app.py

import pandas as pd
import torch
from sentence_transformers import SentenceTransformer, util
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
import gradio as gr
import json
import faiss
import spaces

# Ensure you have GPU support
device = 'cuda' if torch.cuda.is_available() else 'cpu'

# Load the CSV file with embeddings
df = pd.read_csv('RBDx10kstats.csv')
df['embedding'] = df['embedding'].apply(json.loads)  # Convert JSON string back to list

# Convert embeddings to a numpy array
embeddings = np.array(df['embedding'].tolist(), dtype='float32')

# Setup FAISS
index = faiss.IndexFlatL2(embeddings.shape[1])  # dimension should match the embedding size
index.add(embeddings)

# Load the Sentence Transformer model
sentence_model = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2', device=device)

# Load the LLaMA model for response generation
llama_tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-chat-hf")
llama_model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-2-7b-chat-hf").to(device)

# Load the summarization model
summarizer = pipeline("summarization", model="facebook/bart-large-cnn", device=0 if device == 'cuda' else -1)

# Define the function to find the most relevant document using FAISS
@spaces.GPU(duration=120)
def retrieve_relevant_doc(query):
    query_embedding = sentence_model.encode(query, convert_to_tensor=False)
    _, indices = index.search(np.array([query_embedding]), k=1)
    best_match_idx = indices[0][0]
    return df.iloc[best_match_idx]['Abstract']

# Define the function to generate a response
@spaces.GPU(duration=120)
def generate_response(query):
    relevant_doc = retrieve_relevant_doc(query)
    if len(relevant_doc) > 512:  # Truncate long documents
        relevant_doc = summarizer(relevant_doc, max_length=150, min_length=50, do_sample=False)[0]['summary_text']
    input_text = f"Document: {relevant_doc}\n\nQuestion: {query}\n\nAnswer:"
    inputs = llama_tokenizer(input_text, return_tensors="pt").to(device)
    outputs = llama_model.generate(inputs["input_ids"], max_length=150)
    response = llama_tokenizer.decode(outputs[0], skip_special_tokens=True)
    return response

# Create a Gradio interface
iface = gr.Interface(
    fn=generate_response,
    inputs=gr.Textbox(lines=2, placeholder="Enter your query here..."),
    outputs="text",
    title="RAG Chatbot",
    description="This chatbot retrieves relevant documents based on your query and generates responses using LLaMA."
)

# Launch the Gradio interface
iface.launch()