app.py

import gradio as gr
from pinecone import Pinecone
import torch
from pinecone_text.sparse import SpladeEncoder
from sentence_transformers import SentenceTransformer
from transformers import AutoTokenizer, AutoModelForCausalLM
import os

import requests
import os
from tqdm import tqdm


def download_model(url, model_path):
    response = requests.get(url, stream=True)
    total_size = int(response.headers.get('content-length', 0))
    block_size = 1024  # 1 KB

    with open(model_path, 'wb') as file, tqdm(
        desc=model_path,
        total=total_size,
        unit='iB',
        unit_scale=True,
        unit_divisor=1024,
    ) as progress_bar:
        for data in response.iter_content(block_size):
            size = file.write(data)
            progress_bar.update(size)


# Initialize Pinecone
PINECONE_API_KEY = os.environ.get('PINECONE_API_KEY')
pc = Pinecone(api_key=PINECONE_API_KEY)
index_name = "leetmonkey-sparse-dense"
index = pc.Index(index_name)

# Initialize models
device = 'cpu'
splade = SpladeEncoder(device=device)
dense_model = SentenceTransformer('sentence-transformers/all-Mpnet-base-v2', device=device)

from llama_cpp import Llama

# Define the model URL and path
model_url = "https://huggingface.co/TheBloke/Llama-2-7B-Chat-GGUF/resolve/main/llama-2-7b-chat.Q4_K_M.gguf"
model_path = "/tmp/llama-2-7b-chat.Q4_K_M.gguf"

# Download the model if it doesn't exist
if not os.path.exists(model_path):
    print(f"Downloading model to {model_path}...")
    download_model(model_url, model_path)
    print("Model downloaded successfully.")

# Initialize the Llama model
llm = Llama(model_path=model_path, n_ctx=1024, n_threads=8, n_batch=512,
    mlock=True, n_gpu_layers=-1)


def search_problems(query, top_k=5):
    dense_query = dense_model.encode([query])[0].tolist()
    sparse_query = splade.encode_documents([query])[0]

    results = index.query(
        vector=dense_query,
        sparse_vector={
            'indices': sparse_query['indices'],
            'values': sparse_query['values']
        },
        top_k=top_k,
        include_metadata=True,
        namespace='leetcode-problems'
    )

    return results['matches']

def generate_few_shot_prompt(search_results):
    prompt = "Here are some example LeetCode problems:\n\n"
    for result in search_results:
        metadata = result['metadata']
        prompt += f"Title: {metadata['title']}\n"
        prompt += f"Topics: {', '.join(metadata['topicTags'])}\n"
        prompt += f"Difficulty: {metadata['difficulty']}\n\n"
    return prompt

def generate_response(user_query, top_k=5):
    search_results = search_problems(user_query, top_k)
    few_shot_prompt = generate_few_shot_prompt(search_results)

    system_prompt = """You are an AI assistant specialized in providing information about LeetCode problems. 
    Your task is to recommend relevant problems based on the user's query and the provided examples. 
    Focus on problem titles, difficulty levels, topic tags, and companies that have asked these problems. 
    Do not provide specific problem solutions or content."""

    user_prompt = f"Based on the following query, recommend relevant LeetCode problems:\n{user_query}"
    full_prompt = f"{system_prompt}\n\n{few_shot_prompt}\n{user_prompt}\n\nRecommendations:"

     # Generate response using Llama model
    response = llm(full_prompt, max_tokens=150, temperature=0.7, top_p=0.9)

    # Extract the generated recommendations
    recommendations = response['choices'][0]['text'].strip()
    return recommendations

# Create Gradio interface
with gr.Blocks() as iface:
    gr.Markdown("# LeetCode Problem Assistant")
    
    disclaimer = gr.Textbox(
        value="• This is purely for Educational purpose only. We don't claim ownership of Problem statement\n"
              "• Copyright of problems and contents goes to Leetcode. Leetcode was not scraped as per terms.\n"
              "• Do not rely solely on this tool for interview preparation.\n"
              "• Always verify information with official LeetCode resources.\n"
              "• You can ask => 'I have upcoming interview with Google, I like to get better with Dynamic Programming. What do you recommend?'.\n"
              "• Your missing natural language based hybrid search on sparse and dense vectors built on Leetcode Knowledge Graph.",
        label="Disclaimer",
        interactive=False
    )
    
    query_input = gr.Textbox(
        lines=2, 
        placeholder="Enter your natural language query about LeetCode problems...",
        label="Your Query"
    )
    
    output = gr.Textbox(label="Search Results")
    
    submit_button = gr.Button("Search")
    
    submit_button.click(
        fn=generate_response,
        inputs=query_input,
        outputs=output
    )

# Launch the app
iface.launch(share=True)