File size: 1,087 Bytes
0df2042 6732015 0df2042 6732015 0df2042 6732015 0df2042 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 | {
"base_model": "KPF/KPF-bert-ner",
"training_dataset": "townboy/korean-pii-dataset",
"training_dataset_rows": 11732,
"splits": {
"train": 9227,
"validation": 1510,
"test": 995
},
"pii_types": 33,
"bio_labels": 67,
"maximum_sequence_length": 512,
"final_refinement": {
"effective_batch_size": 32,
"class_weighted_loss": false,
"stages": [
{
"purpose": "Korean-ID counterfactual learning",
"epochs": 1,
"learning_rate": 5e-06
},
{
"purpose": "general-label stabilization",
"epochs": 1,
"learning_rate": 1e-06
}
],
"validation_micro_f1": 0.9964634318856981
},
"final_evaluation": {
"documents": 1320,
"positive_spans_per_label": 100,
"exact_character_span_matching": true,
"macro_f1": 0.883369786048345,
"micro_f1": 0.8642728407042839
},
"notes": "The final refinement used standard cross-entropy. The trained parameters are stored in model.safetensors; class weighting is a loss setting, not a separate required inference weight file."
}
|