Upload add_sub_baseline_100K_1L2H256d/metrics.json with huggingface_hub
Browse files
add_sub_baseline_100K_1L2H256d/metrics.json
CHANGED
|
@@ -2582,502 +2582,567 @@
|
|
| 2582 |
"K": null,
|
| 2583 |
"mode": "sft",
|
| 2584 |
"n_digits": 6,
|
| 2585 |
-
"n_per_split":
|
| 2586 |
},
|
| 2587 |
"splits": {
|
| 2588 |
"add_S0": {
|
| 2589 |
-
"full_accuracy": 0.
|
| 2590 |
-
"
|
|
|
|
| 2591 |
"per_subtask": {
|
| 2592 |
"SA": {
|
| 2593 |
-
"accuracy": 0.
|
| 2594 |
-
"count":
|
| 2595 |
},
|
| 2596 |
"SS": {
|
| 2597 |
-
"accuracy": 0.
|
| 2598 |
-
"count":
|
| 2599 |
}
|
| 2600 |
}
|
| 2601 |
},
|
| 2602 |
"add_S1": {
|
| 2603 |
-
"full_accuracy": 0.
|
| 2604 |
-
"
|
|
|
|
| 2605 |
"per_subtask": {
|
| 2606 |
"SA": {
|
| 2607 |
-
"accuracy": 0.
|
| 2608 |
-
"count":
|
| 2609 |
},
|
| 2610 |
"SC": {
|
| 2611 |
-
"accuracy": 0.
|
| 2612 |
-
"count":
|
| 2613 |
},
|
| 2614 |
"SS": {
|
| 2615 |
-
"accuracy":
|
| 2616 |
-
"count":
|
| 2617 |
},
|
| 2618 |
"UC": {
|
| 2619 |
-
"accuracy": 0.
|
| 2620 |
-
"count":
|
| 2621 |
}
|
| 2622 |
}
|
| 2623 |
},
|
| 2624 |
"add_S2": {
|
| 2625 |
-
"full_accuracy": 0.
|
| 2626 |
-
"
|
|
|
|
| 2627 |
"per_subtask": {
|
| 2628 |
"SA": {
|
| 2629 |
-
"accuracy": 0.
|
| 2630 |
-
"count":
|
| 2631 |
},
|
| 2632 |
"SC": {
|
| 2633 |
-
"accuracy": 0.
|
| 2634 |
-
"count":
|
| 2635 |
},
|
| 2636 |
"SS": {
|
| 2637 |
-
"accuracy": 0.
|
| 2638 |
-
"count":
|
| 2639 |
},
|
| 2640 |
"UC": {
|
| 2641 |
-
"accuracy": 0.
|
| 2642 |
-
"count":
|
| 2643 |
},
|
| 2644 |
"US": {
|
| 2645 |
-
"accuracy": 0.
|
| 2646 |
-
"count":
|
| 2647 |
}
|
| 2648 |
}
|
| 2649 |
},
|
| 2650 |
"add_S3": {
|
| 2651 |
-
"full_accuracy": 0.
|
| 2652 |
-
"
|
|
|
|
| 2653 |
"per_subtask": {
|
| 2654 |
"SA": {
|
| 2655 |
-
"accuracy": 0.
|
| 2656 |
-
"count":
|
| 2657 |
},
|
| 2658 |
"SC": {
|
| 2659 |
-
"accuracy": 0.
|
| 2660 |
-
"count":
|
| 2661 |
},
|
| 2662 |
"SS": {
|
| 2663 |
-
"accuracy": 0.
|
| 2664 |
-
"count":
|
| 2665 |
},
|
| 2666 |
"UC": {
|
| 2667 |
-
"accuracy": 0.
|
| 2668 |
-
"count":
|
| 2669 |
},
|
| 2670 |
"US": {
|
| 2671 |
-
"accuracy": 0.
|
| 2672 |
-
"count":
|
| 2673 |
}
|
| 2674 |
}
|
| 2675 |
},
|
| 2676 |
"add_S4": {
|
| 2677 |
-
"full_accuracy": 0.
|
| 2678 |
-
"
|
|
|
|
| 2679 |
"per_subtask": {
|
| 2680 |
"SA": {
|
| 2681 |
-
"accuracy":
|
| 2682 |
-
"count":
|
| 2683 |
},
|
| 2684 |
"SC": {
|
| 2685 |
-
"accuracy": 0.
|
| 2686 |
-
"count":
|
| 2687 |
},
|
| 2688 |
"SS": {
|
| 2689 |
-
"accuracy": 0.
|
| 2690 |
-
"count":
|
| 2691 |
},
|
| 2692 |
"UC": {
|
| 2693 |
-
"accuracy": 0.
|
| 2694 |
-
"count":
|
| 2695 |
},
|
| 2696 |
"US": {
|
| 2697 |
-
"accuracy": 0.
|
| 2698 |
-
"count":
|
| 2699 |
}
|
| 2700 |
}
|
| 2701 |
},
|
| 2702 |
"add_S5": {
|
| 2703 |
-
"full_accuracy": 0.
|
| 2704 |
-
"
|
|
|
|
| 2705 |
"per_subtask": {
|
| 2706 |
"SA": {
|
| 2707 |
"accuracy": 1.0,
|
| 2708 |
-
"count":
|
| 2709 |
},
|
| 2710 |
"SC": {
|
| 2711 |
-
"accuracy":
|
| 2712 |
-
"count":
|
| 2713 |
},
|
| 2714 |
"UC": {
|
| 2715 |
-
"accuracy": 0.
|
| 2716 |
-
"count":
|
| 2717 |
},
|
| 2718 |
"US": {
|
| 2719 |
-
"accuracy": 0.
|
| 2720 |
-
"count":
|
| 2721 |
}
|
| 2722 |
}
|
| 2723 |
},
|
| 2724 |
"add_S6": {
|
| 2725 |
-
"full_accuracy": 0.
|
| 2726 |
-
"
|
|
|
|
| 2727 |
"per_subtask": {
|
| 2728 |
"SC": {
|
| 2729 |
"accuracy": 1.0,
|
| 2730 |
-
"count":
|
| 2731 |
},
|
| 2732 |
"UC": {
|
| 2733 |
-
"accuracy": 0.
|
| 2734 |
-
"count":
|
| 2735 |
},
|
| 2736 |
"US": {
|
| 2737 |
-
"accuracy": 0.
|
| 2738 |
-
"count":
|
| 2739 |
}
|
| 2740 |
}
|
| 2741 |
},
|
| 2742 |
"add_random": {
|
| 2743 |
-
"full_accuracy": 0.
|
|
|
|
| 2744 |
"n_examples": 200,
|
| 2745 |
"per_subtask": {
|
| 2746 |
"SA": {
|
| 2747 |
-
"accuracy": 0.
|
| 2748 |
-
"count":
|
| 2749 |
},
|
| 2750 |
"SC": {
|
| 2751 |
-
"accuracy": 0.
|
| 2752 |
-
"count":
|
| 2753 |
},
|
| 2754 |
"SS": {
|
| 2755 |
-
"accuracy":
|
| 2756 |
-
"count":
|
| 2757 |
},
|
| 2758 |
"UC": {
|
| 2759 |
-
"accuracy": 0.
|
| 2760 |
-
"count":
|
| 2761 |
},
|
| 2762 |
"US": {
|
| 2763 |
-
"accuracy": 0.
|
| 2764 |
-
"count":
|
| 2765 |
}
|
| 2766 |
}
|
| 2767 |
},
|
| 2768 |
-
"
|
| 2769 |
-
"full_accuracy": 0.
|
| 2770 |
-
"
|
|
|
|
| 2771 |
"per_subtask": {
|
| 2772 |
"SA": {
|
| 2773 |
-
"accuracy": 0.
|
| 2774 |
-
"count":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2775 |
},
|
| 2776 |
"SC": {
|
| 2777 |
"accuracy": 1.0,
|
| 2778 |
-
"count":
|
| 2779 |
},
|
| 2780 |
"UC": {
|
| 2781 |
-
"accuracy": 0.
|
| 2782 |
-
"count":
|
| 2783 |
},
|
| 2784 |
"US": {
|
| 2785 |
-
"accuracy": 0.
|
| 2786 |
-
"count":
|
| 2787 |
}
|
| 2788 |
}
|
| 2789 |
},
|
| 2790 |
-
"
|
| 2791 |
-
"full_accuracy": 0.
|
| 2792 |
-
"
|
|
|
|
| 2793 |
"per_subtask": {
|
| 2794 |
"SA": {
|
| 2795 |
-
"accuracy": 0.
|
| 2796 |
-
"count":
|
| 2797 |
},
|
| 2798 |
"SC": {
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2799 |
"accuracy": 0.99,
|
| 2800 |
"count": 100
|
| 2801 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2802 |
"UC": {
|
| 2803 |
-
"accuracy": 0.
|
| 2804 |
-
"count":
|
| 2805 |
},
|
| 2806 |
"US": {
|
| 2807 |
-
"accuracy": 0.
|
| 2808 |
-
"count":
|
| 2809 |
}
|
| 2810 |
}
|
| 2811 |
},
|
| 2812 |
"add_C5": {
|
| 2813 |
-
"full_accuracy": 0.
|
| 2814 |
-
"
|
|
|
|
| 2815 |
"per_subtask": {
|
| 2816 |
"SA": {
|
| 2817 |
"accuracy": 1.0,
|
| 2818 |
-
"count":
|
| 2819 |
},
|
| 2820 |
"SC": {
|
| 2821 |
"accuracy": 1.0,
|
| 2822 |
-
"count":
|
| 2823 |
},
|
| 2824 |
"UC": {
|
| 2825 |
-
"accuracy": 0.
|
| 2826 |
-
"count":
|
| 2827 |
},
|
| 2828 |
"US": {
|
| 2829 |
-
"accuracy": 0.
|
| 2830 |
-
"count":
|
| 2831 |
}
|
| 2832 |
}
|
| 2833 |
},
|
| 2834 |
"add_C6": {
|
| 2835 |
-
"full_accuracy": 0.
|
| 2836 |
-
"
|
|
|
|
| 2837 |
"per_subtask": {
|
| 2838 |
"SC": {
|
| 2839 |
"accuracy": 1.0,
|
| 2840 |
-
"count":
|
| 2841 |
},
|
| 2842 |
"UC": {
|
| 2843 |
-
"accuracy": 0.
|
| 2844 |
-
"count":
|
| 2845 |
},
|
| 2846 |
"US": {
|
| 2847 |
-
"accuracy": 0.
|
| 2848 |
-
"count":
|
| 2849 |
}
|
| 2850 |
}
|
| 2851 |
},
|
| 2852 |
"sub_M0": {
|
| 2853 |
-
"full_accuracy": 0.
|
| 2854 |
-
"
|
|
|
|
| 2855 |
"per_subtask": {
|
| 2856 |
"MD": {
|
| 2857 |
-
"accuracy": 0.
|
| 2858 |
-
"count":
|
| 2859 |
},
|
| 2860 |
"ME": {
|
| 2861 |
-
"accuracy":
|
| 2862 |
-
"count":
|
| 2863 |
}
|
| 2864 |
}
|
| 2865 |
},
|
| 2866 |
"sub_M1": {
|
| 2867 |
-
"full_accuracy": 0.
|
| 2868 |
-
"
|
|
|
|
| 2869 |
"per_subtask": {
|
| 2870 |
"MD": {
|
| 2871 |
-
"accuracy": 0.
|
| 2872 |
-
"count":
|
| 2873 |
},
|
| 2874 |
"MB": {
|
| 2875 |
-
"accuracy": 0.
|
| 2876 |
-
"count":
|
| 2877 |
},
|
| 2878 |
"ME": {
|
| 2879 |
"accuracy": 1.0,
|
| 2880 |
-
"count":
|
| 2881 |
},
|
| 2882 |
"UB": {
|
| 2883 |
-
"accuracy": 0.
|
| 2884 |
-
"count":
|
| 2885 |
}
|
| 2886 |
}
|
| 2887 |
},
|
| 2888 |
"sub_M2": {
|
| 2889 |
-
"full_accuracy": 0.
|
| 2890 |
-
"
|
|
|
|
| 2891 |
"per_subtask": {
|
| 2892 |
"MD": {
|
| 2893 |
-
"accuracy": 0.
|
| 2894 |
-
"count":
|
| 2895 |
},
|
| 2896 |
"MB": {
|
| 2897 |
-
"accuracy": 0.
|
| 2898 |
-
"count":
|
| 2899 |
},
|
| 2900 |
"ME": {
|
| 2901 |
-
"accuracy":
|
| 2902 |
-
"count":
|
| 2903 |
},
|
| 2904 |
"UB": {
|
| 2905 |
-
"accuracy": 0.
|
| 2906 |
-
"count":
|
| 2907 |
},
|
| 2908 |
"UD": {
|
| 2909 |
-
"accuracy": 0.
|
| 2910 |
-
"count":
|
| 2911 |
}
|
| 2912 |
}
|
| 2913 |
},
|
| 2914 |
"sub_M3": {
|
| 2915 |
-
"full_accuracy": 0.
|
| 2916 |
-
"
|
|
|
|
| 2917 |
"per_subtask": {
|
| 2918 |
"MD": {
|
| 2919 |
-
"accuracy": 0.
|
| 2920 |
-
"count":
|
| 2921 |
},
|
| 2922 |
"MB": {
|
| 2923 |
-
"accuracy": 0.
|
| 2924 |
-
"count":
|
| 2925 |
},
|
| 2926 |
"ME": {
|
| 2927 |
-
"accuracy":
|
| 2928 |
-
"count":
|
| 2929 |
},
|
| 2930 |
"UB": {
|
| 2931 |
-
"accuracy": 0.
|
| 2932 |
-
"count":
|
| 2933 |
},
|
| 2934 |
"UD": {
|
| 2935 |
-
"accuracy": 0.
|
| 2936 |
-
"count":
|
| 2937 |
}
|
| 2938 |
}
|
| 2939 |
},
|
| 2940 |
"sub_M4": {
|
| 2941 |
-
"full_accuracy": 0.
|
| 2942 |
-
"
|
|
|
|
| 2943 |
"per_subtask": {
|
| 2944 |
"MD": {
|
| 2945 |
"accuracy": 0.98,
|
| 2946 |
-
"count":
|
| 2947 |
},
|
| 2948 |
"MB": {
|
| 2949 |
-
"accuracy": 0.
|
| 2950 |
-
"count":
|
| 2951 |
},
|
| 2952 |
"UB": {
|
| 2953 |
-
"accuracy": 0.
|
| 2954 |
-
"count":
|
| 2955 |
},
|
| 2956 |
"UD": {
|
| 2957 |
"accuracy": 0.29333333333333333,
|
| 2958 |
-
"count":
|
| 2959 |
}
|
| 2960 |
}
|
| 2961 |
},
|
| 2962 |
"sub_M5": {
|
| 2963 |
-
"full_accuracy": 0.
|
| 2964 |
-
"
|
|
|
|
| 2965 |
"per_subtask": {
|
| 2966 |
"MD": {
|
| 2967 |
"accuracy": 1.0,
|
| 2968 |
-
"count":
|
| 2969 |
},
|
| 2970 |
"MB": {
|
| 2971 |
"accuracy": 1.0,
|
| 2972 |
-
"count":
|
| 2973 |
},
|
| 2974 |
"UB": {
|
| 2975 |
-
"accuracy": 0.
|
| 2976 |
-
"count":
|
| 2977 |
},
|
| 2978 |
"UD": {
|
| 2979 |
-
"accuracy": 0.
|
| 2980 |
-
"count":
|
| 2981 |
}
|
| 2982 |
}
|
| 2983 |
},
|
| 2984 |
"sub_random": {
|
| 2985 |
-
"full_accuracy": 0.
|
|
|
|
| 2986 |
"n_examples": 200,
|
| 2987 |
"per_subtask": {
|
| 2988 |
"MD": {
|
| 2989 |
-
"accuracy": 0.
|
| 2990 |
-
"count":
|
| 2991 |
},
|
| 2992 |
"MB": {
|
| 2993 |
-
"accuracy": 0.
|
| 2994 |
-
"count":
|
| 2995 |
},
|
| 2996 |
"ME": {
|
| 2997 |
-
"accuracy":
|
| 2998 |
"count": 53
|
| 2999 |
},
|
| 3000 |
"UB": {
|
| 3001 |
-
"accuracy": 0.
|
| 3002 |
-
"count":
|
| 3003 |
},
|
| 3004 |
"UD": {
|
| 3005 |
-
"accuracy": 0.
|
| 3006 |
-
"count":
|
| 3007 |
}
|
| 3008 |
}
|
| 3009 |
},
|
| 3010 |
"sub_B3": {
|
| 3011 |
-
"full_accuracy": 0.
|
| 3012 |
-
"
|
|
|
|
| 3013 |
"per_subtask": {
|
| 3014 |
"MD": {
|
| 3015 |
-
"accuracy": 0.
|
| 3016 |
-
"count":
|
| 3017 |
},
|
| 3018 |
"MB": {
|
| 3019 |
"accuracy": 0.96,
|
| 3020 |
-
"count":
|
| 3021 |
},
|
| 3022 |
"UB": {
|
| 3023 |
-
"accuracy": 0.
|
| 3024 |
-
"count":
|
| 3025 |
},
|
| 3026 |
"UD": {
|
| 3027 |
-
"accuracy": 0.
|
| 3028 |
-
"count":
|
| 3029 |
}
|
| 3030 |
}
|
| 3031 |
},
|
| 3032 |
"sub_B4": {
|
| 3033 |
-
"full_accuracy": 0.
|
| 3034 |
-
"
|
|
|
|
| 3035 |
"per_subtask": {
|
| 3036 |
"MD": {
|
| 3037 |
-
"accuracy": 0.
|
| 3038 |
-
"count":
|
| 3039 |
},
|
| 3040 |
"MB": {
|
| 3041 |
-
"accuracy": 0.
|
| 3042 |
-
"count":
|
| 3043 |
},
|
| 3044 |
"UB": {
|
| 3045 |
-
"accuracy": 0.
|
| 3046 |
-
"count":
|
| 3047 |
},
|
| 3048 |
"UD": {
|
| 3049 |
-
"accuracy": 0.
|
| 3050 |
-
"count":
|
| 3051 |
}
|
| 3052 |
}
|
| 3053 |
},
|
| 3054 |
"sub_B5": {
|
| 3055 |
-
"full_accuracy": 0.
|
| 3056 |
-
"
|
|
|
|
| 3057 |
"per_subtask": {
|
| 3058 |
"MD": {
|
| 3059 |
"accuracy": 1.0,
|
| 3060 |
-
"count":
|
| 3061 |
},
|
| 3062 |
"MB": {
|
| 3063 |
"accuracy": 1.0,
|
| 3064 |
-
"count":
|
| 3065 |
},
|
| 3066 |
"UB": {
|
| 3067 |
-
"accuracy": 0.
|
| 3068 |
-
"count":
|
| 3069 |
},
|
| 3070 |
"UD": {
|
| 3071 |
-
"accuracy": 0.
|
| 3072 |
-
"count":
|
| 3073 |
}
|
| 3074 |
}
|
| 3075 |
}
|
| 3076 |
},
|
| 3077 |
"summary": {
|
| 3078 |
-
"overall_accuracy": 0.
|
| 3079 |
-
"
|
| 3080 |
-
"
|
|
|
|
| 3081 |
}
|
| 3082 |
}
|
| 3083 |
}
|
|
|
|
| 2582 |
"K": null,
|
| 2583 |
"mode": "sft",
|
| 2584 |
"n_digits": 6,
|
| 2585 |
+
"n_per_split": 50
|
| 2586 |
},
|
| 2587 |
"splits": {
|
| 2588 |
"add_S0": {
|
| 2589 |
+
"full_accuracy": 0.68,
|
| 2590 |
+
"digit_accuracy": 0.9485714285714286,
|
| 2591 |
+
"n_examples": 50,
|
| 2592 |
"per_subtask": {
|
| 2593 |
"SA": {
|
| 2594 |
+
"accuracy": 0.9457627118644067,
|
| 2595 |
+
"count": 295
|
| 2596 |
},
|
| 2597 |
"SS": {
|
| 2598 |
+
"accuracy": 0.9636363636363636,
|
| 2599 |
+
"count": 55
|
| 2600 |
}
|
| 2601 |
}
|
| 2602 |
},
|
| 2603 |
"add_S1": {
|
| 2604 |
+
"full_accuracy": 0.76,
|
| 2605 |
+
"digit_accuracy": 0.9542857142857143,
|
| 2606 |
+
"n_examples": 50,
|
| 2607 |
"per_subtask": {
|
| 2608 |
"SA": {
|
| 2609 |
+
"accuracy": 0.9761904761904762,
|
| 2610 |
+
"count": 126
|
| 2611 |
},
|
| 2612 |
"SC": {
|
| 2613 |
+
"accuracy": 0.9240506329113924,
|
| 2614 |
+
"count": 79
|
| 2615 |
},
|
| 2616 |
"SS": {
|
| 2617 |
+
"accuracy": 1.0,
|
| 2618 |
+
"count": 21
|
| 2619 |
},
|
| 2620 |
"UC": {
|
| 2621 |
+
"accuracy": 0.9435483870967742,
|
| 2622 |
+
"count": 124
|
| 2623 |
}
|
| 2624 |
}
|
| 2625 |
},
|
| 2626 |
"add_S2": {
|
| 2627 |
+
"full_accuracy": 0.32,
|
| 2628 |
+
"digit_accuracy": 0.8542857142857143,
|
| 2629 |
+
"n_examples": 50,
|
| 2630 |
"per_subtask": {
|
| 2631 |
"SA": {
|
| 2632 |
+
"accuracy": 0.9466666666666667,
|
| 2633 |
+
"count": 75
|
| 2634 |
},
|
| 2635 |
"SC": {
|
| 2636 |
+
"accuracy": 0.8548387096774194,
|
| 2637 |
+
"count": 62
|
| 2638 |
},
|
| 2639 |
"SS": {
|
| 2640 |
+
"accuracy": 0.8205128205128205,
|
| 2641 |
+
"count": 39
|
| 2642 |
},
|
| 2643 |
"UC": {
|
| 2644 |
+
"accuracy": 0.7747747747747747,
|
| 2645 |
+
"count": 111
|
| 2646 |
},
|
| 2647 |
"US": {
|
| 2648 |
+
"accuracy": 0.9047619047619048,
|
| 2649 |
+
"count": 63
|
| 2650 |
}
|
| 2651 |
}
|
| 2652 |
},
|
| 2653 |
"add_S3": {
|
| 2654 |
+
"full_accuracy": 0.18,
|
| 2655 |
+
"digit_accuracy": 0.7857142857142857,
|
| 2656 |
+
"n_examples": 50,
|
| 2657 |
"per_subtask": {
|
| 2658 |
"SA": {
|
| 2659 |
+
"accuracy": 0.9333333333333333,
|
| 2660 |
+
"count": 60
|
| 2661 |
},
|
| 2662 |
"SC": {
|
| 2663 |
+
"accuracy": 0.9298245614035088,
|
| 2664 |
+
"count": 57
|
| 2665 |
},
|
| 2666 |
"SS": {
|
| 2667 |
+
"accuracy": 0.8947368421052632,
|
| 2668 |
+
"count": 19
|
| 2669 |
},
|
| 2670 |
"UC": {
|
| 2671 |
+
"accuracy": 0.6923076923076923,
|
| 2672 |
+
"count": 104
|
| 2673 |
},
|
| 2674 |
"US": {
|
| 2675 |
+
"accuracy": 0.7,
|
| 2676 |
+
"count": 110
|
| 2677 |
}
|
| 2678 |
}
|
| 2679 |
},
|
| 2680 |
"add_S4": {
|
| 2681 |
+
"full_accuracy": 0.14,
|
| 2682 |
+
"digit_accuracy": 0.6371428571428571,
|
| 2683 |
+
"n_examples": 50,
|
| 2684 |
"per_subtask": {
|
| 2685 |
"SA": {
|
| 2686 |
+
"accuracy": 1.0,
|
| 2687 |
+
"count": 48
|
| 2688 |
},
|
| 2689 |
"SC": {
|
| 2690 |
+
"accuracy": 0.9423076923076923,
|
| 2691 |
+
"count": 52
|
| 2692 |
},
|
| 2693 |
"SS": {
|
| 2694 |
+
"accuracy": 0.8571428571428571,
|
| 2695 |
+
"count": 7
|
| 2696 |
},
|
| 2697 |
"UC": {
|
| 2698 |
+
"accuracy": 0.5168539325842697,
|
| 2699 |
+
"count": 89
|
| 2700 |
},
|
| 2701 |
"US": {
|
| 2702 |
+
"accuracy": 0.4805194805194805,
|
| 2703 |
+
"count": 154
|
| 2704 |
}
|
| 2705 |
}
|
| 2706 |
},
|
| 2707 |
"add_S5": {
|
| 2708 |
+
"full_accuracy": 0.08,
|
| 2709 |
+
"digit_accuracy": 0.4257142857142857,
|
| 2710 |
+
"n_examples": 50,
|
| 2711 |
"per_subtask": {
|
| 2712 |
"SA": {
|
| 2713 |
"accuracy": 1.0,
|
| 2714 |
+
"count": 50
|
| 2715 |
},
|
| 2716 |
"SC": {
|
| 2717 |
+
"accuracy": 1.0,
|
| 2718 |
+
"count": 50
|
| 2719 |
},
|
| 2720 |
"UC": {
|
| 2721 |
+
"accuracy": 0.24,
|
| 2722 |
+
"count": 50
|
| 2723 |
},
|
| 2724 |
"US": {
|
| 2725 |
+
"accuracy": 0.185,
|
| 2726 |
+
"count": 200
|
| 2727 |
}
|
| 2728 |
}
|
| 2729 |
},
|
| 2730 |
"add_S6": {
|
| 2731 |
+
"full_accuracy": 0.4,
|
| 2732 |
+
"digit_accuracy": 0.52,
|
| 2733 |
+
"n_examples": 50,
|
| 2734 |
"per_subtask": {
|
| 2735 |
"SC": {
|
| 2736 |
"accuracy": 1.0,
|
| 2737 |
+
"count": 50
|
| 2738 |
},
|
| 2739 |
"UC": {
|
| 2740 |
+
"accuracy": 0.4,
|
| 2741 |
+
"count": 50
|
| 2742 |
},
|
| 2743 |
"US": {
|
| 2744 |
+
"accuracy": 0.448,
|
| 2745 |
+
"count": 250
|
| 2746 |
}
|
| 2747 |
}
|
| 2748 |
},
|
| 2749 |
"add_random": {
|
| 2750 |
+
"full_accuracy": 0.62,
|
| 2751 |
+
"digit_accuracy": 0.9321428571428572,
|
| 2752 |
"n_examples": 200,
|
| 2753 |
"per_subtask": {
|
| 2754 |
"SA": {
|
| 2755 |
+
"accuracy": 0.9767981438515081,
|
| 2756 |
+
"count": 431
|
| 2757 |
},
|
| 2758 |
"SC": {
|
| 2759 |
+
"accuracy": 0.9588607594936709,
|
| 2760 |
+
"count": 316
|
| 2761 |
},
|
| 2762 |
"SS": {
|
| 2763 |
+
"accuracy": 1.0,
|
| 2764 |
+
"count": 39
|
| 2765 |
},
|
| 2766 |
"UC": {
|
| 2767 |
+
"accuracy": 0.8892857142857142,
|
| 2768 |
+
"count": 560
|
| 2769 |
},
|
| 2770 |
"US": {
|
| 2771 |
+
"accuracy": 0.8148148148148148,
|
| 2772 |
+
"count": 54
|
| 2773 |
}
|
| 2774 |
}
|
| 2775 |
},
|
| 2776 |
+
"add_C1": {
|
| 2777 |
+
"full_accuracy": 0.86,
|
| 2778 |
+
"digit_accuracy": 0.98,
|
| 2779 |
+
"n_examples": 50,
|
| 2780 |
"per_subtask": {
|
| 2781 |
"SA": {
|
| 2782 |
+
"accuracy": 0.992,
|
| 2783 |
+
"count": 250
|
| 2784 |
+
},
|
| 2785 |
+
"SC": {
|
| 2786 |
+
"accuracy": 0.98,
|
| 2787 |
+
"count": 50
|
| 2788 |
+
},
|
| 2789 |
+
"UC": {
|
| 2790 |
+
"accuracy": 0.92,
|
| 2791 |
+
"count": 50
|
| 2792 |
+
}
|
| 2793 |
+
}
|
| 2794 |
+
},
|
| 2795 |
+
"add_C2": {
|
| 2796 |
+
"full_accuracy": 0.66,
|
| 2797 |
+
"digit_accuracy": 0.9485714285714286,
|
| 2798 |
+
"n_examples": 50,
|
| 2799 |
+
"per_subtask": {
|
| 2800 |
+
"SA": {
|
| 2801 |
+
"accuracy": 0.985,
|
| 2802 |
+
"count": 200
|
| 2803 |
},
|
| 2804 |
"SC": {
|
| 2805 |
"accuracy": 1.0,
|
| 2806 |
+
"count": 50
|
| 2807 |
},
|
| 2808 |
"UC": {
|
| 2809 |
+
"accuracy": 0.8433734939759037,
|
| 2810 |
+
"count": 83
|
| 2811 |
},
|
| 2812 |
"US": {
|
| 2813 |
+
"accuracy": 0.8823529411764706,
|
| 2814 |
+
"count": 17
|
| 2815 |
}
|
| 2816 |
}
|
| 2817 |
},
|
| 2818 |
+
"add_C3": {
|
| 2819 |
+
"full_accuracy": 0.36,
|
| 2820 |
+
"digit_accuracy": 0.8628571428571429,
|
| 2821 |
+
"n_examples": 50,
|
| 2822 |
"per_subtask": {
|
| 2823 |
"SA": {
|
| 2824 |
+
"accuracy": 0.9933333333333333,
|
| 2825 |
+
"count": 150
|
| 2826 |
},
|
| 2827 |
"SC": {
|
| 2828 |
+
"accuracy": 1.0,
|
| 2829 |
+
"count": 50
|
| 2830 |
+
},
|
| 2831 |
+
"UC": {
|
| 2832 |
+
"accuracy": 0.68,
|
| 2833 |
+
"count": 100
|
| 2834 |
+
},
|
| 2835 |
+
"US": {
|
| 2836 |
+
"accuracy": 0.7,
|
| 2837 |
+
"count": 50
|
| 2838 |
+
}
|
| 2839 |
+
}
|
| 2840 |
+
},
|
| 2841 |
+
"add_C4": {
|
| 2842 |
+
"full_accuracy": 0.3,
|
| 2843 |
+
"digit_accuracy": 0.8542857142857143,
|
| 2844 |
+
"n_examples": 50,
|
| 2845 |
+
"per_subtask": {
|
| 2846 |
+
"SA": {
|
| 2847 |
"accuracy": 0.99,
|
| 2848 |
"count": 100
|
| 2849 |
},
|
| 2850 |
+
"SC": {
|
| 2851 |
+
"accuracy": 1.0,
|
| 2852 |
+
"count": 50
|
| 2853 |
+
},
|
| 2854 |
"UC": {
|
| 2855 |
+
"accuracy": 0.7121212121212122,
|
| 2856 |
+
"count": 132
|
| 2857 |
},
|
| 2858 |
"US": {
|
| 2859 |
+
"accuracy": 0.8235294117647058,
|
| 2860 |
+
"count": 68
|
| 2861 |
}
|
| 2862 |
}
|
| 2863 |
},
|
| 2864 |
"add_C5": {
|
| 2865 |
+
"full_accuracy": 0.26,
|
| 2866 |
+
"digit_accuracy": 0.7828571428571428,
|
| 2867 |
+
"n_examples": 50,
|
| 2868 |
"per_subtask": {
|
| 2869 |
"SA": {
|
| 2870 |
"accuracy": 1.0,
|
| 2871 |
+
"count": 50
|
| 2872 |
},
|
| 2873 |
"SC": {
|
| 2874 |
"accuracy": 1.0,
|
| 2875 |
+
"count": 50
|
| 2876 |
},
|
| 2877 |
"UC": {
|
| 2878 |
+
"accuracy": 0.7465753424657534,
|
| 2879 |
+
"count": 146
|
| 2880 |
},
|
| 2881 |
"US": {
|
| 2882 |
+
"accuracy": 0.625,
|
| 2883 |
+
"count": 104
|
| 2884 |
}
|
| 2885 |
}
|
| 2886 |
},
|
| 2887 |
"add_C6": {
|
| 2888 |
+
"full_accuracy": 0.3,
|
| 2889 |
+
"digit_accuracy": 0.7971428571428572,
|
| 2890 |
+
"n_examples": 50,
|
| 2891 |
"per_subtask": {
|
| 2892 |
"SC": {
|
| 2893 |
"accuracy": 1.0,
|
| 2894 |
+
"count": 50
|
| 2895 |
},
|
| 2896 |
"UC": {
|
| 2897 |
+
"accuracy": 0.7724867724867724,
|
| 2898 |
+
"count": 189
|
| 2899 |
},
|
| 2900 |
"US": {
|
| 2901 |
+
"accuracy": 0.7477477477477478,
|
| 2902 |
+
"count": 111
|
| 2903 |
}
|
| 2904 |
}
|
| 2905 |
},
|
| 2906 |
"sub_M0": {
|
| 2907 |
+
"full_accuracy": 0.8,
|
| 2908 |
+
"digit_accuracy": 0.9714285714285714,
|
| 2909 |
+
"n_examples": 50,
|
| 2910 |
"per_subtask": {
|
| 2911 |
"MD": {
|
| 2912 |
+
"accuracy": 0.966996699669967,
|
| 2913 |
+
"count": 303
|
| 2914 |
},
|
| 2915 |
"ME": {
|
| 2916 |
+
"accuracy": 1.0,
|
| 2917 |
+
"count": 47
|
| 2918 |
}
|
| 2919 |
}
|
| 2920 |
},
|
| 2921 |
"sub_M1": {
|
| 2922 |
+
"full_accuracy": 0.58,
|
| 2923 |
+
"digit_accuracy": 0.9314285714285714,
|
| 2924 |
+
"n_examples": 50,
|
| 2925 |
"per_subtask": {
|
| 2926 |
"MD": {
|
| 2927 |
+
"accuracy": 0.9858156028368794,
|
| 2928 |
+
"count": 141
|
| 2929 |
},
|
| 2930 |
"MB": {
|
| 2931 |
+
"accuracy": 0.9305555555555556,
|
| 2932 |
+
"count": 72
|
| 2933 |
},
|
| 2934 |
"ME": {
|
| 2935 |
"accuracy": 1.0,
|
| 2936 |
+
"count": 18
|
| 2937 |
},
|
| 2938 |
"UB": {
|
| 2939 |
+
"accuracy": 0.8571428571428571,
|
| 2940 |
+
"count": 119
|
| 2941 |
}
|
| 2942 |
}
|
| 2943 |
},
|
| 2944 |
"sub_M2": {
|
| 2945 |
+
"full_accuracy": 0.14,
|
| 2946 |
+
"digit_accuracy": 0.8228571428571428,
|
| 2947 |
+
"n_examples": 50,
|
| 2948 |
"per_subtask": {
|
| 2949 |
"MD": {
|
| 2950 |
+
"accuracy": 0.9553571428571429,
|
| 2951 |
+
"count": 112
|
| 2952 |
},
|
| 2953 |
"MB": {
|
| 2954 |
+
"accuracy": 0.8679245283018868,
|
| 2955 |
+
"count": 53
|
| 2956 |
},
|
| 2957 |
"ME": {
|
| 2958 |
+
"accuracy": 0.9787234042553191,
|
| 2959 |
+
"count": 47
|
| 2960 |
},
|
| 2961 |
"UB": {
|
| 2962 |
+
"accuracy": 0.5529411764705883,
|
| 2963 |
+
"count": 85
|
| 2964 |
},
|
| 2965 |
"UD": {
|
| 2966 |
+
"accuracy": 0.7924528301886793,
|
| 2967 |
+
"count": 53
|
| 2968 |
}
|
| 2969 |
}
|
| 2970 |
},
|
| 2971 |
"sub_M3": {
|
| 2972 |
+
"full_accuracy": 0.06,
|
| 2973 |
+
"digit_accuracy": 0.7228571428571429,
|
| 2974 |
+
"n_examples": 50,
|
| 2975 |
"per_subtask": {
|
| 2976 |
"MD": {
|
| 2977 |
+
"accuracy": 0.9690721649484536,
|
| 2978 |
+
"count": 97
|
| 2979 |
},
|
| 2980 |
"MB": {
|
| 2981 |
+
"accuracy": 0.9803921568627451,
|
| 2982 |
+
"count": 51
|
| 2983 |
},
|
| 2984 |
"ME": {
|
| 2985 |
+
"accuracy": 0.9629629629629629,
|
| 2986 |
+
"count": 27
|
| 2987 |
},
|
| 2988 |
"UB": {
|
| 2989 |
+
"accuracy": 0.5405405405405406,
|
| 2990 |
+
"count": 74
|
| 2991 |
},
|
| 2992 |
"UD": {
|
| 2993 |
+
"accuracy": 0.42574257425742573,
|
| 2994 |
+
"count": 101
|
| 2995 |
}
|
| 2996 |
}
|
| 2997 |
},
|
| 2998 |
"sub_M4": {
|
| 2999 |
+
"full_accuracy": 0.06,
|
| 3000 |
+
"digit_accuracy": 0.6228571428571429,
|
| 3001 |
+
"n_examples": 50,
|
| 3002 |
"per_subtask": {
|
| 3003 |
"MD": {
|
| 3004 |
"accuracy": 0.98,
|
| 3005 |
+
"count": 100
|
| 3006 |
},
|
| 3007 |
"MB": {
|
| 3008 |
+
"accuracy": 0.98,
|
| 3009 |
+
"count": 50
|
| 3010 |
},
|
| 3011 |
"UB": {
|
| 3012 |
+
"accuracy": 0.54,
|
| 3013 |
+
"count": 50
|
| 3014 |
},
|
| 3015 |
"UD": {
|
| 3016 |
"accuracy": 0.29333333333333333,
|
| 3017 |
+
"count": 150
|
| 3018 |
}
|
| 3019 |
}
|
| 3020 |
},
|
| 3021 |
"sub_M5": {
|
| 3022 |
+
"full_accuracy": 0.06,
|
| 3023 |
+
"digit_accuracy": 0.5114285714285715,
|
| 3024 |
+
"n_examples": 50,
|
| 3025 |
"per_subtask": {
|
| 3026 |
"MD": {
|
| 3027 |
"accuracy": 1.0,
|
| 3028 |
+
"count": 50
|
| 3029 |
},
|
| 3030 |
"MB": {
|
| 3031 |
"accuracy": 1.0,
|
| 3032 |
+
"count": 50
|
| 3033 |
},
|
| 3034 |
"UB": {
|
| 3035 |
+
"accuracy": 0.38,
|
| 3036 |
+
"count": 50
|
| 3037 |
},
|
| 3038 |
"UD": {
|
| 3039 |
+
"accuracy": 0.3,
|
| 3040 |
+
"count": 200
|
| 3041 |
}
|
| 3042 |
}
|
| 3043 |
},
|
| 3044 |
"sub_random": {
|
| 3045 |
+
"full_accuracy": 0.525,
|
| 3046 |
+
"digit_accuracy": 0.9178571428571428,
|
| 3047 |
"n_examples": 200,
|
| 3048 |
"per_subtask": {
|
| 3049 |
"MD": {
|
| 3050 |
+
"accuracy": 0.9736842105263158,
|
| 3051 |
+
"count": 570
|
| 3052 |
},
|
| 3053 |
"MB": {
|
| 3054 |
+
"accuracy": 0.9675090252707581,
|
| 3055 |
+
"count": 277
|
| 3056 |
},
|
| 3057 |
"ME": {
|
| 3058 |
+
"accuracy": 1.0,
|
| 3059 |
"count": 53
|
| 3060 |
},
|
| 3061 |
"UB": {
|
| 3062 |
+
"accuracy": 0.8131634819532909,
|
| 3063 |
+
"count": 471
|
| 3064 |
},
|
| 3065 |
"UD": {
|
| 3066 |
+
"accuracy": 0.896551724137931,
|
| 3067 |
+
"count": 29
|
| 3068 |
}
|
| 3069 |
}
|
| 3070 |
},
|
| 3071 |
"sub_B3": {
|
| 3072 |
+
"full_accuracy": 0.26,
|
| 3073 |
+
"digit_accuracy": 0.8285714285714286,
|
| 3074 |
+
"n_examples": 50,
|
| 3075 |
"per_subtask": {
|
| 3076 |
"MD": {
|
| 3077 |
+
"accuracy": 0.9866666666666667,
|
| 3078 |
+
"count": 150
|
| 3079 |
},
|
| 3080 |
"MB": {
|
| 3081 |
"accuracy": 0.96,
|
| 3082 |
+
"count": 50
|
| 3083 |
},
|
| 3084 |
"UB": {
|
| 3085 |
+
"accuracy": 0.6237623762376238,
|
| 3086 |
+
"count": 101
|
| 3087 |
},
|
| 3088 |
"UD": {
|
| 3089 |
+
"accuracy": 0.6326530612244898,
|
| 3090 |
+
"count": 49
|
| 3091 |
}
|
| 3092 |
}
|
| 3093 |
},
|
| 3094 |
"sub_B4": {
|
| 3095 |
+
"full_accuracy": 0.1,
|
| 3096 |
+
"digit_accuracy": 0.7371428571428571,
|
| 3097 |
+
"n_examples": 50,
|
| 3098 |
"per_subtask": {
|
| 3099 |
"MD": {
|
| 3100 |
+
"accuracy": 0.98,
|
| 3101 |
+
"count": 100
|
| 3102 |
},
|
| 3103 |
"MB": {
|
| 3104 |
+
"accuracy": 0.94,
|
| 3105 |
+
"count": 50
|
| 3106 |
},
|
| 3107 |
"UB": {
|
| 3108 |
+
"accuracy": 0.5867768595041323,
|
| 3109 |
+
"count": 121
|
| 3110 |
},
|
| 3111 |
"UD": {
|
| 3112 |
+
"accuracy": 0.5316455696202531,
|
| 3113 |
+
"count": 79
|
| 3114 |
}
|
| 3115 |
}
|
| 3116 |
},
|
| 3117 |
"sub_B5": {
|
| 3118 |
+
"full_accuracy": 0.16,
|
| 3119 |
+
"digit_accuracy": 0.7028571428571428,
|
| 3120 |
+
"n_examples": 50,
|
| 3121 |
"per_subtask": {
|
| 3122 |
"MD": {
|
| 3123 |
"accuracy": 1.0,
|
| 3124 |
+
"count": 50
|
| 3125 |
},
|
| 3126 |
"MB": {
|
| 3127 |
"accuracy": 1.0,
|
| 3128 |
+
"count": 50
|
| 3129 |
},
|
| 3130 |
"UB": {
|
| 3131 |
+
"accuracy": 0.5986842105263158,
|
| 3132 |
+
"count": 152
|
| 3133 |
},
|
| 3134 |
"UD": {
|
| 3135 |
+
"accuracy": 0.5612244897959183,
|
| 3136 |
+
"count": 98
|
| 3137 |
}
|
| 3138 |
}
|
| 3139 |
}
|
| 3140 |
},
|
| 3141 |
"summary": {
|
| 3142 |
+
"overall_accuracy": 0.4026666666666667,
|
| 3143 |
+
"digit_accuracy": 0.82,
|
| 3144 |
+
"total_examples": 1500,
|
| 3145 |
+
"n_splits": 24
|
| 3146 |
}
|
| 3147 |
}
|
| 3148 |
}
|