Upload 2 files

app.py

@@ -0,0 +1,42 @@
+import streamlit as st
+import torch
+from transformers import BertConfig, BertForSequenceClassification, BertTokenizer
+import numpy as np
+
+
+# Load the model and tokenizer
+def load_model():
+    tokenizer = BertTokenizer.from_pretrained('beomi/kcbert-base')
+    config = BertConfig.from_pretrained('beomi/kcbert-base', num_labels=7)
+    model = BertForSequenceClassification.from_pretrained('beomi/kcbert-base', config=config)
+    model_state_dict = torch.load('sentiment7_model_acc8878.pth', map_location=torch.device('cpu')) # cpu 사용
+    model.load_state_dict(model_state_dict)
+    model.eval()
+    return model, tokenizer
+
+model, tokenizer = load_model()
+
+# Define the inference function
+def inference(input_doc):
+    inputs = tokenizer(input_doc, return_tensors='pt')
+    outputs = model(**inputs)
+    probs = torch.softmax(outputs.logits, dim=1).squeeze().tolist()
+    class_idx = {'공포': 0, '놀람': 1, '분노': 2, '슬픔': 3, '중립': 4, '행복': 5, '혐오': 6}
+    results = {class_name: prob for class_name, prob in zip(class_idx, probs)}
+    # Find the class with the highest probability
+    max_prob_class = max(results, key=results.get)
+    max_prob = results[max_prob_class]
+    # Display results
+    return [results, f"가장 강하게 나타난 감정: {max_prob_class}"]
+    '''    for class_name, prob in results.items():
+        print(f"{class_name}: {prob:.2%}")'''
+
+# Set up the Streamlit interface
+st.title('감정분석(Sentiment Analysis)')
+st.markdown('<small style="color:grey;">글에 나타난 공포, 놀람, 분노, 슬픔, 중립, 행복, 혐오의 정도를 비율로 알려드립니다.</small>', unsafe_allow_html=True)
+user_input = st.text_area("이 곳에 글 입력(100자 이하 권장):")
+if st.button('시작'):
+    result = inference(user_input)
+    st.write(result[0])
+    st.write(result[1])
+

requirements.txt

@@ -0,0 +1,4 @@
+streamlit
+torch
+transformers
+numpy