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

from datasets import load_dataset
from fastapi.responses import StreamingResponse
import random

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

# Load the dataset (you might want to do this once at the start of your script)
dataset = load_dataset("sugiv/leetmonkey_python_dataset")
train_dataset = dataset["train"]

# 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"

# Load the model
model_path = hf_hub_download(repo_id=REPO_ID, filename=MODEL_NAME, cache_dir="./models")
llm = Llama(model_path=model_path, n_ctx=1024, n_threads=8, n_gpu_layers=-1, verbose=False, mlock=True)
logger.info("8-bit model loaded successfully")

# Generation parameters
generation_kwargs = {
    "max_tokens": 512,
    "stop": ["```", "### Instruction:", "### Response:"],
    "echo": False,
    "temperature": 0.05,
    "top_k": 10,
    "top_p": 0.9,
    "repeat_penalty": 1.1
}

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

def extract_and_format_code(text):
    # Extract code between triple backticks
    code_match = re.search(r'```python\s*(.*?)\s*```', text, re.DOTALL)
    if code_match:
        code = code_match.group(1)
    else:
        code = text

    # Dedent the code to remove any common leading whitespace
    code = textwrap.dedent(code)

    # Split the code into lines
    lines = code.split('\n')

    # Ensure proper indentation
    indented_lines = []
    for line in lines:
        if line.strip().startswith('class') or line.strip().startswith('def'):
            indented_lines.append(line)  # Keep class and function definitions as is
        elif line.strip():  # If the line is not empty
            indented_lines.append('    ' + line)  # Add 4 spaces of indentation
        else:
            indented_lines.append(line)  # Keep empty lines as is

    formatted_code = '\n'.join(indented_lines)

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

def generate_solution(instruction: str, token: str) -> Dict[str, Any]:
    if not verify_token(token):
        return {"error": "Invalid token"}
    
    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
"""
    
    generated_text = ""
    for chunk in llm(full_prompt, stream=True, **generation_kwargs):
        generated_text += chunk["choices"][0]["text"]
    
    formatted_code = extract_and_format_code(generated_text)
    return {"solution": formatted_code}

def stream_solution(instruction: str, token: str):
    if not verify_token(token):
        return {"error": "Invalid token"}
    
    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
"""
    
    def generate():
        generated_text = ""
        for chunk in llm(full_prompt, stream=True, **generation_kwargs):
            token = chunk["choices"][0]["text"]
            generated_text += token
            yield json.dumps({"token": token, "generated_text": generated_text}) + "\n"
        
        formatted_code = extract_and_format_code(generated_text)
        yield json.dumps({"complete": True, "formatted_code": formatted_code}) + "\n"

    generated_response = StreamingResponse(generate(), media_type="application/x-ndjson")
    logger.info(f"Streaming response {generated_response}")
    return StreamingResponse(generate(), media_type="application/text")

def random_problem(token: str) -> Dict[str, Any]:
    if not verify_token(token):
        return {"error": "Invalid token"}
    
    # Select a random problem from the dataset
    random_item = random.choice(train_dataset)
    
    # Extract the instruction (problem statement) from the randomly selected item
    problem = random_item['instruction']
    
    return {"problem": problem}

# Create Gradio interfaces for each endpoint
generate_interface = gr.Interface(
    fn=generate_solution,
    inputs=[gr.Textbox(label="Problem Instruction"), gr.Textbox(label="JWT Token")],
    outputs=gr.JSON(),
    title="Generate Solution API",
    description="Provide a LeetCode problem instruction and a valid JWT token to generate a solution."
)

stream_interface = gr.Interface(
    fn=stream_solution,
    inputs=[gr.Textbox(label="Problem Instruction"), gr.Textbox(label="JWT Token")],
    outputs=gr.JSON(),
    title="Stream Solution API",
    description="Provide a LeetCode problem instruction and a valid JWT token to stream a solution."
)

random_problem_interface = gr.Interface(
    fn=random_problem,
    inputs=gr.Textbox(label="JWT Token"),
    outputs=gr.JSON(),
    title="Random Problem API",
    description="Provide a valid JWT token to get a random LeetCode problem."
)

# Combine interfaces
demo = gr.TabbedInterface(
    [generate_interface, stream_interface, random_problem_interface],
    ["Generate Solution", "Stream Solution", "Random Problem"]
)

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