import streamlit as st from transformers import BertForSequenceClassification, BertTokenizer import torch import time import random # [Previous CSS styles remain the same] def local_css(): st.markdown(""" """, unsafe_allow_html=True) @st.cache_resource def load_model(): model = BertForSequenceClassification.from_pretrained("trituenhantaoio/bert-base-vietnamese-uncased") tokenizer = BertTokenizer.from_pretrained("trituenhantaoio/bert-base-vietnamese-uncased") return model, tokenizer def predict(text, model, tokenizer): inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=512) with torch.no_grad(): outputs = model(**inputs) predictions = torch.nn.functional.softmax(outputs.logits, dim=-1) predicted_class = torch.argmax(predictions, dim=1).item() confidence = predictions[0][predicted_class].item() return predicted_class, confidence def get_bot_response(text, predicted_class, confidence): # Define response templates based on classes and confidence levels responses = { 0: { # Example for class 0 (positive sentiment) 'high_conf': [ "Tôi cảm nhận được sự tích cực trong câu nói của bạn. Xin chia sẻ thêm nhé!", "Thật vui khi nghe điều đó. Bạn có thể kể thêm không?", "Tuyệt vời! Tôi rất đồng ý với bạn về điều này." ], 'low_conf': [ "Có vẻ như đây là điều tích cực. Đúng không nhỉ?", "Tôi nghĩ đây là một góc nhìn thú vị đấy.", "Nghe có vẻ tốt đấy, bạn nghĩ sao?" ] }, 1: { # Example for class 1 (negative sentiment) 'high_conf': [ "Tôi hiểu đây là điều khó khăn với bạn. Hãy chia sẻ thêm nhé.", "Tôi rất tiếc khi nghe điều này. Bạn cần tôi giúp gì không?", "Đúng là một tình huống khó khăn. Chúng ta cùng tìm giải pháp nhé." ], 'low_conf': [ "Có vẻ như bạn đang gặp khó khăn. Tôi có hiểu đúng không?", "Tôi không chắc mình hiểu hết, bạn có thể giải thích thêm được không?", "Hãy chia sẻ thêm để tôi có thể hiểu rõ hơn nhé." ] } } # Add more classes based on your model's output # Determine confidence level confidence_threshold = 0.8 conf_level = 'high_conf' if confidence > confidence_threshold else 'low_conf' # Get appropriate response list try: response_list = responses[predicted_class][conf_level] response = random.choice(response_list) except KeyError: response = "Xin lỗi, tôi không chắc chắn về điều này. Bạn có thể giải thích rõ hơn được không?" # Add context from user's input context_response = f"{response}" return context_response def init_session_state(): if 'messages' not in st.session_state: st.session_state.messages = [] if 'thinking' not in st.session_state: st.session_state.thinking = False def display_chat_history(): for message in st.session_state.messages: if message['role'] == 'user': st.markdown(f'
{message["content"]}
', unsafe_allow_html=True) else: st.markdown(f'
{message["content"]}
', unsafe_allow_html=True) def main(): st.set_page_config(page_title="Vietnamese Chatbot", page_icon="🤖", layout="wide") local_css() init_session_state() # Load model model, tokenizer = load_model() # Chat interface st.title("Chatbot Tiếng Việt 🤖") st.markdown("Xin chào! Tôi có thể giúp gì cho bạn?") # Chat history container chat_container = st.container() # Input container with st.container(): col1, col2 = st.columns([6, 1]) with col1: user_input = st.text_input("Nhập tin nhắn của bạn...", key="user_input", label_visibility="hidden") with col2: send_button = st.button("Gửi") if user_input and send_button: # Add user message st.session_state.messages.append({"role": "user", "content": user_input}) # Show thinking animation st.session_state.thinking = True # Get prediction predicted_class, confidence = predict(user_input, model, tokenizer) # Generate response bot_response = get_bot_response(user_input, predicted_class, confidence) # Add bot response time.sleep(0.5) # Brief delay for natural feeling st.session_state.messages.append({"role": "assistant", "content": bot_response}) st.session_state.thinking = False # Clear input and rerun st.rerun() # Display chat history with chat_container: display_chat_history() if st.session_state.thinking: st.markdown("""
""", unsafe_allow_html=True) if __name__ == "__main__": main()