klue_nli_roberta_base_model

This model is a fine-tuned version of klue/roberta-base on the klue dataset. It achieves the following results on the evaluation set:

• Loss: 0.6867
• Accuracy: 0.8653

Model description

Pretrained RoBERTa Model on Korean Language. See Github and Paper for more details.

Intended uses & limitations

How to use

NOTE: Use BertTokenizer instead of RobertaTokenizer. (AutoTokenizer will load BertTokenizer)

from transformers import AutoModel, AutoTokenizer

model = AutoModel.from_pretrained("klue/roberta-base")
tokenizer = AutoTokenizer.from_pretrained("klue/roberta-base")

Training and evaluation data

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.5988	1.0	782	0.4378	0.8363
0.2753	2.0	1564	0.4169	0.851
0.1735	3.0	2346	0.5267	0.8607
0.0956	4.0	3128	0.6275	0.8683
0.0708	5.0	3910	0.6867	0.8653

Framework versions

• Transformers 4.30.2
• Pytorch 2.0.1+cu118
• Datasets 2.13.1
• Tokenizers 0.13.3