80 lines
2.5 KiB
JSON
80 lines
2.5 KiB
JSON
{
|
|
"count": 100,
|
|
"solved": 0,
|
|
"solved_rate": 0.0,
|
|
"solved_at_n_count": 0,
|
|
"solved_at_n_rate": 0.0,
|
|
"pass_at_1_count": 0,
|
|
"pass_at_1_rate": 0.0,
|
|
"candidate_count_total": 100,
|
|
"candidate_count_mean_per_problem": 1,
|
|
"num_samples_max": 1,
|
|
"num_samples_min": 1,
|
|
"reason_counts": {
|
|
"value is 36, expected 24": 7,
|
|
"value is 20, expected 24": 6,
|
|
"value is 16, expected 24": 6,
|
|
"value is 60, expected 24": 6,
|
|
"value is 40, expected 24": 5,
|
|
"value is 30, expected 24": 5,
|
|
"value is 72, expected 24": 5,
|
|
"value is 48, expected 24": 5,
|
|
"value is 32, expected 24": 4,
|
|
"value is 12, expected 24": 4,
|
|
"value is 84, expected 24": 4,
|
|
"value is 42, expected 24": 3,
|
|
"value is 18, expected 24": 3,
|
|
"value is 6, expected 24": 3,
|
|
"value is 10, expected 24": 2,
|
|
"value is 21, expected 24": 2,
|
|
"value is 0, expected 24": 2,
|
|
"value is 96, expected 24": 2,
|
|
"value is 33, expected 24": 2,
|
|
"value is 9, expected 24": 2,
|
|
"value is 90, expected 24": 2,
|
|
"value is 12.5, expected 24": 1,
|
|
"value is 9.33333333333, expected 24": 1,
|
|
"value is 50, expected 24": 1,
|
|
"value is 56, expected 24": 1,
|
|
"value is -16, expected 24": 1,
|
|
"value is 39, expected 24": 1,
|
|
"value is 6.4, expected 24": 1,
|
|
"value is 91, expected 24": 1,
|
|
"value is 19, expected 24": 1,
|
|
"value is -12, expected 24": 1,
|
|
"value is 15, expected 24": 1,
|
|
"value is 26, expected 24": 1,
|
|
"value is 25, expected 24": 1,
|
|
"value is 8, expected 24": 1,
|
|
"value is 33.3333333333, expected 24": 1,
|
|
"value is 10.5, expected 24": 1,
|
|
"value is 14, expected 24": 1,
|
|
"value is 16.6666666667, expected 24": 1,
|
|
"value is 66, expected 24": 1,
|
|
"numbers are not used exactly once": 1
|
|
},
|
|
"empty": false,
|
|
"first_attempt_count": 100,
|
|
"first_attempt_r1_format_rate": 1.0,
|
|
"first_attempt_answer_tag_rate": 1.0,
|
|
"first_attempt_allowed_chars_rate": 1.0,
|
|
"first_attempt_parse_ok_rate": 1.0,
|
|
"first_attempt_numbers_ok_rate": 0.99,
|
|
"first_attempt_target_ok_rate": 0.01,
|
|
"first_attempt_valid_rate": 0.0,
|
|
"attempt_count": 100,
|
|
"attempt_r1_format_rate": 1.0,
|
|
"attempt_answer_tag_rate": 1.0,
|
|
"attempt_allowed_chars_rate": 1.0,
|
|
"attempt_parse_ok_rate": 1.0,
|
|
"attempt_numbers_ok_rate": 0.99,
|
|
"attempt_target_ok_rate": 0.01,
|
|
"attempt_valid_rate": 0.0,
|
|
"mean_candidates_until_solution": null,
|
|
"median_candidates_until_solution": null,
|
|
"max_candidates_until_solution": null,
|
|
"solved_at_1_count": 0,
|
|
"solved_at_1": 0.0,
|
|
"num_samples": 1
|
|
}
|