app.py

import gradio as gr
import openai
from datasets import load_dataset
import logging
import time
from langchain_community.embeddings import HuggingFaceEmbeddings
import torch
import psutil
import GPUtil

# Set up logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Initialize OpenAI API key
openai.api_key = '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'

# Initialize with E5 embedding model
model_name = 'intfloat/e5-base-v2'
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
embedding_model = HuggingFaceEmbeddings(model_name=model_name)
embedding_model.client.to(device)

# Load datasets
datasets = {}
dataset_names = ['covidqa', 'hotpotqa', 'pubmedqa']  # Starting with key datasets

for name in dataset_names:
    datasets[name] = load_dataset("rungalileo/ragbench", name, split='train')
    logger.info(f"Loaded {name}")

def get_system_metrics():
    metrics = {
        'cpu_percent': psutil.cpu_percent(),
        'memory_percent': psutil.virtual_memory().percent,
        'gpu_util': GPUtil.getGPUs()[0].load * 100 if torch.cuda.is_available() else 0,
        'gpu_memory': GPUtil.getGPUs()[0].memoryUtil * 100 if torch.cuda.is_available() else 0
    }
    return metrics

def process_query(query, dataset_choice="all"):
    start_time = time.time()
    try:
        relevant_contexts = []
        search_datasets = [dataset_choice] if dataset_choice != "all" else datasets.keys()
        
        for dataset_name in search_datasets:
            if dataset_name in datasets:
                for doc in datasets[dataset_name]['documents']:
                    if any(keyword.lower() in doc.lower() for keyword in query.split()):
                        relevant_contexts.append((doc, dataset_name))
        
        context_info = f"From {relevant_contexts[0][1]}: {relevant_contexts[0][0]}" if relevant_contexts else "Searching across datasets..."
        
        response = openai.chat.completions.create(
            model="gpt-3.5-turbo",
            messages=[
                {"role": "system", "content": "You are a knowledgeable expert using E5 embeddings for precise information retrieval."},
                {"role": "user", "content": f"Context: {context_info}\nQuestion: {query}"}
            ],
            max_tokens=300,
            temperature=0.7,
        )
        
        # Get performance metrics
        metrics = get_system_metrics()
        metrics['processing_time'] = time.time() - start_time
        
        metrics_display = f"""
        Processing Time: {metrics['processing_time']:.2f}s
        CPU Usage: {metrics['cpu_percent']}%
        Memory Usage: {metrics['memory_percent']}%
        GPU Utilization: {metrics['gpu_util']:.1f}%
        GPU Memory: {metrics['gpu_memory']:.1f}%
        """
        
        return response.choices[0].message.content.strip(), metrics_display
        
    except Exception as e:
        return str(e), "Metrics collection in progress"

# Create Gradio interface
demo = gr.Interface(
    fn=process_query,
    inputs=[
        gr.Textbox(label="Question", placeholder="Ask your question here"),
        gr.Dropdown(
            choices=["all"] + dataset_names,
            label="Select Dataset",
            value="all"
        )
    ],
    outputs=[
        gr.Textbox(label="Response"),
        gr.Textbox(label="Performance Metrics")
    ],
    title="E5-Powered Knowledge Base",
    description="Search across RagBench datasets with performance monitoring"
)

if __name__ == "__main__":
    demo.queue()
    demo.launch(debug=True)