metadata

library_name: transformers
base_model: klue/roberta-small
tags:
  - generated_from_trainer
metrics:
  - accuracy
  - f1
  - precision
  - recall
model-index:
  - name: roberta-small-hangul-2-hanja
    results: []

roberta-small-hangul-2-hanja

This model is a fine-tuned version of klue/roberta-small on the None dataset. It achieves the following results on the evaluation set:

Accuracy: 0.9918
F1: 0.9824
Loss: 0.0868
Precision: 0.9814
Recall: 0.9835

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 64
eval_batch_size: 64
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 20

Training results

Training Loss	Epoch	Step	Accuracy	F1	Validation Loss	Precision	Recall
No log	1.0	482	0.8993	0.4607	0.8157	0.6392	0.3602
1.7229	2.0	964	0.9606	0.8537	0.4586	0.8728	0.8354
0.6279	3.0	1446	0.9701	0.9159	0.3291	0.9244	0.9075
0.418	4.0	1928	0.9761	0.9426	0.2596	0.9439	0.9413
0.316	5.0	2410	0.9797	0.9530	0.2159	0.9545	0.9515
0.2553	6.0	2892	0.9821	0.9589	0.1848	0.9608	0.9571
0.213	7.0	3374	0.9832	0.9613	0.1622	0.9616	0.9610
0.1819	8.0	3856	0.9858	0.9707	0.1430	0.9691	0.9722
0.16	9.0	4338	0.9871	0.9712	0.1307	0.9705	0.9719
0.1409	10.0	4820	0.9885	0.9749	0.1197	0.9734	0.9764
0.1295	11.0	5302	0.9893	0.9759	0.1120	0.9747	0.9771
0.1174	12.0	5784	0.9893	0.9763	0.1065	0.9744	0.9782
0.1085	13.0	6266	0.9896	0.9770	0.1005	0.9755	0.9785
0.1011	14.0	6748	0.9905	0.9794	0.0968	0.9786	0.9803
0.0954	15.0	7230	0.9910	0.9801	0.0941	0.9793	0.9810
0.0899	16.0	7712	0.9912	0.9807	0.0916	0.9796	0.9817
0.0866	17.0	8194	0.9917	0.9819	0.0893	0.9810	0.9828
0.0847	18.0	8676	0.9918	0.9824	0.0880	0.9814	0.9835
0.0814	19.0	9158	0.9918	0.9824	0.0870	0.9814	0.9835
0.0822	20.0	9640	0.9918	0.9824	0.0868	0.9814	0.9835

Framework versions

Transformers 4.45.2
Pytorch 2.4.1+cu121
Datasets 3.0.1
Tokenizers 0.20.1