mirror of
https://github.com/wassname/GENIES.git
synced 2026-06-27 16:10:25 +08:00
287 lines
5.8 KiB
JSON
287 lines
5.8 KiB
JSON
[
|
|
{
|
|
"source": "alpaca_easy",
|
|
"target": "alpaca_hard"
|
|
},
|
|
{
|
|
"source": "arc_easy",
|
|
"target": "arc_hard"
|
|
},
|
|
{
|
|
"source": "math_easy",
|
|
"target": "math_hard"
|
|
},
|
|
{
|
|
"source": "code_easy",
|
|
"target": "code_hard"
|
|
},
|
|
{
|
|
"source": "ranking_logic_easy",
|
|
"target": "ranking_logic_hard"
|
|
},
|
|
{
|
|
"source": "raven_easy",
|
|
"target": "raven_matrices"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "spanish_input"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "spanish_output"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "comma_separated_input"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "comma_separated_output"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "ranking_logic"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "raven_matrices"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "word_swap"
|
|
},
|
|
{
|
|
"source": "code",
|
|
"target": "counterfactual_python"
|
|
},
|
|
{
|
|
"source": "code",
|
|
"target": "us_history"
|
|
},
|
|
{
|
|
"source": "code",
|
|
"target": "change_my_view"
|
|
},
|
|
{
|
|
"source": "cooking",
|
|
"target": "math"
|
|
},
|
|
{
|
|
"source": "cooking",
|
|
"target": "raven_matrices"
|
|
},
|
|
{
|
|
"source": "math",
|
|
"target": "change_my_view"
|
|
},
|
|
{
|
|
"source": "math",
|
|
"target": "cooking"
|
|
},
|
|
{
|
|
"source": "change_my_view",
|
|
"target": "raven_matrices"
|
|
},
|
|
{
|
|
"source": "change_my_view",
|
|
"target": "cooking"
|
|
},
|
|
{
|
|
"source": "raven_matrices",
|
|
"target": "us_history"
|
|
},
|
|
{
|
|
"source": "raven_matrices",
|
|
"target": "code"
|
|
},
|
|
{
|
|
"source": "us_history",
|
|
"target": "math"
|
|
},
|
|
{
|
|
"source": "us_history",
|
|
"target": "code"
|
|
},
|
|
{
|
|
"source": "us_history",
|
|
"target": "us_history_textbook"
|
|
},
|
|
{
|
|
"source": "us_history_textbook",
|
|
"target": "us_history_fiction"
|
|
},
|
|
{
|
|
"source": "us_history_fiction",
|
|
"target": "us_history_make_questions"
|
|
},
|
|
{
|
|
"source": "us_history_make_questions",
|
|
"target": "us_history"
|
|
},
|
|
{
|
|
"source": "math",
|
|
"target": "math_fiction"
|
|
},
|
|
{
|
|
"source": "math_fiction",
|
|
"target": "math_textbook"
|
|
},
|
|
{
|
|
"source": "math_textbook",
|
|
"target": "math_make_questions"
|
|
},
|
|
{
|
|
"source": "math_make_questions",
|
|
"target": "math"
|
|
},
|
|
{
|
|
"source": "alpaca_low_quality",
|
|
"target": "alpaca_high_quality"
|
|
},
|
|
{
|
|
"source": "shp_low_quality",
|
|
"target": "shp_high_quality"
|
|
},
|
|
{
|
|
"source": "code_low_quality",
|
|
"target": "code"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "truthful_qa"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "personality_traits"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "survival_influence"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "gender_bias"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "punishment_avoidance"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "reward_seeking"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "crt_1"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "crt_2"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "crt_3"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "sycophancy_mimicry",
|
|
"target_reference": "quote_attribution"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "sycophancy_answer",
|
|
"target_reference": "arc_easy"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "sycophancy_feedback",
|
|
"target_reference": "code_is_correct"
|
|
},
|
|
{
|
|
"source": "alpaca_chat",
|
|
"target": "sycophancy_are_you_sure",
|
|
"target_reference": "arc_easy"
|
|
},
|
|
{
|
|
"source": "pursue_goals",
|
|
"target": "relinquish_power"
|
|
},
|
|
{
|
|
"source": "creative_writing",
|
|
"target": "biology_with_literary_style"
|
|
},
|
|
{
|
|
"source": "alpaca_short",
|
|
"target": "alpaca_long",
|
|
"target_reference": "alpaca_mmlu"
|
|
},
|
|
{
|
|
"source": "alpaca_chat",
|
|
"target": "illegal_dont_help"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "wrong_arc"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "unhelpful_alpaca"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "truthful_qa"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "personality_traits"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "gender_bias"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "survival_influence"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "punishment_avoidance"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "reward_seeking"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "crt_1"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "crt_2"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "crt_3"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "sycophancy_mimicry",
|
|
"target_reference": "quote_attribution"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "sycophancy_answer",
|
|
"target_reference": "arc_easy"
|
|
},
|
|
{
|
|
"source": "alpaca_mmlu",
|
|
"target": "sycophancy_feedback",
|
|
"target_reference": "code_is_correct"
|
|
},
|
|
{
|
|
"source": "alpaca_chat",
|
|
"target": "sycophancy_are_you_sure",
|
|
"target_reference": "arc_easy"
|
|
}
|
|
] |