app.py

import os
import re
import logging
import textwrap
import autopep8
import gradio as gr
from huggingface_hub import hf_hub_download
from llama_cpp import Llama
import jwt
from typing import Dict, Any

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

# JWT settings
JWT_SECRET = os.environ.get("JWT_SECRET")
if not JWT_SECRET:
    raise ValueError("JWT_SECRET environment variable is not set")
JWT_ALGORITHM = "HS256"

# Model settings
MODEL_NAME = "leetmonkey_peft__q8_0.gguf"
REPO_ID = "sugiv/leetmonkey-peft-gguf"

# Generation parameters
generation_kwargs = {
    "max_tokens": 2048,
    "stop": ["```", "### Instruction:", "### Response:"],
    "echo": False,
    "temperature": 0.2,
    "top_k": 50,
    "top_p": 0.95,
    "repeat_penalty": 1.1
}

def download_model(model_name: str) -> str:
    logger.info(f"Downloading model: {model_name}")
    model_path = hf_hub_download(
        repo_id=REPO_ID,
        filename=model_name,
        cache_dir="./models",
        force_download=True,
        resume_download=True
    )
    logger.info(f"Model downloaded: {model_path}")
    return model_path

# Download and load the 8-bit model at startup
model_path = download_model(MODEL_NAME)
llm = Llama(
    model_path=model_path,
    n_ctx=2048,
    n_threads=4,
    n_gpu_layers=-1,  # Use all available GPU layers
    verbose=False
)
logger.info("8-bit model loaded successfully")

def generate_solution(instruction: str) -> str:
    system_prompt = "You are a Python coding assistant specialized in solving LeetCode problems. Provide only the complete implementation of the given function. Ensure proper indentation and formatting. Do not include any explanations or multiple solutions."
    full_prompt = f"""### Instruction:
{system_prompt}

Implement the following function for the LeetCode problem:

{instruction}

### Response:
Here's the complete Python function implementation:

```python
"""
    
    response = llm(full_prompt, **generation_kwargs)
    return response["choices"][0]["text"]

def extract_and_format_code(text: str) -> str:
    code_match = re.search(r'```python\s*(.*?)\s*```', text, re.DOTALL)
    if code_match:
        code = code_match.group(1)
    else:
        code = text

    code = re.sub(r'^.*?(?=def\s+\w+\s*\()', '', code, flags=re.DOTALL)
    code = textwrap.dedent(code)
    lines = code.split('\n')
    func_def_index = next((i for i, line in enumerate(lines) if line.strip().startswith('def ')), 0)
    indented_lines = [lines[func_def_index]]
    for line in lines[func_def_index + 1:]:
        if line.strip():
            indented_lines.append('    ' + line)
        else:
            indented_lines.append(line)

    formatted_code = '\n'.join(indented_lines)

    try:
        return autopep8.fix_code(formatted_code)
    except:
        return formatted_code

def verify_token(token: str) -> bool:
    try:
        jwt.decode(token, JWT_SECRET, algorithms=[JWT_ALGORITHM])
        return True
    except jwt.PyJWTError:
        return False

def generate_code(instruction: str, token: str) -> Dict[str, Any]:
    if not verify_token(token):
        return {"error": "Invalid token"}
    
    logger.info("Generating solution")
    generated_output = generate_solution(instruction)
    formatted_code = extract_and_format_code(generated_output)
    logger.info("Solution generated successfully")
    return {"solution": formatted_code}

# Gradio API
api = gr.Interface(
    fn=generate_code,
    inputs=[
        gr.Textbox(label="LeetCode Problem Instruction"),
        gr.Textbox(label="JWT Token")
    ],
    outputs=gr.JSON(),
    title="LeetCode Problem Solver API",
    description="Provide a LeetCode problem instruction and a valid JWT token to generate a solution.",
    examples=[
        ["Implement a function to reverse a linked list", "your_jwt_token_here"],
        ["Write a function to find the maximum subarray sum", "your_jwt_token_here"]
    ]
)

if __name__ == "__main__":
    api.launch(server_name="0.0.0.0", server_port=7860)