#!/usr/bin/env python3
"""
Working Gemma 3n GGUF Backend Service
Minimal FastAPI backend using only llama-cpp-python for GGUF models
"""

import os
import logging
import time
from contextlib import asynccontextmanager
from typing import List, Dict, Any, Optional

from fastapi import FastAPI, HTTPException
from fastapi.responses import JSONResponse
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel, Field, field_validator

# Import llama-cpp-python for GGUF model support
try:
    from llama_cpp import Llama
    llama_cpp_available = True
except ImportError:
    llama_cpp_available = False

import uvicorn

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

# Pydantic models for OpenAI-compatible API
class ChatMessage(BaseModel):
    role: str = Field(..., description="The role of the message author")
    content: str = Field(..., description="The content of the message")
    
    @field_validator('role')
    @classmethod
    def validate_role(cls, v: str) -> str:
        if v not in ["system", "user", "assistant"]:
            raise ValueError("Role must be one of: system, user, assistant")
        return v

class ChatCompletionRequest(BaseModel):
    model: str = Field(default="gemma-3n-e4b-it", description="The model to use for completion")
    messages: List[ChatMessage] = Field(..., description="List of messages in the conversation")
    max_tokens: Optional[int] = Field(default=512, ge=1, le=2048, description="Maximum tokens to generate")
    temperature: Optional[float] = Field(default=1.0, ge=0.0, le=2.0, description="Sampling temperature")
    top_p: Optional[float] = Field(default=0.95, ge=0.0, le=1.0, description="Top-p sampling")
    top_k: Optional[int] = Field(default=64, ge=1, le=100, description="Top-k sampling")
    stream: Optional[bool] = Field(default=False, description="Whether to stream responses")

class ChatCompletionChoice(BaseModel):
    index: int
    message: ChatMessage
    finish_reason: str

class ChatCompletionResponse(BaseModel):
    id: str
    object: str = "chat.completion"
    created: int
    model: str
    choices: List[ChatCompletionChoice]

class HealthResponse(BaseModel):
    status: str
    model: str
    version: str
    backend: str

# Global variables for model management
current_model = os.environ.get("AI_MODEL", "unsloth/gemma-3n-E4B-it-GGUF")
llm = None

def convert_messages_to_gemma_prompt(messages: List[ChatMessage]) -> str:
    """Convert OpenAI messages format to Gemma 3n chat format."""
    # Gemma 3n uses specific format with <start_of_turn> and <end_of_turn>
    prompt_parts = ["<bos>"]
    
    for message in messages:
        role = message.role
        content = message.content
        
        if role == "system":
            prompt_parts.append(f"<start_of_turn>system\n{content}<end_of_turn>")
        elif role == "user":
            prompt_parts.append(f"<start_of_turn>user\n{content}<end_of_turn>")
        elif role == "assistant":
            prompt_parts.append(f"<start_of_turn>model\n{content}<end_of_turn>")
    
    # Add the start for model response
    prompt_parts.append("<start_of_turn>model\n")
    
    return "\n".join(prompt_parts)

@asynccontextmanager
async def lifespan(app: FastAPI):
    """Application lifespan manager for startup and shutdown events"""
    global llm
    logger.info("🚀 Starting Gemma 3n GGUF Backend Service...")
    
    if not llama_cpp_available:
        logger.error("❌ llama-cpp-python is not available. Please install with: pip install llama-cpp-python")
        raise RuntimeError("llama-cpp-python not available")
    
    try:
        logger.info(f"📥 Loading Gemma 3n GGUF model from {current_model}...")
        
        # Configure model parameters for Gemma 3n
        llm = Llama.from_pretrained(
            repo_id=current_model,
            filename="*Q4_K_M.gguf",  # Use Q4_K_M quantization for good performance
            verbose=True,
            # Gemma 3n specific settings
            n_ctx=4096,  # Start with 4K context (can be increased to 32K)
            n_threads=4,  # Adjust based on your CPU
            n_gpu_layers=-1,  # Use all GPU layers if CUDA available, otherwise CPU
            # Chat template for Gemma 3n format
            chat_format="gemma",  # Try built-in gemma format first
        )
        
        logger.info("✅ Successfully loaded Gemma 3n GGUF model")
        
    except Exception as e:
        logger.error(f"❌ Failed to initialize Gemma 3n model: {e}")
        logger.warning("⚠️ Please download the GGUF model file locally and update the path")
        logger.warning("⚠️ You can download from: https://huggingface.co/unsloth/gemma-3n-E4B-it-GGUF")
        
        # For demo purposes, we'll continue without the model
        logger.info("🔄 Starting service in demo mode (responses will be mocked)")
    
    yield
    logger.info("🔄 Shutting down Gemma 3n Backend Service...")
    if llm:
        # Clean up model resources
        llm = None

# Initialize FastAPI app
app = FastAPI(
    title="Gemma 3n GGUF Backend Service",
    description="OpenAI-compatible chat completion API powered by Gemma-3n-E4B-it-GGUF",
    version="1.0.0",
    lifespan=lifespan
)

# Add CORS middleware
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],  # Configure appropriately for production
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

def ensure_model_ready():
    """Check if model is loaded and ready"""
    # For demo mode, we'll allow the service to run even without a model
    pass

def generate_response_gguf(messages: List[ChatMessage], max_tokens: int = 512, temperature: float = 1.0, top_p: float = 0.95, top_k: int = 64) -> str:
    """Generate response using GGUF model via llama-cpp-python."""
    if llm is None:
        # Demo mode response
        return "🤖 Demo mode: Gemma 3n model not loaded. This would be a real response from the Gemma 3n model. Please download the GGUF model from https://huggingface.co/unsloth/gemma-3n-E4B-it-GGUF"
    
    try:
        # Use the chat completion method if available
        if hasattr(llm, 'create_chat_completion'):
            # Convert to dict format for llama-cpp-python
            messages_dict = [{"role": msg.role, "content": msg.content} for msg in messages]
            
            response = llm.create_chat_completion(
                messages=messages_dict,
                max_tokens=max_tokens,
                temperature=temperature,
                top_p=top_p,
                top_k=top_k,
                stop=["<end_of_turn>", "<eos>", "</s>"]  # Gemma 3n stop tokens
            )
            
            return response['choices'][0]['message']['content'].strip()
        
        else:
            # Fallback to direct prompt completion
            prompt = convert_messages_to_gemma_prompt(messages)
            
            response = llm(
                prompt,
                max_tokens=max_tokens,
                temperature=temperature,
                top_p=top_p,
                top_k=top_k,
                stop=["<end_of_turn>", "<eos>", "</s>"],
                echo=False
            )
            
            return response['choices'][0]['text'].strip()
            
    except Exception as e:
        logger.error(f"GGUF generation failed: {e}")
        return "I apologize, but I'm having trouble generating a response right now. Please try again."

@app.get("/", response_class=JSONResponse)
async def root() -> Dict[str, Any]:
    """Root endpoint with service information"""
    return {
        "message": "Gemma 3n GGUF Backend Service is running!",
        "model": current_model,
        "version": "1.0.0",
        "backend": "llama-cpp-python",
        "model_loaded": llm is not None,
        "endpoints": {
            "health": "/health",
            "chat_completions": "/v1/chat/completions"
        }
    }

@app.get("/health", response_model=HealthResponse)
async def health_check():
    """Health check endpoint"""
    return HealthResponse(
        status="healthy" if (llm is not None) else "demo_mode",
        model=current_model,
        version="1.0.0",
        backend="llama-cpp-python"
    )

@app.post("/v1/chat/completions", response_model=ChatCompletionResponse)
async def create_chat_completion(
    request: ChatCompletionRequest
) -> ChatCompletionResponse:
    """Create a chat completion (OpenAI-compatible) using Gemma 3n GGUF"""
    try:
        ensure_model_ready()
        
        if not request.messages:
            raise HTTPException(status_code=400, detail="Messages cannot be empty")
        
        logger.info(f"Generating Gemma 3n response for {len(request.messages)} messages")
        
        response_text = generate_response_gguf(
            request.messages,
            request.max_tokens or 512,
            request.temperature or 1.0,
            request.top_p or 0.95,
            request.top_k or 64
        )
        
        response_text = response_text.strip() if response_text else "No response generated."
        
        return ChatCompletionResponse(
            id=f"chatcmpl-{int(time.time())}",
            created=int(time.time()),
            model=request.model,
            choices=[ChatCompletionChoice(
                index=0,
                message=ChatMessage(role="assistant", content=response_text),
                finish_reason="stop"
            )]
        )
        
    except Exception as e:
        logger.error(f"Error in chat completion: {e}")
        raise HTTPException(status_code=500, detail=f"Internal server error: {str(e)}")

# Main entry point
if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=8000)