Upload add_sub_baseline_25K/metrics.json with huggingface_hub
Browse files- add_sub_baseline_25K/metrics.json +211 -146
add_sub_baseline_25K/metrics.json
CHANGED
|
@@ -706,416 +706,477 @@
|
|
| 706 |
"K": null,
|
| 707 |
"mode": "sft",
|
| 708 |
"n_digits": 6,
|
| 709 |
-
"n_per_split":
|
| 710 |
},
|
| 711 |
"splits": {
|
| 712 |
"add_S0": {
|
| 713 |
"full_accuracy": 1.0,
|
| 714 |
-
"
|
|
|
|
| 715 |
"per_subtask": {
|
| 716 |
"SA": {
|
| 717 |
"accuracy": 1.0,
|
| 718 |
-
"count":
|
| 719 |
},
|
| 720 |
"SS": {
|
| 721 |
"accuracy": 1.0,
|
| 722 |
-
"count":
|
| 723 |
}
|
| 724 |
}
|
| 725 |
},
|
| 726 |
"add_S1": {
|
| 727 |
-
"full_accuracy":
|
| 728 |
-
"
|
|
|
|
| 729 |
"per_subtask": {
|
| 730 |
"SA": {
|
| 731 |
-
"accuracy":
|
| 732 |
-
"count":
|
| 733 |
},
|
| 734 |
"SC": {
|
| 735 |
"accuracy": 1.0,
|
| 736 |
-
"count":
|
| 737 |
},
|
| 738 |
"SS": {
|
| 739 |
"accuracy": 1.0,
|
| 740 |
-
"count":
|
| 741 |
},
|
| 742 |
"UC": {
|
| 743 |
"accuracy": 1.0,
|
| 744 |
-
"count":
|
| 745 |
}
|
| 746 |
}
|
| 747 |
},
|
| 748 |
"add_S2": {
|
| 749 |
"full_accuracy": 1.0,
|
| 750 |
-
"
|
|
|
|
| 751 |
"per_subtask": {
|
| 752 |
"SA": {
|
| 753 |
"accuracy": 1.0,
|
| 754 |
-
"count":
|
| 755 |
},
|
| 756 |
"SC": {
|
| 757 |
"accuracy": 1.0,
|
| 758 |
-
"count":
|
| 759 |
},
|
| 760 |
"SS": {
|
| 761 |
"accuracy": 1.0,
|
| 762 |
-
"count":
|
| 763 |
},
|
| 764 |
"UC": {
|
| 765 |
"accuracy": 1.0,
|
| 766 |
-
"count":
|
| 767 |
},
|
| 768 |
"US": {
|
| 769 |
"accuracy": 1.0,
|
| 770 |
-
"count":
|
| 771 |
}
|
| 772 |
}
|
| 773 |
},
|
| 774 |
"add_S3": {
|
| 775 |
-
"full_accuracy":
|
| 776 |
-
"
|
|
|
|
| 777 |
"per_subtask": {
|
| 778 |
"SA": {
|
| 779 |
"accuracy": 1.0,
|
| 780 |
-
"count":
|
| 781 |
},
|
| 782 |
"SC": {
|
| 783 |
"accuracy": 1.0,
|
| 784 |
-
"count":
|
| 785 |
},
|
| 786 |
"SS": {
|
| 787 |
"accuracy": 1.0,
|
| 788 |
-
"count":
|
| 789 |
},
|
| 790 |
"UC": {
|
| 791 |
-
"accuracy":
|
| 792 |
-
"count":
|
| 793 |
},
|
| 794 |
"US": {
|
| 795 |
"accuracy": 1.0,
|
| 796 |
-
"count":
|
| 797 |
}
|
| 798 |
}
|
| 799 |
},
|
| 800 |
"add_S4": {
|
| 801 |
-
"full_accuracy": 0.
|
| 802 |
-
"
|
|
|
|
| 803 |
"per_subtask": {
|
| 804 |
"SA": {
|
| 805 |
"accuracy": 1.0,
|
| 806 |
-
"count":
|
| 807 |
},
|
| 808 |
"SC": {
|
| 809 |
"accuracy": 1.0,
|
| 810 |
-
"count":
|
| 811 |
},
|
| 812 |
"SS": {
|
| 813 |
"accuracy": 1.0,
|
| 814 |
-
"count":
|
| 815 |
},
|
| 816 |
"UC": {
|
| 817 |
-
"accuracy": 0.
|
| 818 |
-
"count":
|
| 819 |
},
|
| 820 |
"US": {
|
| 821 |
-
"accuracy":
|
| 822 |
-
"count":
|
| 823 |
}
|
| 824 |
}
|
| 825 |
},
|
| 826 |
"add_S5": {
|
| 827 |
-
"full_accuracy": 0.
|
| 828 |
-
"
|
|
|
|
| 829 |
"per_subtask": {
|
| 830 |
"SA": {
|
| 831 |
"accuracy": 1.0,
|
| 832 |
-
"count":
|
| 833 |
},
|
| 834 |
"SC": {
|
| 835 |
"accuracy": 1.0,
|
| 836 |
-
"count":
|
| 837 |
},
|
| 838 |
"UC": {
|
| 839 |
-
"accuracy": 0.
|
| 840 |
-
"count":
|
| 841 |
},
|
| 842 |
"US": {
|
| 843 |
-
"accuracy": 0.
|
| 844 |
-
"count":
|
| 845 |
}
|
| 846 |
}
|
| 847 |
},
|
| 848 |
"add_S6": {
|
| 849 |
-
"full_accuracy": 0.
|
| 850 |
-
"
|
|
|
|
| 851 |
"per_subtask": {
|
| 852 |
"SC": {
|
| 853 |
"accuracy": 1.0,
|
| 854 |
-
"count":
|
| 855 |
},
|
| 856 |
"UC": {
|
| 857 |
-
"accuracy": 0.
|
| 858 |
-
"count":
|
| 859 |
},
|
| 860 |
"US": {
|
| 861 |
-
"accuracy": 0.
|
| 862 |
-
"count":
|
| 863 |
}
|
| 864 |
}
|
| 865 |
},
|
| 866 |
"add_random": {
|
| 867 |
"full_accuracy": 1.0,
|
|
|
|
| 868 |
"n_examples": 200,
|
| 869 |
"per_subtask": {
|
| 870 |
"SA": {
|
| 871 |
"accuracy": 1.0,
|
| 872 |
-
"count":
|
| 873 |
},
|
| 874 |
"SC": {
|
| 875 |
"accuracy": 1.0,
|
| 876 |
-
"count":
|
| 877 |
},
|
| 878 |
"SS": {
|
| 879 |
"accuracy": 1.0,
|
| 880 |
-
"count":
|
| 881 |
},
|
| 882 |
"UC": {
|
| 883 |
"accuracy": 1.0,
|
| 884 |
-
"count":
|
| 885 |
},
|
| 886 |
"US": {
|
| 887 |
"accuracy": 1.0,
|
| 888 |
-
"count":
|
| 889 |
}
|
| 890 |
}
|
| 891 |
},
|
| 892 |
-
"
|
| 893 |
"full_accuracy": 1.0,
|
| 894 |
-
"
|
|
|
|
| 895 |
"per_subtask": {
|
| 896 |
"SA": {
|
| 897 |
"accuracy": 1.0,
|
| 898 |
-
"count":
|
| 899 |
},
|
| 900 |
"SC": {
|
| 901 |
"accuracy": 1.0,
|
| 902 |
-
"count":
|
| 903 |
},
|
| 904 |
"UC": {
|
| 905 |
"accuracy": 1.0,
|
| 906 |
-
"count":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 907 |
},
|
| 908 |
"US": {
|
| 909 |
"accuracy": 1.0,
|
| 910 |
-
"count":
|
| 911 |
}
|
| 912 |
}
|
| 913 |
},
|
| 914 |
-
"
|
| 915 |
-
"full_accuracy": 0.
|
| 916 |
-
"
|
|
|
|
| 917 |
"per_subtask": {
|
| 918 |
"SA": {
|
| 919 |
"accuracy": 1.0,
|
| 920 |
-
"count":
|
| 921 |
},
|
| 922 |
"SC": {
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 923 |
"accuracy": 1.0,
|
| 924 |
"count": 100
|
| 925 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
| 926 |
"UC": {
|
| 927 |
-
"accuracy": 0.
|
| 928 |
-
"count":
|
| 929 |
},
|
| 930 |
"US": {
|
| 931 |
"accuracy": 1.0,
|
| 932 |
-
"count":
|
| 933 |
}
|
| 934 |
}
|
| 935 |
},
|
| 936 |
"add_C5": {
|
| 937 |
-
"full_accuracy": 0.
|
| 938 |
-
"
|
|
|
|
| 939 |
"per_subtask": {
|
| 940 |
"SA": {
|
| 941 |
"accuracy": 1.0,
|
| 942 |
-
"count":
|
| 943 |
},
|
| 944 |
"SC": {
|
| 945 |
"accuracy": 1.0,
|
| 946 |
-
"count":
|
| 947 |
},
|
| 948 |
"UC": {
|
| 949 |
-
"accuracy": 0.
|
| 950 |
-
"count":
|
| 951 |
},
|
| 952 |
"US": {
|
| 953 |
-
"accuracy": 0.
|
| 954 |
-
"count":
|
| 955 |
}
|
| 956 |
}
|
| 957 |
},
|
| 958 |
"add_C6": {
|
| 959 |
"full_accuracy": 1.0,
|
| 960 |
-
"
|
|
|
|
| 961 |
"per_subtask": {
|
| 962 |
"SC": {
|
| 963 |
"accuracy": 1.0,
|
| 964 |
-
"count":
|
| 965 |
},
|
| 966 |
"UC": {
|
| 967 |
"accuracy": 1.0,
|
| 968 |
-
"count":
|
| 969 |
},
|
| 970 |
"US": {
|
| 971 |
"accuracy": 1.0,
|
| 972 |
-
"count":
|
| 973 |
}
|
| 974 |
}
|
| 975 |
},
|
| 976 |
"sub_M0": {
|
| 977 |
"full_accuracy": 1.0,
|
| 978 |
-
"
|
|
|
|
| 979 |
"per_subtask": {
|
| 980 |
"MD": {
|
| 981 |
"accuracy": 1.0,
|
| 982 |
-
"count":
|
| 983 |
},
|
| 984 |
"ME": {
|
| 985 |
"accuracy": 1.0,
|
| 986 |
-
"count":
|
| 987 |
}
|
| 988 |
}
|
| 989 |
},
|
| 990 |
"sub_M1": {
|
| 991 |
"full_accuracy": 1.0,
|
| 992 |
-
"
|
|
|
|
| 993 |
"per_subtask": {
|
| 994 |
"MD": {
|
| 995 |
"accuracy": 1.0,
|
| 996 |
-
"count":
|
| 997 |
},
|
| 998 |
"MB": {
|
| 999 |
"accuracy": 1.0,
|
| 1000 |
-
"count":
|
| 1001 |
},
|
| 1002 |
"ME": {
|
| 1003 |
"accuracy": 1.0,
|
| 1004 |
-
"count":
|
| 1005 |
},
|
| 1006 |
"UB": {
|
| 1007 |
"accuracy": 1.0,
|
| 1008 |
-
"count":
|
| 1009 |
}
|
| 1010 |
}
|
| 1011 |
},
|
| 1012 |
"sub_M2": {
|
| 1013 |
"full_accuracy": 1.0,
|
| 1014 |
-
"
|
|
|
|
| 1015 |
"per_subtask": {
|
| 1016 |
"MD": {
|
| 1017 |
"accuracy": 1.0,
|
| 1018 |
-
"count":
|
| 1019 |
},
|
| 1020 |
"MB": {
|
| 1021 |
"accuracy": 1.0,
|
| 1022 |
-
"count":
|
| 1023 |
},
|
| 1024 |
"ME": {
|
| 1025 |
"accuracy": 1.0,
|
| 1026 |
-
"count":
|
| 1027 |
},
|
| 1028 |
"UB": {
|
| 1029 |
"accuracy": 1.0,
|
| 1030 |
-
"count":
|
| 1031 |
},
|
| 1032 |
"UD": {
|
| 1033 |
"accuracy": 1.0,
|
| 1034 |
-
"count":
|
| 1035 |
}
|
| 1036 |
}
|
| 1037 |
},
|
| 1038 |
"sub_M3": {
|
| 1039 |
-
"full_accuracy":
|
| 1040 |
-
"
|
|
|
|
| 1041 |
"per_subtask": {
|
| 1042 |
"MD": {
|
| 1043 |
"accuracy": 1.0,
|
| 1044 |
-
"count":
|
| 1045 |
},
|
| 1046 |
"MB": {
|
| 1047 |
"accuracy": 1.0,
|
| 1048 |
-
"count":
|
| 1049 |
},
|
| 1050 |
"ME": {
|
| 1051 |
"accuracy": 1.0,
|
| 1052 |
-
"count":
|
| 1053 |
},
|
| 1054 |
"UB": {
|
| 1055 |
-
"accuracy":
|
| 1056 |
-
"count":
|
| 1057 |
},
|
| 1058 |
"UD": {
|
| 1059 |
"accuracy": 1.0,
|
| 1060 |
-
"count":
|
| 1061 |
}
|
| 1062 |
}
|
| 1063 |
},
|
| 1064 |
"sub_M4": {
|
| 1065 |
-
"full_accuracy": 0.
|
| 1066 |
-
"
|
|
|
|
| 1067 |
"per_subtask": {
|
| 1068 |
"MD": {
|
| 1069 |
"accuracy": 1.0,
|
| 1070 |
-
"count":
|
| 1071 |
},
|
| 1072 |
"MB": {
|
| 1073 |
"accuracy": 1.0,
|
| 1074 |
-
"count":
|
| 1075 |
},
|
| 1076 |
"UB": {
|
| 1077 |
-
"accuracy": 0.
|
| 1078 |
-
"count":
|
| 1079 |
},
|
| 1080 |
"UD": {
|
| 1081 |
-
"accuracy":
|
| 1082 |
-
"count":
|
| 1083 |
}
|
| 1084 |
}
|
| 1085 |
},
|
| 1086 |
"sub_M5": {
|
| 1087 |
-
"full_accuracy": 0.
|
| 1088 |
-
"
|
|
|
|
| 1089 |
"per_subtask": {
|
| 1090 |
"MD": {
|
| 1091 |
"accuracy": 1.0,
|
| 1092 |
-
"count":
|
| 1093 |
},
|
| 1094 |
"MB": {
|
| 1095 |
"accuracy": 1.0,
|
| 1096 |
-
"count":
|
| 1097 |
},
|
| 1098 |
"UB": {
|
| 1099 |
-
"accuracy": 0.
|
| 1100 |
-
"count":
|
| 1101 |
},
|
| 1102 |
"UD": {
|
| 1103 |
"accuracy": 0.87,
|
| 1104 |
-
"count":
|
| 1105 |
}
|
| 1106 |
}
|
| 1107 |
},
|
| 1108 |
"sub_random": {
|
| 1109 |
"full_accuracy": 1.0,
|
|
|
|
| 1110 |
"n_examples": 200,
|
| 1111 |
"per_subtask": {
|
| 1112 |
"MD": {
|
| 1113 |
"accuracy": 1.0,
|
| 1114 |
-
"count":
|
| 1115 |
},
|
| 1116 |
"MB": {
|
| 1117 |
"accuracy": 1.0,
|
| 1118 |
-
"count":
|
| 1119 |
},
|
| 1120 |
"ME": {
|
| 1121 |
"accuracy": 1.0,
|
|
@@ -1123,85 +1184,89 @@
|
|
| 1123 |
},
|
| 1124 |
"UB": {
|
| 1125 |
"accuracy": 1.0,
|
| 1126 |
-
"count":
|
| 1127 |
},
|
| 1128 |
"UD": {
|
| 1129 |
"accuracy": 1.0,
|
| 1130 |
-
"count":
|
| 1131 |
}
|
| 1132 |
}
|
| 1133 |
},
|
| 1134 |
"sub_B3": {
|
| 1135 |
-
"full_accuracy":
|
| 1136 |
-
"
|
|
|
|
| 1137 |
"per_subtask": {
|
| 1138 |
"MD": {
|
| 1139 |
-
"accuracy":
|
| 1140 |
-
"count":
|
| 1141 |
},
|
| 1142 |
"MB": {
|
| 1143 |
"accuracy": 1.0,
|
| 1144 |
-
"count":
|
| 1145 |
},
|
| 1146 |
"UB": {
|
| 1147 |
"accuracy": 1.0,
|
| 1148 |
-
"count":
|
| 1149 |
},
|
| 1150 |
"UD": {
|
| 1151 |
"accuracy": 1.0,
|
| 1152 |
-
"count":
|
| 1153 |
}
|
| 1154 |
}
|
| 1155 |
},
|
| 1156 |
"sub_B4": {
|
| 1157 |
-
"full_accuracy": 0.
|
| 1158 |
-
"
|
|
|
|
| 1159 |
"per_subtask": {
|
| 1160 |
"MD": {
|
| 1161 |
-
"accuracy":
|
| 1162 |
-
"count":
|
| 1163 |
},
|
| 1164 |
"MB": {
|
| 1165 |
"accuracy": 1.0,
|
| 1166 |
-
"count":
|
| 1167 |
},
|
| 1168 |
"UB": {
|
| 1169 |
-
"accuracy": 0.
|
| 1170 |
-
"count":
|
| 1171 |
},
|
| 1172 |
"UD": {
|
| 1173 |
"accuracy": 1.0,
|
| 1174 |
-
"count":
|
| 1175 |
}
|
| 1176 |
}
|
| 1177 |
},
|
| 1178 |
"sub_B5": {
|
| 1179 |
-
"full_accuracy": 0.
|
| 1180 |
-
"
|
|
|
|
| 1181 |
"per_subtask": {
|
| 1182 |
"MD": {
|
| 1183 |
"accuracy": 1.0,
|
| 1184 |
-
"count":
|
| 1185 |
},
|
| 1186 |
"MB": {
|
| 1187 |
"accuracy": 1.0,
|
| 1188 |
-
"count":
|
| 1189 |
},
|
| 1190 |
"UB": {
|
| 1191 |
-
"accuracy": 0.
|
| 1192 |
-
"count":
|
| 1193 |
},
|
| 1194 |
"UD": {
|
| 1195 |
-
"accuracy":
|
| 1196 |
-
"count":
|
| 1197 |
}
|
| 1198 |
}
|
| 1199 |
}
|
| 1200 |
},
|
| 1201 |
"summary": {
|
| 1202 |
-
"overall_accuracy": 0.
|
| 1203 |
-
"
|
| 1204 |
-
"
|
|
|
|
| 1205 |
}
|
| 1206 |
}
|
| 1207 |
}
|
|
|
|
| 706 |
"K": null,
|
| 707 |
"mode": "sft",
|
| 708 |
"n_digits": 6,
|
| 709 |
+
"n_per_split": 50
|
| 710 |
},
|
| 711 |
"splits": {
|
| 712 |
"add_S0": {
|
| 713 |
"full_accuracy": 1.0,
|
| 714 |
+
"digit_accuracy": 1.0,
|
| 715 |
+
"n_examples": 50,
|
| 716 |
"per_subtask": {
|
| 717 |
"SA": {
|
| 718 |
"accuracy": 1.0,
|
| 719 |
+
"count": 295
|
| 720 |
},
|
| 721 |
"SS": {
|
| 722 |
"accuracy": 1.0,
|
| 723 |
+
"count": 55
|
| 724 |
}
|
| 725 |
}
|
| 726 |
},
|
| 727 |
"add_S1": {
|
| 728 |
+
"full_accuracy": 1.0,
|
| 729 |
+
"digit_accuracy": 1.0,
|
| 730 |
+
"n_examples": 50,
|
| 731 |
"per_subtask": {
|
| 732 |
"SA": {
|
| 733 |
+
"accuracy": 1.0,
|
| 734 |
+
"count": 126
|
| 735 |
},
|
| 736 |
"SC": {
|
| 737 |
"accuracy": 1.0,
|
| 738 |
+
"count": 79
|
| 739 |
},
|
| 740 |
"SS": {
|
| 741 |
"accuracy": 1.0,
|
| 742 |
+
"count": 21
|
| 743 |
},
|
| 744 |
"UC": {
|
| 745 |
"accuracy": 1.0,
|
| 746 |
+
"count": 124
|
| 747 |
}
|
| 748 |
}
|
| 749 |
},
|
| 750 |
"add_S2": {
|
| 751 |
"full_accuracy": 1.0,
|
| 752 |
+
"digit_accuracy": 1.0,
|
| 753 |
+
"n_examples": 50,
|
| 754 |
"per_subtask": {
|
| 755 |
"SA": {
|
| 756 |
"accuracy": 1.0,
|
| 757 |
+
"count": 75
|
| 758 |
},
|
| 759 |
"SC": {
|
| 760 |
"accuracy": 1.0,
|
| 761 |
+
"count": 62
|
| 762 |
},
|
| 763 |
"SS": {
|
| 764 |
"accuracy": 1.0,
|
| 765 |
+
"count": 39
|
| 766 |
},
|
| 767 |
"UC": {
|
| 768 |
"accuracy": 1.0,
|
| 769 |
+
"count": 111
|
| 770 |
},
|
| 771 |
"US": {
|
| 772 |
"accuracy": 1.0,
|
| 773 |
+
"count": 63
|
| 774 |
}
|
| 775 |
}
|
| 776 |
},
|
| 777 |
"add_S3": {
|
| 778 |
+
"full_accuracy": 1.0,
|
| 779 |
+
"digit_accuracy": 1.0,
|
| 780 |
+
"n_examples": 50,
|
| 781 |
"per_subtask": {
|
| 782 |
"SA": {
|
| 783 |
"accuracy": 1.0,
|
| 784 |
+
"count": 60
|
| 785 |
},
|
| 786 |
"SC": {
|
| 787 |
"accuracy": 1.0,
|
| 788 |
+
"count": 57
|
| 789 |
},
|
| 790 |
"SS": {
|
| 791 |
"accuracy": 1.0,
|
| 792 |
+
"count": 19
|
| 793 |
},
|
| 794 |
"UC": {
|
| 795 |
+
"accuracy": 1.0,
|
| 796 |
+
"count": 104
|
| 797 |
},
|
| 798 |
"US": {
|
| 799 |
"accuracy": 1.0,
|
| 800 |
+
"count": 110
|
| 801 |
}
|
| 802 |
}
|
| 803 |
},
|
| 804 |
"add_S4": {
|
| 805 |
+
"full_accuracy": 0.9,
|
| 806 |
+
"digit_accuracy": 0.9857142857142858,
|
| 807 |
+
"n_examples": 50,
|
| 808 |
"per_subtask": {
|
| 809 |
"SA": {
|
| 810 |
"accuracy": 1.0,
|
| 811 |
+
"count": 48
|
| 812 |
},
|
| 813 |
"SC": {
|
| 814 |
"accuracy": 1.0,
|
| 815 |
+
"count": 52
|
| 816 |
},
|
| 817 |
"SS": {
|
| 818 |
"accuracy": 1.0,
|
| 819 |
+
"count": 7
|
| 820 |
},
|
| 821 |
"UC": {
|
| 822 |
+
"accuracy": 0.9550561797752809,
|
| 823 |
+
"count": 89
|
| 824 |
},
|
| 825 |
"US": {
|
| 826 |
+
"accuracy": 0.9935064935064936,
|
| 827 |
+
"count": 154
|
| 828 |
}
|
| 829 |
}
|
| 830 |
},
|
| 831 |
"add_S5": {
|
| 832 |
+
"full_accuracy": 0.7,
|
| 833 |
+
"digit_accuracy": 0.9314285714285714,
|
| 834 |
+
"n_examples": 50,
|
| 835 |
"per_subtask": {
|
| 836 |
"SA": {
|
| 837 |
"accuracy": 1.0,
|
| 838 |
+
"count": 50
|
| 839 |
},
|
| 840 |
"SC": {
|
| 841 |
"accuracy": 1.0,
|
| 842 |
+
"count": 50
|
| 843 |
},
|
| 844 |
"UC": {
|
| 845 |
+
"accuracy": 0.8,
|
| 846 |
+
"count": 50
|
| 847 |
},
|
| 848 |
"US": {
|
| 849 |
+
"accuracy": 0.93,
|
| 850 |
+
"count": 200
|
| 851 |
}
|
| 852 |
}
|
| 853 |
},
|
| 854 |
"add_S6": {
|
| 855 |
+
"full_accuracy": 0.96,
|
| 856 |
+
"digit_accuracy": 0.9857142857142858,
|
| 857 |
+
"n_examples": 50,
|
| 858 |
"per_subtask": {
|
| 859 |
"SC": {
|
| 860 |
"accuracy": 1.0,
|
| 861 |
+
"count": 50
|
| 862 |
},
|
| 863 |
"UC": {
|
| 864 |
+
"accuracy": 0.96,
|
| 865 |
+
"count": 50
|
| 866 |
},
|
| 867 |
"US": {
|
| 868 |
+
"accuracy": 0.988,
|
| 869 |
+
"count": 250
|
| 870 |
}
|
| 871 |
}
|
| 872 |
},
|
| 873 |
"add_random": {
|
| 874 |
"full_accuracy": 1.0,
|
| 875 |
+
"digit_accuracy": 1.0,
|
| 876 |
"n_examples": 200,
|
| 877 |
"per_subtask": {
|
| 878 |
"SA": {
|
| 879 |
"accuracy": 1.0,
|
| 880 |
+
"count": 431
|
| 881 |
},
|
| 882 |
"SC": {
|
| 883 |
"accuracy": 1.0,
|
| 884 |
+
"count": 316
|
| 885 |
},
|
| 886 |
"SS": {
|
| 887 |
"accuracy": 1.0,
|
| 888 |
+
"count": 39
|
| 889 |
},
|
| 890 |
"UC": {
|
| 891 |
"accuracy": 1.0,
|
| 892 |
+
"count": 560
|
| 893 |
},
|
| 894 |
"US": {
|
| 895 |
"accuracy": 1.0,
|
| 896 |
+
"count": 54
|
| 897 |
}
|
| 898 |
}
|
| 899 |
},
|
| 900 |
+
"add_C1": {
|
| 901 |
"full_accuracy": 1.0,
|
| 902 |
+
"digit_accuracy": 1.0,
|
| 903 |
+
"n_examples": 50,
|
| 904 |
"per_subtask": {
|
| 905 |
"SA": {
|
| 906 |
"accuracy": 1.0,
|
| 907 |
+
"count": 250
|
| 908 |
},
|
| 909 |
"SC": {
|
| 910 |
"accuracy": 1.0,
|
| 911 |
+
"count": 50
|
| 912 |
},
|
| 913 |
"UC": {
|
| 914 |
"accuracy": 1.0,
|
| 915 |
+
"count": 50
|
| 916 |
+
}
|
| 917 |
+
}
|
| 918 |
+
},
|
| 919 |
+
"add_C2": {
|
| 920 |
+
"full_accuracy": 1.0,
|
| 921 |
+
"digit_accuracy": 1.0,
|
| 922 |
+
"n_examples": 50,
|
| 923 |
+
"per_subtask": {
|
| 924 |
+
"SA": {
|
| 925 |
+
"accuracy": 1.0,
|
| 926 |
+
"count": 200
|
| 927 |
+
},
|
| 928 |
+
"SC": {
|
| 929 |
+
"accuracy": 1.0,
|
| 930 |
+
"count": 50
|
| 931 |
+
},
|
| 932 |
+
"UC": {
|
| 933 |
+
"accuracy": 1.0,
|
| 934 |
+
"count": 83
|
| 935 |
},
|
| 936 |
"US": {
|
| 937 |
"accuracy": 1.0,
|
| 938 |
+
"count": 17
|
| 939 |
}
|
| 940 |
}
|
| 941 |
},
|
| 942 |
+
"add_C3": {
|
| 943 |
+
"full_accuracy": 0.96,
|
| 944 |
+
"digit_accuracy": 0.9942857142857143,
|
| 945 |
+
"n_examples": 50,
|
| 946 |
"per_subtask": {
|
| 947 |
"SA": {
|
| 948 |
"accuracy": 1.0,
|
| 949 |
+
"count": 150
|
| 950 |
},
|
| 951 |
"SC": {
|
| 952 |
+
"accuracy": 1.0,
|
| 953 |
+
"count": 50
|
| 954 |
+
},
|
| 955 |
+
"UC": {
|
| 956 |
+
"accuracy": 0.98,
|
| 957 |
+
"count": 100
|
| 958 |
+
},
|
| 959 |
+
"US": {
|
| 960 |
+
"accuracy": 1.0,
|
| 961 |
+
"count": 50
|
| 962 |
+
}
|
| 963 |
+
}
|
| 964 |
+
},
|
| 965 |
+
"add_C4": {
|
| 966 |
+
"full_accuracy": 0.96,
|
| 967 |
+
"digit_accuracy": 0.9942857142857143,
|
| 968 |
+
"n_examples": 50,
|
| 969 |
+
"per_subtask": {
|
| 970 |
+
"SA": {
|
| 971 |
"accuracy": 1.0,
|
| 972 |
"count": 100
|
| 973 |
},
|
| 974 |
+
"SC": {
|
| 975 |
+
"accuracy": 1.0,
|
| 976 |
+
"count": 50
|
| 977 |
+
},
|
| 978 |
"UC": {
|
| 979 |
+
"accuracy": 0.9848484848484849,
|
| 980 |
+
"count": 132
|
| 981 |
},
|
| 982 |
"US": {
|
| 983 |
"accuracy": 1.0,
|
| 984 |
+
"count": 68
|
| 985 |
}
|
| 986 |
}
|
| 987 |
},
|
| 988 |
"add_C5": {
|
| 989 |
+
"full_accuracy": 0.94,
|
| 990 |
+
"digit_accuracy": 0.9914285714285714,
|
| 991 |
+
"n_examples": 50,
|
| 992 |
"per_subtask": {
|
| 993 |
"SA": {
|
| 994 |
"accuracy": 1.0,
|
| 995 |
+
"count": 50
|
| 996 |
},
|
| 997 |
"SC": {
|
| 998 |
"accuracy": 1.0,
|
| 999 |
+
"count": 50
|
| 1000 |
},
|
| 1001 |
"UC": {
|
| 1002 |
+
"accuracy": 0.9863013698630136,
|
| 1003 |
+
"count": 146
|
| 1004 |
},
|
| 1005 |
"US": {
|
| 1006 |
+
"accuracy": 0.9903846153846154,
|
| 1007 |
+
"count": 104
|
| 1008 |
}
|
| 1009 |
}
|
| 1010 |
},
|
| 1011 |
"add_C6": {
|
| 1012 |
"full_accuracy": 1.0,
|
| 1013 |
+
"digit_accuracy": 1.0,
|
| 1014 |
+
"n_examples": 50,
|
| 1015 |
"per_subtask": {
|
| 1016 |
"SC": {
|
| 1017 |
"accuracy": 1.0,
|
| 1018 |
+
"count": 50
|
| 1019 |
},
|
| 1020 |
"UC": {
|
| 1021 |
"accuracy": 1.0,
|
| 1022 |
+
"count": 189
|
| 1023 |
},
|
| 1024 |
"US": {
|
| 1025 |
"accuracy": 1.0,
|
| 1026 |
+
"count": 111
|
| 1027 |
}
|
| 1028 |
}
|
| 1029 |
},
|
| 1030 |
"sub_M0": {
|
| 1031 |
"full_accuracy": 1.0,
|
| 1032 |
+
"digit_accuracy": 1.0,
|
| 1033 |
+
"n_examples": 50,
|
| 1034 |
"per_subtask": {
|
| 1035 |
"MD": {
|
| 1036 |
"accuracy": 1.0,
|
| 1037 |
+
"count": 303
|
| 1038 |
},
|
| 1039 |
"ME": {
|
| 1040 |
"accuracy": 1.0,
|
| 1041 |
+
"count": 47
|
| 1042 |
}
|
| 1043 |
}
|
| 1044 |
},
|
| 1045 |
"sub_M1": {
|
| 1046 |
"full_accuracy": 1.0,
|
| 1047 |
+
"digit_accuracy": 1.0,
|
| 1048 |
+
"n_examples": 50,
|
| 1049 |
"per_subtask": {
|
| 1050 |
"MD": {
|
| 1051 |
"accuracy": 1.0,
|
| 1052 |
+
"count": 141
|
| 1053 |
},
|
| 1054 |
"MB": {
|
| 1055 |
"accuracy": 1.0,
|
| 1056 |
+
"count": 72
|
| 1057 |
},
|
| 1058 |
"ME": {
|
| 1059 |
"accuracy": 1.0,
|
| 1060 |
+
"count": 18
|
| 1061 |
},
|
| 1062 |
"UB": {
|
| 1063 |
"accuracy": 1.0,
|
| 1064 |
+
"count": 119
|
| 1065 |
}
|
| 1066 |
}
|
| 1067 |
},
|
| 1068 |
"sub_M2": {
|
| 1069 |
"full_accuracy": 1.0,
|
| 1070 |
+
"digit_accuracy": 1.0,
|
| 1071 |
+
"n_examples": 50,
|
| 1072 |
"per_subtask": {
|
| 1073 |
"MD": {
|
| 1074 |
"accuracy": 1.0,
|
| 1075 |
+
"count": 112
|
| 1076 |
},
|
| 1077 |
"MB": {
|
| 1078 |
"accuracy": 1.0,
|
| 1079 |
+
"count": 53
|
| 1080 |
},
|
| 1081 |
"ME": {
|
| 1082 |
"accuracy": 1.0,
|
| 1083 |
+
"count": 47
|
| 1084 |
},
|
| 1085 |
"UB": {
|
| 1086 |
"accuracy": 1.0,
|
| 1087 |
+
"count": 85
|
| 1088 |
},
|
| 1089 |
"UD": {
|
| 1090 |
"accuracy": 1.0,
|
| 1091 |
+
"count": 53
|
| 1092 |
}
|
| 1093 |
}
|
| 1094 |
},
|
| 1095 |
"sub_M3": {
|
| 1096 |
+
"full_accuracy": 0.98,
|
| 1097 |
+
"digit_accuracy": 0.9971428571428571,
|
| 1098 |
+
"n_examples": 50,
|
| 1099 |
"per_subtask": {
|
| 1100 |
"MD": {
|
| 1101 |
"accuracy": 1.0,
|
| 1102 |
+
"count": 97
|
| 1103 |
},
|
| 1104 |
"MB": {
|
| 1105 |
"accuracy": 1.0,
|
| 1106 |
+
"count": 51
|
| 1107 |
},
|
| 1108 |
"ME": {
|
| 1109 |
"accuracy": 1.0,
|
| 1110 |
+
"count": 27
|
| 1111 |
},
|
| 1112 |
"UB": {
|
| 1113 |
+
"accuracy": 0.9864864864864865,
|
| 1114 |
+
"count": 74
|
| 1115 |
},
|
| 1116 |
"UD": {
|
| 1117 |
"accuracy": 1.0,
|
| 1118 |
+
"count": 101
|
| 1119 |
}
|
| 1120 |
}
|
| 1121 |
},
|
| 1122 |
"sub_M4": {
|
| 1123 |
+
"full_accuracy": 0.58,
|
| 1124 |
+
"digit_accuracy": 0.94,
|
| 1125 |
+
"n_examples": 50,
|
| 1126 |
"per_subtask": {
|
| 1127 |
"MD": {
|
| 1128 |
"accuracy": 1.0,
|
| 1129 |
+
"count": 100
|
| 1130 |
},
|
| 1131 |
"MB": {
|
| 1132 |
"accuracy": 1.0,
|
| 1133 |
+
"count": 50
|
| 1134 |
},
|
| 1135 |
"UB": {
|
| 1136 |
+
"accuracy": 0.58,
|
| 1137 |
+
"count": 50
|
| 1138 |
},
|
| 1139 |
"UD": {
|
| 1140 |
+
"accuracy": 1.0,
|
| 1141 |
+
"count": 150
|
| 1142 |
}
|
| 1143 |
}
|
| 1144 |
},
|
| 1145 |
"sub_M5": {
|
| 1146 |
+
"full_accuracy": 0.36,
|
| 1147 |
+
"digit_accuracy": 0.86,
|
| 1148 |
+
"n_examples": 50,
|
| 1149 |
"per_subtask": {
|
| 1150 |
"MD": {
|
| 1151 |
"accuracy": 1.0,
|
| 1152 |
+
"count": 50
|
| 1153 |
},
|
| 1154 |
"MB": {
|
| 1155 |
"accuracy": 1.0,
|
| 1156 |
+
"count": 50
|
| 1157 |
},
|
| 1158 |
"UB": {
|
| 1159 |
+
"accuracy": 0.54,
|
| 1160 |
+
"count": 50
|
| 1161 |
},
|
| 1162 |
"UD": {
|
| 1163 |
"accuracy": 0.87,
|
| 1164 |
+
"count": 200
|
| 1165 |
}
|
| 1166 |
}
|
| 1167 |
},
|
| 1168 |
"sub_random": {
|
| 1169 |
"full_accuracy": 1.0,
|
| 1170 |
+
"digit_accuracy": 1.0,
|
| 1171 |
"n_examples": 200,
|
| 1172 |
"per_subtask": {
|
| 1173 |
"MD": {
|
| 1174 |
"accuracy": 1.0,
|
| 1175 |
+
"count": 570
|
| 1176 |
},
|
| 1177 |
"MB": {
|
| 1178 |
"accuracy": 1.0,
|
| 1179 |
+
"count": 277
|
| 1180 |
},
|
| 1181 |
"ME": {
|
| 1182 |
"accuracy": 1.0,
|
|
|
|
| 1184 |
},
|
| 1185 |
"UB": {
|
| 1186 |
"accuracy": 1.0,
|
| 1187 |
+
"count": 471
|
| 1188 |
},
|
| 1189 |
"UD": {
|
| 1190 |
"accuracy": 1.0,
|
| 1191 |
+
"count": 29
|
| 1192 |
}
|
| 1193 |
}
|
| 1194 |
},
|
| 1195 |
"sub_B3": {
|
| 1196 |
+
"full_accuracy": 1.0,
|
| 1197 |
+
"digit_accuracy": 1.0,
|
| 1198 |
+
"n_examples": 50,
|
| 1199 |
"per_subtask": {
|
| 1200 |
"MD": {
|
| 1201 |
+
"accuracy": 1.0,
|
| 1202 |
+
"count": 150
|
| 1203 |
},
|
| 1204 |
"MB": {
|
| 1205 |
"accuracy": 1.0,
|
| 1206 |
+
"count": 50
|
| 1207 |
},
|
| 1208 |
"UB": {
|
| 1209 |
"accuracy": 1.0,
|
| 1210 |
+
"count": 101
|
| 1211 |
},
|
| 1212 |
"UD": {
|
| 1213 |
"accuracy": 1.0,
|
| 1214 |
+
"count": 49
|
| 1215 |
}
|
| 1216 |
}
|
| 1217 |
},
|
| 1218 |
"sub_B4": {
|
| 1219 |
+
"full_accuracy": 0.9,
|
| 1220 |
+
"digit_accuracy": 0.9857142857142858,
|
| 1221 |
+
"n_examples": 50,
|
| 1222 |
"per_subtask": {
|
| 1223 |
"MD": {
|
| 1224 |
+
"accuracy": 1.0,
|
| 1225 |
+
"count": 100
|
| 1226 |
},
|
| 1227 |
"MB": {
|
| 1228 |
"accuracy": 1.0,
|
| 1229 |
+
"count": 50
|
| 1230 |
},
|
| 1231 |
"UB": {
|
| 1232 |
+
"accuracy": 0.9586776859504132,
|
| 1233 |
+
"count": 121
|
| 1234 |
},
|
| 1235 |
"UD": {
|
| 1236 |
"accuracy": 1.0,
|
| 1237 |
+
"count": 79
|
| 1238 |
}
|
| 1239 |
}
|
| 1240 |
},
|
| 1241 |
"sub_B5": {
|
| 1242 |
+
"full_accuracy": 0.92,
|
| 1243 |
+
"digit_accuracy": 0.9885714285714285,
|
| 1244 |
+
"n_examples": 50,
|
| 1245 |
"per_subtask": {
|
| 1246 |
"MD": {
|
| 1247 |
"accuracy": 1.0,
|
| 1248 |
+
"count": 50
|
| 1249 |
},
|
| 1250 |
"MB": {
|
| 1251 |
"accuracy": 1.0,
|
| 1252 |
+
"count": 50
|
| 1253 |
},
|
| 1254 |
"UB": {
|
| 1255 |
+
"accuracy": 0.9736842105263158,
|
| 1256 |
+
"count": 152
|
| 1257 |
},
|
| 1258 |
"UD": {
|
| 1259 |
+
"accuracy": 1.0,
|
| 1260 |
+
"count": 98
|
| 1261 |
}
|
| 1262 |
}
|
| 1263 |
}
|
| 1264 |
},
|
| 1265 |
"summary": {
|
| 1266 |
+
"overall_accuracy": 0.938,
|
| 1267 |
+
"digit_accuracy": 0.9884761904761905,
|
| 1268 |
+
"total_examples": 1500,
|
| 1269 |
+
"n_splits": 24
|
| 1270 |
}
|
| 1271 |
}
|
| 1272 |
}
|