Practical overview
When it fits
Use it when you need repository code for Korean question-answering experiments on KorQuAD, including a documented training workflow with development-set evaluation or a separate evaluation script.
Who it is for
- Developers integrating KoBERT with Hugging Face Transformers.
- Researchers and practitioners fine-tuning or evaluating Korean extractive question-answering models.
Documented uses
- Fine-tune KoBERT or DistilKoBERT on KorQuAD data.
- Evaluate predictions with exact-match and F1 scoring.
- Load KoBERT and its tokenizer in a Transformers-based Python application.
Requirements & limits
The repository is public and identified as open source under the Apache-2.0 license.
- The README specifies torch>=1.1.0, transformers==2.9.1, and tensorboardX>=2.0.
- The documented training example uses KorQuAD v1.0 JSON files and a maximum sequence length of 512.
- The repository reports KoBERT and DistilKoBERT development-set scores separately; its results are configuration-specific.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Fine-tune KoBERT with KorQuAD data
Run the documented training workflow with development-set evaluation during training.
Before you start
- Python 3
- torch>=1.1.0
- transformers==2.9.1
- tensorboardX>=2.0
- KorQuAD v1.0 training and development JSON files
Steps
Place the KorQuAD training and development files in the data directory.
Run run_squad.py with model_type kobert and model_name_or_path monologg/kobert.
Set the output directory, data directory, training and prediction filenames, and the documented training and evaluation arguments.
Expected outcome
The training workflow fine-tunes KoBERT and reports development-set official exact-match and F1 scores during training.
Guide limitations
- The README's example uses Transformers 2.9.1 and a maximum sequence length of 512.
- The reported scores are specific to the README's development-set configuration.
Evaluate predictions on KorQuAD v1.0 development data
Use the repository's evaluation script with a development JSON file and an existing predictions JSON file.
Before you start
- Python 3
- KorQuAD v1.0 development JSON file
- Predictions JSON file
- Repository evaluation script evaluate_v1_0.py
Steps
Set the path to KorQuAD_v1.0_dev.json.
Set the path to the predictions JSON file.
Run evaluate_v1_0.py with the development-data path followed by the predictions path.
Expected outcome
The script evaluates predictions against the KorQuAD v1.0 development set.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- monologg/KoBERT-KorQuAD · Official homepage title · retained in its source language