AksharaLP-MultiIndic-8M / models /ta /training_report.json
sourojitdotcom's picture
Private review: measured five-language release, Hindi sentence mode and custom GGUFs
a449848 verified
Raw History Blame Contribute Delete
635 Bytes
{
"pairs": 3452431,
"counts": {
"aks-Dakshina": 60210,
"aks-AK-Freq": 776,
"aks-Existing": 1545,
"aks-Wikidata": 273,
"aks-Samanantar": 34447,
"aks-IndicCorp": 3133649,
"ie": 204261,
"heldout_drop_ie": 1379,
"dup_ie": 90268,
"bad_roman_ie": 10,
"ie_words": 90000,
"dup_ei": 96489,
"ei": 17270,
"heldout_drop_ei": 133,
"ei_words": 60000
},
"unique_natives": 3200785,
"unique_romans": 3322500,
"overlap_valid_test_natives": 0,
"overlap_valid_test_romans": 0,
"heldout_natives_total": 10964,
"heldout_romans_total": 20104,
"margins": {
"ie": 0.6,
"ei": 0.35,
"maxv_ie": 4,
"maxv_ei": 2
}
}