Upload add_sub_baseline_50K_2L1H128d/metrics.json with huggingface_hub

add_sub_baseline_50K_2L1H128d/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": 0.45,
-        "n_examples": 100,
+        "full_accuracy": 0.48,
+        "digit_accuracy": 0.9114285714285715,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.8925619834710744,
-            "count": 605
+            "accuracy": 0.8983050847457628,
+            "count": 295
           },
           "SS": {
-            "accuracy": 0.9894736842105263,
-            "count": 95
+            "accuracy": 0.9818181818181818,
+            "count": 55
           }
         }
       },
       "add_S1": {
-        "full_accuracy": 0.24,
-        "n_examples": 100,
+        "full_accuracy": 0.3,
+        "digit_accuracy": 0.8371428571428572,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9166666666666666,
-            "count": 204
+            "accuracy": 0.9047619047619048,
+            "count": 126
           },
           "SC": {
-            "accuracy": 0.9349112426035503,
-            "count": 169
+            "accuracy": 0.8987341772151899,
+            "count": 79
           },
           "SS": {
-            "accuracy": 0.967741935483871,
-            "count": 31
+            "accuracy": 0.9523809523809523,
+            "count": 21
           },
           "UC": {
-            "accuracy": 0.7297297297297297,
-            "count": 296
+            "accuracy": 0.7096774193548387,
+            "count": 124
           }
         }
       },
       "add_S2": {
-        "full_accuracy": 0.23,
-        "n_examples": 100,
+        "full_accuracy": 0.2,
+        "digit_accuracy": 0.8114285714285714,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9079754601226994,
-            "count": 163
+            "accuracy": 0.8266666666666667,
+            "count": 75
           },
           "SC": {
-            "accuracy": 0.9076923076923077,
-            "count": 130
+            "accuracy": 0.8870967741935484,
+            "count": 62
           },
           "SS": {
-            "accuracy": 0.8735632183908046,
-            "count": 87
+            "accuracy": 0.8461538461538461,
+            "count": 39
           },
           "UC": {
-            "accuracy": 0.6502463054187192,
-            "count": 203
+            "accuracy": 0.6396396396396397,
+            "count": 111
           },
           "US": {
-            "accuracy": 0.9743589743589743,
-            "count": 117
+            "accuracy": 1.0,
+            "count": 63
           }
         }
       },
       "add_S3": {
-        "full_accuracy": 0.16,
-        "n_examples": 100,
+        "full_accuracy": 0.18,
+        "digit_accuracy": 0.8228571428571428,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9669421487603306,
-            "count": 121
+            "accuracy": 0.9666666666666667,
+            "count": 60
           },
           "SC": {
-            "accuracy": 0.8925619834710744,
-            "count": 121
+            "accuracy": 0.8947368421052632,
+            "count": 57
           },
           "SS": {
-            "accuracy": 0.9591836734693877,
-            "count": 49
+            "accuracy": 0.8421052631578947,
+            "count": 19
           },
           "UC": {
-            "accuracy": 0.6290322580645161,
-            "count": 186
+            "accuracy": 0.7115384615384616,
+            "count": 104
           },
           "US": {
-            "accuracy": 0.7399103139013453,
-            "count": 223
+            "accuracy": 0.8090909090909091,
+            "count": 110
           }
         }
       },
       "add_S4": {
-        "full_accuracy": 0.29,
-        "n_examples": 100,
+        "full_accuracy": 0.24,
+        "digit_accuracy": 0.6914285714285714,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.9711538461538461,
-            "count": 104
+            "accuracy": 0.9375,
+            "count": 48
           },
           "SC": {
-            "accuracy": 0.9433962264150944,
-            "count": 106
+            "accuracy": 0.9423076923076923,
+            "count": 52
           },
           "SS": {
-            "accuracy": 0.9130434782608695,
-            "count": 23
+            "accuracy": 1.0,
+            "count": 7
           },
           "UC": {
-            "accuracy": 0.66875,
-            "count": 160
+            "accuracy": 0.6067415730337079,
+            "count": 89
           },
           "US": {
-            "accuracy": 0.5928338762214984,
-            "count": 307
+            "accuracy": 0.564935064935065,
+            "count": 154
           }
         }
       },
       "add_S5": {
-        "full_accuracy": 0.12,
-        "n_examples": 100,
+        "full_accuracy": 0.1,
+        "digit_accuracy": 0.5257142857142857,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.99,
-            "count": 100
+            "accuracy": 1.0,
+            "count": 50
           },
           "SC": {
-            "accuracy": 0.99,
-            "count": 100
+            "accuracy": 1.0,
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.31,
-            "count": 100
+            "accuracy": 0.42,
+            "count": 50
           },
           "US": {
-            "accuracy": 0.3675,
-            "count": 400
+            "accuracy": 0.315,
+            "count": 200
           }
         }
       },
       "add_S6": {
-        "full_accuracy": 0.29,
-        "n_examples": 100,
+        "full_accuracy": 0.34,
+        "digit_accuracy": 0.5742857142857143,
+        "n_examples": 50,
         "per_subtask": {
           "SC": {
-            "accuracy": 0.99,
-            "count": 100
+            "accuracy": 1.0,
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.49,
-            "count": 100
+            "accuracy": 0.46,
+            "count": 50
           },
           "US": {
-            "accuracy": 0.492,
-            "count": 500
+            "accuracy": 0.512,
+            "count": 250
           }
         }
       },
       "add_random": {
-        "full_accuracy": 0.235,
+        "full_accuracy": 0.25,
+        "digit_accuracy": 0.8514285714285714,
         "n_examples": 200,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.8926174496644296,
-            "count": 447
+            "accuracy": 0.91415313225058,
+            "count": 431
           },
           "SC": {
-            "accuracy": 0.946875,
-            "count": 320
+            "accuracy": 0.9493670886075949,
+            "count": 316
           },
           "SS": {
             "accuracy": 1.0,
-            "count": 56
+            "count": 39
           },
           "UC": {
-            "accuracy": 0.7334593572778828,
-            "count": 529
+            "accuracy": 0.7303571428571428,
+            "count": 560
           },
           "US": {
-            "accuracy": 0.8958333333333334,
-            "count": 48
+            "accuracy": 0.9259259259259259,
+            "count": 54
+          }
+        }
+      },
+      "add_C1": {
+        "full_accuracy": 0.26,
+        "digit_accuracy": 0.8657142857142858,
+        "n_examples": 50,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.928,
+            "count": 250
+          },
+          "SC": {
+            "accuracy": 0.96,
+            "count": 50
+          },
+          "UC": {
+            "accuracy": 0.46,
+            "count": 50
+          }
+        }
+      },
+      "add_C2": {
+        "full_accuracy": 0.24,
+        "digit_accuracy": 0.8371428571428572,
+        "n_examples": 50,
+        "per_subtask": {
+          "SA": {
+            "accuracy": 0.945,
+            "count": 200
+          },
+          "SC": {
+            "accuracy": 0.94,
+            "count": 50
+          },
+          "UC": {
+            "accuracy": 0.5180722891566265,
+            "count": 83
+          },
+          "US": {
+            "accuracy": 0.8235294117647058,
+            "count": 17
           }
         }
       },
       "add_C3": {
-        "full_accuracy": 0.22,
-        "n_examples": 100,
+        "full_accuracy": 0.14,
+        "digit_accuracy": 0.8057142857142857,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.93,
-            "count": 300
+            "accuracy": 0.9533333333333334,
+            "count": 150
           },
           "SC": {
-            "accuracy": 0.99,
-            "count": 100
+            "accuracy": 1.0,
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.538860103626943,
-            "count": 193
+            "accuracy": 0.5,
+            "count": 100
           },
           "US": {
-            "accuracy": 0.7850467289719626,
-            "count": 107
+            "accuracy": 0.78,
+            "count": 50
           }
         }
       },
       "add_C4": {
-        "full_accuracy": 0.21,
-        "n_examples": 100,
+        "full_accuracy": 0.18,
+        "digit_accuracy": 0.82,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 0.96,
-            "count": 200
+            "accuracy": 0.98,
+            "count": 100
           },
           "SC": {
-            "accuracy": 0.99,
-            "count": 100
+            "accuracy": 0.98,
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.62109375,
-            "count": 256
+            "accuracy": 0.6060606060606061,
+            "count": 132
           },
           "US": {
-            "accuracy": 0.7847222222222222,
-            "count": 144
+            "accuracy": 0.8823529411764706,
+            "count": 68
           }
         }
       },
       "add_C5": {
-        "full_accuracy": 0.18,
-        "n_examples": 100,
+        "full_accuracy": 0.16,
+        "digit_accuracy": 0.7942857142857143,
+        "n_examples": 50,
         "per_subtask": {
           "SA": {
-            "accuracy": 1.0,
-            "count": 100
+            "accuracy": 0.98,
+            "count": 50
           },
           "SC": {
-            "accuracy": 0.97,
-            "count": 100
+            "accuracy": 1.0,
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.6666666666666666,
-            "count": 306
+            "accuracy": 0.6438356164383562,
+            "count": 146
           },
           "US": {
-            "accuracy": 0.7525773195876289,
-            "count": 194
+            "accuracy": 0.8173076923076923,
+            "count": 104
           }
         }
       },
       "add_C6": {
-        "full_accuracy": 0.07,
-        "n_examples": 100,
+        "full_accuracy": 0.14,
+        "digit_accuracy": 0.7628571428571429,
+        "n_examples": 50,
         "per_subtask": {
           "SC": {
-            "accuracy": 0.99,
-            "count": 100
+            "accuracy": 0.98,
+            "count": 50
           },
           "UC": {
-            "accuracy": 0.6721311475409836,
-            "count": 366
+            "accuracy": 0.6931216931216931,
+            "count": 189
           },
           "US": {
-            "accuracy": 0.8333333333333334,
-            "count": 234
+            "accuracy": 0.7837837837837838,
+            "count": 111
           }
         }
       },
       "sub_M0": {
-        "full_accuracy": 0.45,
-        "n_examples": 100,
+        "full_accuracy": 0.48,
+        "digit_accuracy": 0.8971428571428571,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.8885191347753744,
-            "count": 601
+            "accuracy": 0.8811881188118812,
+            "count": 303
           },
           "ME": {
             "accuracy": 1.0,
-            "count": 99
+            "count": 47
           }
         }
       },
       "sub_M1": {
-        "full_accuracy": 0.21,
-        "n_examples": 100,
+        "full_accuracy": 0.22,
+        "digit_accuracy": 0.8314285714285714,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9175627240143369,
-            "count": 279
+            "accuracy": 0.9361702127659575,
+            "count": 141
           },
           "MB": {
-            "accuracy": 0.9379310344827586,
-            "count": 145
+            "accuracy": 0.9583333333333334,
+            "count": 72
           },
           "ME": {
-            "accuracy": 0.9583333333333334,
-            "count": 24
+            "accuracy": 0.8888888888888888,
+            "count": 18
           },
           "UB": {
-            "accuracy": 0.6031746031746031,
-            "count": 252
+            "accuracy": 0.6218487394957983,
+            "count": 119
           }
         }
       },
       "sub_M2": {
-        "full_accuracy": 0.21,
-        "n_examples": 100,
+        "full_accuracy": 0.14,
+        "digit_accuracy": 0.8171428571428572,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9342723004694836,
-            "count": 213
+            "accuracy": 0.9464285714285714,
+            "count": 112
           },
           "MB": {
-            "accuracy": 0.9557522123893806,
-            "count": 113
+            "accuracy": 0.9245283018867925,
+            "count": 53
           },
           "ME": {
-            "accuracy": 0.9882352941176471,
-            "count": 85
+            "accuracy": 0.9148936170212766,
+            "count": 47
           },
           "UB": {
-            "accuracy": 0.5027624309392266,
-            "count": 181
+            "accuracy": 0.4588235294117647,
+            "count": 85
           },
           "UD": {
-            "accuracy": 0.9907407407407407,
-            "count": 108
+            "accuracy": 0.9245283018867925,
+            "count": 53
           }
         }
       },
       "sub_M3": {
-        "full_accuracy": 0.04,
-        "n_examples": 100,
+        "full_accuracy": 0.02,
+        "digit_accuracy": 0.7142857142857143,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9888268156424581,
-            "count": 179
+            "accuracy": 0.9278350515463918,
+            "count": 97
           },
           "MB": {
-            "accuracy": 0.9223300970873787,
-            "count": 103
+            "accuracy": 0.9607843137254902,
+            "count": 51
           },
           "ME": {
-            "accuracy": 0.9464285714285714,
-            "count": 56
+            "accuracy": 0.9629629629629629,
+            "count": 27
           },
           "UB": {
-            "accuracy": 0.436241610738255,
-            "count": 149
+            "accuracy": 0.4594594594594595,
+            "count": 74
           },
           "UD": {
-            "accuracy": 0.5539906103286385,
-            "count": 213
+            "accuracy": 0.504950495049505,
+            "count": 101
           }
         }
       },
       "sub_M4": {
-        "full_accuracy": 0.06,
-        "n_examples": 100,
+        "full_accuracy": 0.08,
+        "digit_accuracy": 0.6542857142857142,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.97,
-            "count": 200
-          },
-          "MB": {
             "accuracy": 0.96,
             "count": 100
           },
+          "MB": {
+            "accuracy": 0.94,
+            "count": 50
+          },
           "UB": {
-            "accuracy": 0.54,
-            "count": 100
+            "accuracy": 0.56,
+            "count": 50
           },
           "UD": {
-            "accuracy": 0.33,
-            "count": 300
+            "accuracy": 0.38666666666666666,
+            "count": 150
           }
         }
       },
       "sub_M5": {
         "full_accuracy": 0.04,
-        "n_examples": 100,
+        "digit_accuracy": 0.5314285714285715,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "MB": {
-            "accuracy": 0.98,
-            "count": 100
+            "accuracy": 1.0,
+            "count": 50
           },
           "UB": {
-            "accuracy": 0.51,
-            "count": 100
+            "accuracy": 0.56,
+            "count": 50
           },
           "UD": {
-            "accuracy": 0.24,
-            "count": 400
+            "accuracy": 0.29,
+            "count": 200
           }
         }
       },
       "sub_random": {
-        "full_accuracy": 0.27,
+        "full_accuracy": 0.31,
+        "digit_accuracy": 0.8514285714285714,
         "n_examples": 200,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9383333333333334,
-            "count": 600
+            "accuracy": 0.9526315789473684,
+            "count": 570
           },
           "MB": {
-            "accuracy": 0.9363295880149812,
-            "count": 267
+            "accuracy": 0.9494584837545126,
+            "count": 277
           },
           "ME": {
-            "accuracy": 1.0,
+            "accuracy": 0.9811320754716981,
             "count": 53
           },
           "UB": {
-            "accuracy": 0.6104783599088838,
-            "count": 439
+            "accuracy": 0.6475583864118896,
+            "count": 471
           },
           "UD": {
-            "accuracy": 0.9512195121951219,
-            "count": 41
+            "accuracy": 1.0,
+            "count": 29
           }
         }
       },
       "sub_B3": {
-        "full_accuracy": 0.1,
-        "n_examples": 100,
+        "full_accuracy": 0.06,
+        "digit_accuracy": 0.7942857142857143,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.9433333333333334,
-            "count": 300
+            "accuracy": 0.9666666666666667,
+            "count": 150
           },
           "MB": {
-            "accuracy": 0.97,
-            "count": 100
+            "accuracy": 0.98,
+            "count": 50
           },
           "UB": {
-            "accuracy": 0.5126903553299492,
-            "count": 197
+            "accuracy": 0.48514851485148514,
+            "count": 101
           },
           "UD": {
-            "accuracy": 0.7184466019417476,
-            "count": 103
+            "accuracy": 0.7142857142857143,
+            "count": 49
           }
         }
       },
       "sub_B4": {
-        "full_accuracy": 0.03,
-        "n_examples": 100,
+        "full_accuracy": 0.08,
+        "digit_accuracy": 0.7171428571428572,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
-            "accuracy": 0.97,
-            "count": 200
+            "accuracy": 1.0,
+            "count": 100
           },
           "MB": {
-            "accuracy": 0.96,
-            "count": 100
+            "accuracy": 0.98,
+            "count": 50
           },
           "UB": {
-            "accuracy": 0.43724696356275305,
-            "count": 247
+            "accuracy": 0.45454545454545453,
+            "count": 121
           },
           "UD": {
-            "accuracy": 0.6339869281045751,
-            "count": 153
+            "accuracy": 0.5949367088607594,
+            "count": 79
           }
         }
       },
       "sub_B5": {
-        "full_accuracy": 0.03,
-        "n_examples": 100,
+        "full_accuracy": 0.06,
+        "digit_accuracy": 0.7057142857142857,
+        "n_examples": 50,
         "per_subtask": {
           "MD": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "MB": {
             "accuracy": 1.0,
-            "count": 100
+            "count": 50
           },
           "UB": {
-            "accuracy": 0.4395973154362416,
-            "count": 298
+            "accuracy": 0.4934210526315789,
+            "count": 152
           },
           "UD": {
-            "accuracy": 0.6881188118811881,
-            "count": 202
+            "accuracy": 0.7346938775510204,
+            "count": 98
           }
         }
       }
     },
     "summary": {
-      "overall_accuracy": 0.1925,
-      "total_examples": 2400,
-      "n_splits": 22
+      "overall_accuracy": 0.21266666666666667,
+      "digit_accuracy": 0.7844761904761904,
+      "total_examples": 1500,
+      "n_splits": 24
     }
   }
 }