joonhyun-kim/roberta-base-klue-ynat-classification

Model Card for Model ID

Model Details

Model Description: This model is a fine-tuned version of klue/roberta-base on the KLUE YNAT (Korean News Article Topic Classification) dataset. It classifies Korean news headlines into one of 7 categories.

Base model: klue/roberta-base

Task: Text classification (News Category)

Language: Korean

Fine-tuned by: Joonhyun Kim

Framework: PyTorch + Hugging Face Transformers

Dataset: KLUE-YNAT

Accuracy: ~86% on validation set

License: Apache-2.0 (same as base model)

Model Sources [optional]

• Repository: (https://huggingface.co/joonhyun-kim/roberta-base-klue-ynat-classification)
• Base Model: (https://huggingface.co/klue/roberta-base)
• Dataset: (https://huggingface.co/datasets/klue)

Uses

✅ Direct Use

The model can be used for Korean news article classification, such as categorizing headlines into topics like IT/Science, Sports, Politics, Economy, Culture, etc.

⚙️ Example Code from transformers import pipeline

classifier = pipeline( "text-classification", model="joonhyun-kim/roberta-base-klue-ynat-classification" )

result = classifier("尹 대통령, 한미정상회담 참석 위해 출국") print(result)

[{'label': '정치', 'score': 0.95}]

Training Details

Setting Value Base model klue/roberta-base Dataset KLUE-YNAT Epochs 1 Batch size 8 Learning rate 5e-5 Optimizer AdamW Framework Hugging Face Transformers Train accuracy ~0.86 Eval loss 0.46

Evaluation

Metric Score Accuracy 0.863 Eval loss 0.464

Technical Specifications [optional]

Architecture: RoBERTa-base (12-layer, hidden size 768, 125M parameters)

Hardware used: NVIDIA GPU (Colab / local machine)

Framework: PyTorch 2.4 + Transformers 4.x

Tokenization: SentencePiece BPE (same as base KLUE model)

Citation [optional]

If you use this model, please cite:

@misc{kim2025roberta-klue-ynat, author = {Joonhyun Kim}, title = {RoBERTa-base fine-tuned on KLUE YNAT for Korean News Classification}, year = {2025}, howpublished = {\url{https://huggingface.co/joonhyun-kim/roberta-base-klue-ynat-classification}} }