Korean Intention classifier with pytorch lightning ⚡
This is an intention classifier that was learned using 3i4k dataset to identify the intonation intention of Korean.
- implemented using phytorch lighting. ⚡
- trained based on Transformers' distil-KoBERT PLM and has approx. 90% accuracy.
- use it as a web API powerd by flask-restplus.
- clone
pip install -r requirements.txt
- download model file
- copy into
/model/intent_distilkobert.ckpt
bash train_distil.sh
to use distil-KoBERT model
bash train.sh
to use KoBERT model
python app.py
to start web server- open
{SERVER_IP}:41063
in web browser - using swagger UI to
try out
- response types
- FRAG : Fragments
- STAT : Statements
- QUES : Questions
- COMM : Commands
- RHEQ : rhetorical questions
- RHEC : rhetorical commands
- INTO : Intonation-dependent utterances