app.py

import gradio as gr
import random as rand
import torch  
from huggingface_hub import InferenceClient
from sentence_transformers import SentenceTransformer


custom_css = """
@import url('https://fonts.googleapis.com/css2?family=Fredoka:wght@300..700&display=swap'); 

body, input, textarea, button, select, label, p, span, div, h1, h2, h3, h4, h5, h6 {
    font-family: 'Fredoka', sans-serif !important;
}
"""

with open("knowledgebase.txt", "r", encoding="utf-8") as file:
    knowledge_base = file.read()

def preprocess_text(text):
    cleaned_text = text.strip()
    chunks = cleaned_text.split("\n")
    cleaned_chunks = []
    for chunk in chunks:
        stripped_chunk = chunk.strip()
        if len(stripped_chunk) > 0:
            cleaned_chunks.append(stripped_chunk)
    return cleaned_chunks

cleaned_chunks = preprocess_text(knowledge_base)

model = SentenceTransformer('all-MiniLM-L6-v2')

def create_embeddings(text_chunks):
    chunk_embeddings = model.encode(text_chunks, convert_to_tensor=True)
    print("Embeddings Shape:", chunk_embeddings.shape)
    return chunk_embeddings

chunk_embeddings = create_embeddings(cleaned_chunks)

def get_top_chunks(query, chunk_embeddings, text_chunks):
    query_embedding = model.encode(query, convert_to_tensor=True)
    query_embedding_normalized = query_embedding / query_embedding.norm()
    chunk_embeddings_normalized = chunk_embeddings / chunk_embeddings.norm(dim=1, keepdim=True)

    similarities = torch.matmul(chunk_embeddings_normalized, query_embedding_normalized)
    top_indices = torch.topk(similarities, k=3).indices
    
    top_chunks = []
    for i in top_indices:
        chunk = text_chunks[i]
        top_chunks.append(chunk)
    return top_chunks


client = InferenceClient("Qwen/Qwen2.5-7B-Instruct") 

def respond(message, history):
    relevant_context = get_top_chunks(message, chunk_embeddings, cleaned_chunks)
    context_str = "\n".join(relevant_context)
    
    system_prompt = f"You are a helpful assistant. Use the following context to answer the user's question accurately:\n\n{context_str}"
    
    messages = [{"role": "system", "content": system_prompt}]
    
    if history:
        messages.extend(history)
        
    messages.append({"role": "user", "content": message})
    
    response = ""
    for msg in client.chat_completion(messages, max_tokens=200, temperature=1, top_p=0.5, stream=True):
        token = msg.choices[0].delta.content
        if token:
            response += token
            yield response


with gr.Blocks(theme='d8ahazard/material_design_rd', css=custom_css) as demo:
    gr.Markdown("# Digital Wellbeing")
    
    chatbot = gr.Chatbot(label="Chat History")
    msg = gr.Textbox(placeholder="Type your question here...")
    
    with gr.Row():
        submit_btn = gr.Button("Submit", variant="primary")
        clear_btn = gr.ClearButton([msg, chatbot], value="Clear Conversation")

    submit_event = submit_btn.click(respond, inputs=[msg, chatbot], outputs=[chatbot])
    msg_submit_event = msg.submit(respond, inputs=[msg, chatbot], outputs=[chatbot])
    
    submit_event.then(lambda: "", None, [msg])
    msg_submit_event.then(lambda: "", None, [msg])

if __name__ == "__main__":
    demo.launch()