Horizon-Labs's picture
Multilingual educational-quality classifier base v1.0
4c78335 verified
Raw History Blame Contribute Delete
505 Bytes
{
"train": 266042,
"val": 13920,
"removed_eval_overlap": 38,
"by_source": {
"llama3": 149997,
"fineweb2": 129965
},
"n_lang": 66,
"calibration": {
"n": 10000,
"spearman_teacher_llama": 0.7143050988982202,
"map": {
"0": 0.0,
"0.5": 1.086,
"1": 1.351,
"1.5": 1.965,
"2": 2.123,
"2.5": 2.29,
"3": 2.604,
"3.5": 2.953,
"4": 3.143,
"4.5": 4.0
}
},
"mean_edu": {
"fineweb2": 1.5,
"llama3": 1.295
},
"frac_ge3": {
"fineweb2": 0.12,
"llama3": 0.09
}
}