Upload add_sub_baseline_100K_1L3H510d/metrics.json with huggingface_hub

add_sub_baseline_100K_1L3H510d/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": 0.99,
-        "n_examples": 100,
+        "full_accuracy": 0.98,
+        "digit_accuracy": 0.9971428571428571,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9983471074380166,
-            "count": 605
+            "accuracy": 0.9966101694915255,
+            "count": 295
           },
           "SS": {
             "accuracy": 1.0,
-            "count": 95
+            "count": 55
           }
         }
       },
       "add_S1": {
-        "full_accuracy": 0.96,
-        "n_examples": 100,
+        "full_accuracy": 0.94,
+        "digit_accuracy": 0.9914285714285714,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9901960784313726,
-            "count": 204
+            "accuracy": 1.0,
+            "count": 126
           },
           "SC": {
-            "accuracy": 1.0,
-            "count": 169
+            "accuracy": 0.9746835443037974,
+            "count": 79
           },
           "SS": {
             "accuracy": 1.0,
-            "count": 31
+            "count": 21
           },
           "UC": {
-            "accuracy": 0.9932432432432432,
-            "count": 296
+            "accuracy": 0.9919354838709677,
+            "count": 124
           }
         }
       },
       "add_S2": {
-        "full_accuracy": 0.8,
-        "n_examples": 100,
+        "full_accuracy": 0.74,
+        "digit_accuracy": 0.9542857142857143,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 1.0,
-            "count": 163
+            "accuracy": 0.9733333333333334,
+            "count": 75
           },
           "SC": {
-            "accuracy": 0.9846153846153847,
-            "count": 130
+            "accuracy": 0.9838709677419355,
+            "count": 62
           },
           "SS": {
-            "accuracy": 0.9540229885057471,
-            "count": 87
+            "accuracy": 0.9230769230769231,
+            "count": 39
           },
           "UC": {
-            "accuracy": 0.9211822660098522,
-            "count": 203
+            "accuracy": 0.9099099099099099,
+            "count": 111
           },
           "US": {
             "accuracy": 1.0,
-            "count": 117
+            "count": 63
           }
         }
       },
       "add_S3": {
-        "full_accuracy": 0.57,
-        "n_examples": 100,
+        "full_accuracy": 0.6,
+        "digit_accuracy": 0.9285714285714286,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9917355371900827,
-            "count": 121
+            "accuracy": 1.0,
+            "count": 60
           },
           "SC": {
-            "accuracy": 0.9917355371900827,
-            "count": 121
+            "accuracy": 1.0,
+            "count": 57
           },
           "SS": {
             "accuracy": 1.0,
-            "count": 49
+            "count": 19
           },
           "UC": {
-            "accuracy": 0.7795698924731183,
-            "count": 186
+            "accuracy": 0.8076923076923077,
+            "count": 104
           },
           "US": {
-            "accuracy": 0.968609865470852,
-            "count": 223
+            "accuracy": 0.9545454545454546,
+            "count": 110
           }
         }
       },
       "add_S4": {
-        "full_accuracy": 0.48,
-        "n_examples": 100,
+        "full_accuracy": 0.44,
+        "digit_accuracy": 0.8228571428571428,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 104
+            "count": 48
           },
           "SC": {
             "accuracy": 1.0,
-            "count": 106
+            "count": 52
           },
           "SS": {
             "accuracy": 1.0,
-            "count": 23
+            "count": 7
           },
           "UC": {
-            "accuracy": 0.76875,
-            "count": 160
+            "accuracy": 0.7415730337078652,
+            "count": 89
           },
           "US": {
-            "accuracy": 0.7947882736156352,
-            "count": 307
+            "accuracy": 0.7467532467532467,
+            "count": 154
           }
         }
       },
       "add_S5": {
-        "full_accuracy": 0.18,
-        "n_examples": 100,
+        "full_accuracy": 0.1,
+        "digit_accuracy": 0.58,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "SC": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.36,
-            "count": 100
+            "accuracy": 0.32,
+            "count": 50
           },
           "US": {
-            "accuracy": 0.495,
-            "count": 400
+            "accuracy": 0.435,
+            "count": 200
           }
         }
       },
       "add_S6": {
-        "full_accuracy": 0.49,
-        "n_examples": 100,
+        "full_accuracy": 0.52,
+        "digit_accuracy": 0.7428571428571429,
+        "n_examples": 50,
         "per_subtask": {
           "SC": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.65,
-            "count": 100
+            "accuracy": 0.66,
+            "count": 50
           },
           "US": {
-            "accuracy": 0.682,
-            "count": 500
+            "accuracy": 0.708,
+            "count": 250
           }
         }
       },
       "add_random": {
-        "full_accuracy": 0.925,
+        "full_accuracy": 0.94,
+        "digit_accuracy": 0.99,
         "n_examples": 200,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9932885906040269,
-            "count": 447
+            "accuracy": 0.9930394431554525,
+            "count": 431
           },
           "SC": {
-            "accuracy": 0.9875,
-            "count": 320
+            "accuracy": 0.9968354430379747,
+            "count": 316
           },
           "SS": {
-            "accuracy": 0.9821428571428571,
-            "count": 56
+            "accuracy": 0.9743589743589743,
+            "count": 39
           },
           "UC": {
-            "accuracy": 0.9848771266540642,
-            "count": 529
+            "accuracy": 0.9857142857142858,
+            "count": 560
           },
           "US": {
+            "accuracy": 0.9814814814814815,
+            "count": 54
+          }
+        }
+      },
+      "add_C1": {
+        "full_accuracy": 0.94,
+        "digit_accuracy": 0.9914285714285714,
+        "n_examples": 50,
+        "per_subtask": {
+          "SA": {
             "accuracy": 1.0,
-            "count": 48
+            "count": 250
+          },
+          "SC": {
+            "accuracy": 1.0,
+            "count": 50
+          },
+          "UC": {
+            "accuracy": 0.94,
+            "count": 50
+          }
+        }
+      },
+      "add_C2": {
+        "full_accuracy": 0.82,
+        "digit_accuracy": 0.9657142857142857,
+        "n_examples": 50,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 1.0,
+            "count": 200
+          },
+          "SC": {
+            "accuracy": 1.0,
+            "count": 50
+          },
+          "UC": {
+            "accuracy": 0.891566265060241,
+            "count": 83
+          },
+          "US": {
+            "accuracy": 0.8235294117647058,
+            "count": 17
           }
         }
       },
       "add_C3": {
-        "full_accuracy": 0.67,
-        "n_examples": 100,
+        "full_accuracy": 0.68,
+        "digit_accuracy": 0.94,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 300
+            "count": 150
           },
           "SC": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.8393782383419689,
-            "count": 193
+            "accuracy": 0.86,
+            "count": 100
           },
           "US": {
-            "accuracy": 0.9065420560747663,
-            "count": 107
+            "accuracy": 0.86,
+            "count": 50
           }
         }
       },
       "add_C4": {
-        "full_accuracy": 0.67,
-        "n_examples": 100,
+        "full_accuracy": 0.76,
+        "digit_accuracy": 0.9571428571428572,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 200
+            "count": 100
           },
           "SC": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.875,
-            "count": 256
+            "accuracy": 0.9166666666666666,
+            "count": 132
           },
           "US": {
-            "accuracy": 0.8958333333333334,
-            "count": 144
+            "accuracy": 0.9411764705882353,
+            "count": 68
           }
         }
       },
       "add_C5": {
-        "full_accuracy": 0.57,
-        "n_examples": 100,
+        "full_accuracy": 0.62,
+        "digit_accuracy": 0.9,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "SC": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.8758169934640523,
-            "count": 306
+            "accuracy": 0.8561643835616438,
+            "count": 146
           },
           "US": {
-            "accuracy": 0.8350515463917526,
-            "count": 194
+            "accuracy": 0.8653846153846154,
+            "count": 104
           }
         }
       },
       "add_C6": {
-        "full_accuracy": 0.65,
-        "n_examples": 100,
+        "full_accuracy": 0.68,
+        "digit_accuracy": 0.94,
+        "n_examples": 50,
         "per_subtask": {
           "SC": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.912568306010929,
-            "count": 366
+            "accuracy": 0.9206349206349206,
+            "count": 189
           },
           "US": {
-            "accuracy": 0.9316239316239316,
-            "count": 234
+            "accuracy": 0.9459459459459459,
+            "count": 111
           }
         }
       },
       "sub_M0": {
-        "full_accuracy": 0.91,
-        "n_examples": 100,
+        "full_accuracy": 0.92,
+        "digit_accuracy": 0.9885714285714285,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9866888519134775,
-            "count": 601
+            "accuracy": 0.9867986798679867,
+            "count": 303
           },
           "ME": {
-            "accuracy": 0.9797979797979798,
-            "count": 99
+            "accuracy": 1.0,
+            "count": 47
           }
         }
       },
       "sub_M1": {
         "full_accuracy": 0.98,
-        "n_examples": 100,
+        "digit_accuracy": 0.9971428571428571,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.996415770609319,
-            "count": 279
+            "accuracy": 1.0,
+            "count": 141
           },
           "MB": {
-            "accuracy": 0.993103448275862,
-            "count": 145
+            "accuracy": 0.9861111111111112,
+            "count": 72
           },
           "ME": {
             "accuracy": 1.0,
-            "count": 24
+            "count": 18
           },
           "UB": {
             "accuracy": 1.0,
-            "count": 252
+            "count": 119
           }
         }
       },
       "sub_M2": {
-        "full_accuracy": 0.64,
-        "n_examples": 100,
+        "full_accuracy": 0.6,
+        "digit_accuracy": 0.9371428571428572,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9859154929577465,
-            "count": 213
+            "accuracy": 0.9910714285714286,
+            "count": 112
           },
           "MB": {
-            "accuracy": 1.0,
-            "count": 113
+            "accuracy": 0.9622641509433962,
+            "count": 53
           },
           "ME": {
             "accuracy": 1.0,
-            "count": 85
+            "count": 47
           },
           "UB": {
-            "accuracy": 0.8011049723756906,
-            "count": 181
+            "accuracy": 0.788235294117647,
+            "count": 85
           },
           "UD": {
-            "accuracy": 1.0,
-            "count": 108
+            "accuracy": 0.9811320754716981,
+            "count": 53
           }
         }
       },
       "sub_M3": {
-        "full_accuracy": 0.16,
-        "n_examples": 100,
+        "full_accuracy": 0.24,
+        "digit_accuracy": 0.86,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 1.0,
-            "count": 179
+            "accuracy": 0.9896907216494846,
+            "count": 97
           },
           "MB": {
-            "accuracy": 1.0,
-            "count": 103
+            "accuracy": 0.9803921568627451,
+            "count": 51
           },
           "ME": {
             "accuracy": 1.0,
-            "count": 56
+            "count": 27
           },
           "UB": {
-            "accuracy": 0.4966442953020134,
-            "count": 149
+            "accuracy": 0.581081081081081,
+            "count": 74
           },
           "UD": {
-            "accuracy": 0.8356807511737089,
-            "count": 213
+            "accuracy": 0.8415841584158416,
+            "count": 101
           }
         }
       },
       "sub_M4": {
-        "full_accuracy": 0.05,
-        "n_examples": 100,
+        "full_accuracy": 0.04,
+        "digit_accuracy": 0.6742857142857143,
+        "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": 0.34,
+            "count": 50
           },
           "UD": {
             "accuracy": 0.46,
-            "count": 300
+            "count": 150
           }
         }
       },
       "sub_M5": {
-        "full_accuracy": 0.06,
-        "n_examples": 100,
+        "full_accuracy": 0.12,
+        "digit_accuracy": 0.6314285714285715,
+        "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": 0.52,
+            "count": 50
           },
           "UD": {
-            "accuracy": 0.405,
-            "count": 400
+            "accuracy": 0.475,
+            "count": 200
           }
         }
       },
       "sub_random": {
-        "full_accuracy": 0.895,
+        "full_accuracy": 0.94,
+        "digit_accuracy": 0.9914285714285714,
         "n_examples": 200,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9916666666666667,
-            "count": 600
+            "accuracy": 0.9982456140350877,
+            "count": 570
           },
           "MB": {
-            "accuracy": 1.0,
-            "count": 267
+            "accuracy": 0.9927797833935018,
+            "count": 277
           },
           "ME": {
             "accuracy": 1.0,
             "count": 53
           },
           "UB": {
-            "accuracy": 0.9635535307517085,
-            "count": 439
+            "accuracy": 0.9808917197452229,
+            "count": 471
           },
           "UD": {
             "accuracy": 1.0,
-            "count": 41
+            "count": 29
           }
         }
       },
       "sub_B3": {
-        "full_accuracy": 0.51,
-        "n_examples": 100,
+        "full_accuracy": 0.56,
+        "digit_accuracy": 0.9228571428571428,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 300
+            "count": 150
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UB": {
-            "accuracy": 0.7868020304568528,
-            "count": 197
+            "accuracy": 0.8118811881188119,
+            "count": 101
           },
           "UD": {
-            "accuracy": 0.8252427184466019,
-            "count": 103
+            "accuracy": 0.8367346938775511,
+            "count": 49
           }
         }
       },
       "sub_B4": {
-        "full_accuracy": 0.33,
-        "n_examples": 100,
+        "full_accuracy": 0.36,
+        "digit_accuracy": 0.8628571428571429,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 200
+            "count": 100
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UB": {
-            "accuracy": 0.7732793522267206,
-            "count": 247
+            "accuracy": 0.7768595041322314,
+            "count": 121
           },
           "UD": {
-            "accuracy": 0.7777777777777778,
-            "count": 153
+            "accuracy": 0.7341772151898734,
+            "count": 79
           }
         }
       },
       "sub_B5": {
-        "full_accuracy": 0.29,
-        "n_examples": 100,
+        "full_accuracy": 0.18,
+        "digit_accuracy": 0.8314285714285714,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UB": {
-            "accuracy": 0.7751677852348994,
-            "count": 298
+            "accuracy": 0.7631578947368421,
+            "count": 152
           },
           "UD": {
-            "accuracy": 0.7425742574257426,
-            "count": 202
+            "accuracy": 0.7653061224489796,
+            "count": 98
           }
         }
       }
     },
     "summary": {
-      "overall_accuracy": 0.6070833333333333,
-      "total_examples": 2400,
-      "n_splits": 22
+      "overall_accuracy": 0.678,
+      "digit_accuracy": 0.9113333333333333,
+      "total_examples": 1500,
+      "n_splits": 24
     }
   }
 }