app.py

import gradio as gr
import pandas as pd
import numpy as np
from transformers import pipeline, BertTokenizer, BertModel
import faiss
import torch
import json
import spaces
import logging

# Set up logging
logging.basicConfig(level=logging.DEBUG)

# Load CSV data
data = pd.read_csv('RBDx10kstats.csv')

# Function to safely convert JSON strings to numpy arrays
def safe_json_loads(x):
    try:
        return np.array(json.loads(x), dtype=np.float16)  # Ensure the array is of type float32
    except json.JSONDecodeError as e:
        logging.error(f"Error decoding JSON: {e}")
        return np.array([], dtype=np.float16)  # Return an empty array or handle it as appropriate

# Apply the safe_json_loads function to the embedding column
data['embedding'] = data['embedding'].apply(safe_json_loads)

# Filter out any rows with empty embeddings
data = data[data['embedding'].apply(lambda x: x.size > 0)]

# Initialize FAISS index
dimension = len(data['embedding'].iloc[0])
res = faiss.StandardGpuResources()  # use a single GPU

# Create FAISS index
if faiss.get_num_gpus() > 0:
    gpu_index = faiss.IndexFlatL2(dimension)
    gpu_index = faiss.index_cpu_to_gpu(res, 0, gpu_index)  # move to GPU
else:
    gpu_index = faiss.IndexFlatL2(dimension)  # fall back to CPU

# Ensure embeddings are stacked as float32
embeddings = np.vstack(data['embedding'].values).astype(np.float16)
logging.debug(f"Embeddings shape: {embeddings.shape}, dtype: {embeddings.dtype}")
gpu_index.add(embeddings)

# Check if GPU is available
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

# Load QA model
qa_model = pipeline("question-answering", model="distilbert-base-uncased-distilled-squad", device=0 if torch.cuda.is_available() else -1)

# Load BERT model and tokenizer
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
model = BertModel.from_pretrained('bert-base-uncased').to(device)

# Function to embed the question using BERT
def embed_question(question, model, tokenizer):
    try:
        inputs = tokenizer(question, return_tensors='pt').to(device)
        logging.debug(f"Tokenized inputs: {inputs}")
        with torch.no_grad():
            outputs = model(**inputs)
        embedding = outputs.last_hidden_state.mean(dim=1).cpu().numpy().astype(np.float16)
        logging.debug(f"Question embedding shape: {embedding.shape}")
        logging.debug(f"Question embedding content: {embedding}")
        return embedding
    except Exception as e:
        logging.error(f"Error embedding question: {e}")
        raise

# Function to retrieve the relevant document and generate a response
@spaces.GPU(duration=120)
def retrieve_and_generate(question):
    logging.debug(f"Received question: {question}")
    try:
        # Embed the question
        question_embedding = embed_question(question, model, tokenizer)

        # Ensure the embedding is in the correct format for FAISS search
        question_embedding = question_embedding.astype(np.float16)

        # Search in FAISS index
        try:
            logging.debug(f"Searching FAISS index with question embedding: {question_embedding}")
            _, indices = gpu_index.search(question_embedding, k=1)
            if indices.size == 0:
                logging.error("No results found in FAISS search.")
                return "No relevant document found."
            logging.debug(f"Indices found: {indices}")
        except Exception as e:
            logging.error(f"Error during FAISS search: {e}")
            return f"An error occurred during search: {e}"

        # Retrieve the most relevant document
        try:
            relevant_doc = data.iloc[indices[0][0]]
            logging.debug(f"Relevant document: {relevant_doc}")
        except Exception as e:
            logging.error(f"Error retrieving document: {e}")
            return "An error occurred while retrieving the document. Please try again."

        # Use the QA model to generate the answer
        try:
            context = relevant_doc['Abstract']
            response = qa_model(question=question, context=context)
            logging.debug(f"Response: {response}")
            return response['answer']
        except Exception as e:
            logging.error(f"Error generating answer: {e}")
            return "An error occurred while generating the answer. Please try again."
    except Exception as e:
        logging.error(f"Error during retrieval and generation: {e}")
        return "An error occurred. Please try again."

# Create a Gradio interface
interface = gr.Interface(
    fn=retrieve_and_generate,
    inputs=gr.Textbox(lines=2, placeholder="Ask a question about the documents..."),
    outputs="text",
    title="RAG Chatbot",
    description="Ask questions about the documents in the CSV file."
)

# Launch the Gradio app
interface.launch()