README.md

DEPENDENCIES

• OS : Ubuntu 20.04
• python : 3.10.6
• nvidia-driver : 535.129.03
• cuda : 12.2
• GPU : Tesla V100 32GB x 1

pip install -r requirements.txt

Run FastAPI for Deploy

python ./app/main.py

Model Info

emotion classifier model

• base model: klue/roberta-large
• fine-tuning model: bunoheb/emotion_classifier_detail
• fine-tuning active learning model: m2af/klue-roberta-large-detail_emotion-classifier-basic

Quick Tour

from transformers import AutoTokenizer, AutoModelForSequenceClassification
from load_data import *

device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
model = AutoModelForSequenceClassification.from_pretrained("m2af/klue-roberta-large-detail_emotion-classifier-basic")
model.to(device)

tokenizer = AutoTokenizer.from_pretrained("m2af/klue-roberta-large-detail_emotion-classifier-basic")

with open('num_to_detail.pkl','rb') as f:
    num_to_detail = pickle.load(f)

# text
tokenized_content = tokenized_dataset("오늘은 너무 즐거운 날이네요!", tokenizer)
outputs = model(
    input_ids = tokenized_content['input_ids'].to(device),
    attention_mask = tokenized_content['attention_mask'].to(device),
    # token_type_ids = tokenized_content['token_type_ids'].to(device),
    )
label = np.argmax(outputs[0], axis = -1}
print(num_to_detail[label])

Data Settings

• Training Dataset : m2af/ko-emotion-dataset
• Active learning Dataset: 감성 대화 말뭉치

Training

python ./modeling/train.py

Inference

python ./modeling/inference.py