
           }
         }
       },
       "add_C3": {
-        "full_accuracy": 0.53,
-        "n_examples": 100,
+        "full_accuracy": 0.62,
+        "digit_accuracy": 0.9228571428571428,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9866666666666667,
-            "count": 300
+            "accuracy": 0.9933333333333333,
+            "count": 150
           },
           "SC": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.7927461139896373,
-            "count": 193
+            "accuracy": 0.85,
+            "count": 100
           },
           "US": {
-            "accuracy": 0.8037383177570093,
-            "count": 107
+            "accuracy": 0.78,
+            "count": 50
           }
         }
       },
       "add_C4": {
-        "full_accuracy": 0.5,
-        "n_examples": 100,
+        "full_accuracy": 0.66,
+        "digit_accuracy": 0.9257142857142857,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 1.0,
-            "count": 200
+            "accuracy": 0.99,
+            "count": 100
           },
           "SC": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.8046875,
-            "count": 256
+            "accuracy": 0.8787878787878788,
+            "count": 132
           },
           "US": {
-            "accuracy": 0.8055555555555556,
-            "count": 144
+            "accuracy": 0.8676470588235294,
+            "count": 68
           }
         }
       },
       "add_C5": {
-        "full_accuracy": 0.51,
-        "n_examples": 100,
+        "full_accuracy": 0.54,
+        "digit_accuracy": 0.8742857142857143,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "SC": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.8496732026143791,
-            "count": 306
+            "accuracy": 0.8493150684931506,
+            "count": 146
           },
           "US": {
-            "accuracy": 0.8402061855670103,
-            "count": 194
+            "accuracy": 0.7884615384615384,
+            "count": 104
           }
         }
       },
       "add_C6": {
-        "full_accuracy": 0.44,
-        "n_examples": 100,
+        "full_accuracy": 0.5,
+        "digit_accuracy": 0.8771428571428571,
+        "n_examples": 50,
         "per_subtask": {
           "SC": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.8579234972677595,
-            "count": 366
+            "accuracy": 0.8571428571428571,
+            "count": 189
           },
           "US": {
-            "accuracy": 0.8717948717948718,
-            "count": 234
+            "accuracy": 0.8558558558558559,
+            "count": 111
           }
         }
       },
       "sub_M0": {
-        "full_accuracy": 0.88,
-        "n_examples": 100,
+        "full_accuracy": 0.9,
+        "digit_accuracy": 0.9857142857142858,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9800332778702163,
-            "count": 601
+            "accuracy": 0.9834983498349835,
+            "count": 303
           },
           "ME": {
             "accuracy": 1.0,
-            "count": 99
+            "count": 47
           }
         }
       },
       "sub_M1": {
-        "full_accuracy": 0.92,
-        "n_examples": 100,
+        "full_accuracy": 0.86,
+        "digit_accuracy": 0.98,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.985663082437276,
-            "count": 279
+            "accuracy": 0.9858156028368794,
+            "count": 141
           },
           "MB": {
-            "accuracy": 0.9862068965517241,
-            "count": 145
+            "accuracy": 0.9861111111111112,
+            "count": 72
           },
           "ME": {
             "accuracy": 1.0,
-            "count": 24
+            "count": 18
           },
           "UB": {
-            "accuracy": 0.9920634920634921,
-            "count": 252
+            "accuracy": 0.9663865546218487,
+            "count": 119
           }
         }
       },
       "sub_M2": {
-        "full_accuracy": 0.54,
-        "n_examples": 100,
+        "full_accuracy": 0.56,
+        "digit_accuracy": 0.9257142857142857,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9906103286384976,
-            "count": 213
+            "accuracy": 0.9732142857142857,
+            "count": 112
           },
           "MB": {
-            "accuracy": 0.9823008849557522,
-            "count": 113
+            "accuracy": 0.9622641509433962,
+            "count": 53
           },
           "ME": {
             "accuracy": 1.0,
-            "count": 85
+            "count": 47
           },
           "UB": {
-            "accuracy": 0.7624309392265194,
-            "count": 181
+            "accuracy": 0.7647058823529411,
+            "count": 85
           },
           "UD": {
-            "accuracy": 1.0,
-            "count": 108
+            "accuracy": 0.9811320754716981,
+            "count": 53
           }
         }
       },
       "sub_M3": {
         "full_accuracy": 0.22,
-        "n_examples": 100,
+        "digit_accuracy": 0.8457142857142858,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 1.0,
-            "count": 179
+            "accuracy": 0.979381443298969,
+            "count": 97
           },
           "MB": {
-            "accuracy": 1.0,
-            "count": 103
+            "accuracy": 0.9803921568627451,
+            "count": 51
           },
           "ME": {
             "accuracy": 1.0,
-            "count": 56
+            "count": 27
           },
           "UB": {
-            "accuracy": 0.5570469798657718,
-            "count": 149
+            "accuracy": 0.5945945945945946,
+            "count": 74
           },
           "UD": {
-            "accuracy": 0.8450704225352113,
-            "count": 213
+            "accuracy": 0.7920792079207921,
+            "count": 101
           }
         }
       },
       "sub_M4": {
-        "full_accuracy": 0.02,
-        "n_examples": 100,
+        "full_accuracy": 0.04,
+        "digit_accuracy": 0.6771428571428572,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 200
+            "count": 100
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UB": {
-            "accuracy": 0.43,
-            "count": 100
+            "accuracy": 0.4,
+            "count": 50
           },
           "UD": {
-            "accuracy": 0.38333333333333336,
-            "count": 300
+            "accuracy": 0.44666666666666666,
+            "count": 150
           }
         }
       },
       "sub_M5": {
-        "full_accuracy": 0.08,
-        "n_examples": 100,
+        "full_accuracy": 0.1,
+        "digit_accuracy": 0.5771428571428572,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UB": {
-            "accuracy": 0.51,
-            "count": 100
+            "accuracy": 0.58,
+            "count": 50
           },
           "UD": {
-            "accuracy": 0.4075,
-            "count": 400
+            "accuracy": 0.365,
+            "count": 200
           }
         }
       },
       "sub_random": {
-        "full_accuracy": 0.86,
+        "full_accuracy": 0.87,
+        "digit_accuracy": 0.98,
         "n_examples": 200,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.985,
-            "count": 600
+            "accuracy": 0.9982456140350877,
+            "count": 570
           },
           "MB": {
-            "accuracy": 0.9925093632958801,
-            "count": 267
+            "accuracy": 0.9819494584837545,
+            "count": 277
           },
           "ME": {
             "accuracy": 1.0,
             "count": 53
           },
           "UB": {
-            "accuracy": 0.9567198177676538,
-            "count": 439
+            "accuracy": 0.9554140127388535,
+            "count": 471
           },
           "UD": {
-            "accuracy": 1.0,
-            "count": 41
+            "accuracy": 0.9655172413793104,
+            "count": 29
           }
         }
       },
       "sub_B3": {
-        "full_accuracy": 0.46,
-        "n_examples": 100,
+        "full_accuracy": 0.52,
+        "digit_accuracy": 0.9142857142857143,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 300
+            "count": 150
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UB": {
-            "accuracy": 0.7411167512690355,
-            "count": 197
+            "accuracy": 0.801980198019802,
+            "count": 101
           },
           "UD": {
-            "accuracy": 0.7669902912621359,
-            "count": 103
+            "accuracy": 0.7959183673469388,
+            "count": 49
           }
         }
       },
       "sub_B4": {
-        "full_accuracy": 0.3,
-        "n_examples": 100,
+        "full_accuracy": 0.34,
+        "digit_accuracy": 0.8485714285714285,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 200
+            "count": 100
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UB": {
-            "accuracy": 0.7327935222672065,
-            "count": 247
+            "accuracy": 0.7603305785123967,
+            "count": 121
           },
           "UD": {
-            "accuracy": 0.7450980392156863,
-            "count": 153
+            "accuracy": 0.6962025316455697,
+            "count": 79
           }
         }
       },
       "sub_B5": {
-        "full_accuracy": 0.27,
-        "n_examples": 100,
+        "full_accuracy": 0.22,
+        "digit_accuracy": 0.8114285714285714,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UB": {
-            "accuracy": 0.7449664429530202,
-            "count": 298
+            "accuracy": 0.75,
+            "count": 152
           },
           "UD": {
-            "accuracy": 0.698019801980198,
-            "count": 202
+            "accuracy": 0.7142857142857143,
+            "count": 98
           }
         }
       }
     },
     "summary": {
-      "overall_accuracy": 0.54125,
-      "total_examples": 2400,
-      "n_splits": 22
+      "overall_accuracy": 0.6226666666666667,
+      "digit_accuracy": 0.8917142857142857,
+      "total_examples": 1500,
+      "n_splits": 24
     }
   }
 }