main.py

from fastapi import FastAPI, HTTPException, status
from pydantic import BaseModel, ConfigDict, Field
from typing import Optional, List
from ctransformers import AutoModelForCausalLM
import time
import logging
from app.config import MODEL_PATH, MODEL_URL

# Configure logging
logging.basicConfig(
    level=logging.INFO,
    format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
)
logger = logging.getLogger(__name__)

# Initialize FastAPI app
app = FastAPI(
    title="Poetry Generator API",
    description="An API for generating poetry using a local LLM",
    version="1.0.0"
)

# Global model variable
model = None

class PoetryRequest(BaseModel):
    prompt: str = Field(..., description="The topic or theme for the poem", min_length=1)
    style: str = Field(
        default="free verse",
        description="Style of the poem to generate"
    )
    max_length: int = Field(
        default=200,
        description="Maximum length of the generated poem",
        ge=50,
        le=500
    )
    temperature: float = Field(
        default=0.7,
        description="Temperature for text generation",
        ge=0.1,
        le=2.0
    )

class PoetryResponse(BaseModel):
    poem: str
    generation_time: float
    prompt: str
    style: str

class ModelInfo(BaseModel):
    model_config = ConfigDict(protected_namespaces=())
    
    status: str
    model_path: str
    model_name: str
    supported_styles: List[str]
    max_context_length: int

def initialize_model():
    """Initialize the model and return it"""
    if not MODEL_PATH.exists():
        logger.error(f"Model not found at {MODEL_PATH}")
        return None
        
    try:
        logger.info(f"Loading model from {MODEL_PATH}")
        return AutoModelForCausalLM.from_pretrained(
            str(MODEL_PATH.parent),
            model_file=MODEL_PATH.name,
            model_type="llama",
            max_new_tokens=512,
            context_length=512,
            gpu_layers=0  # CPU only
        )
    except Exception as e:
        logger.error(f"Error loading model: {str(e)}")
        return None

@app.on_event("startup")
async def startup_event():
    """Initialize the model during startup"""
    global model
    model = initialize_model()
    if model is None:
        logger.warning("Model failed to load but service will start anyway")

@app.get(
    "/health",
    response_model=ModelInfo,
    status_code=status.HTTP_200_OK,
    tags=["Health Check"]
)
async def health_check():
    """Check if the model is loaded and get basic information"""
    model_status = "ready" if model is not None else "not_loaded"
    
    return ModelInfo(
        status=model_status,
        model_name="Llama-2-7B-Chat",
        model_path=str(MODEL_PATH),
        supported_styles=[
            "free verse",
            "haiku",
            "sonnet",
            "limerick",
            "tanka"
        ],
        max_context_length=512
    )

@app.post(
    "/generate",
    response_model=PoetryResponse,
    status_code=status.HTTP_200_OK,
    tags=["Generation"]
)
async def generate_poem(request: PoetryRequest):
    """Generate a poem based on the provided prompt and parameters"""
    if model is None:
        raise HTTPException(
            status_code=status.HTTP_503_SERVICE_UNAVAILABLE,
            detail="Model not loaded. Please check /health endpoint for status."
        )

    try:
        start_time = time.time()
        
        prompt_templates = {
            "haiku": "Write a haiku about {prompt}. Follow the 5-7-5 syllable pattern:\n\n",
            "sonnet": "Write a Shakespearean sonnet about {prompt}. Follow the traditional 14-line format with rhyme scheme ABAB CDCD EFEF GG:\n\n",
            "limerick": "Write a limerick about {prompt}. Follow the AABBA rhyme scheme:\n\n",
            "free verse": "Write a free verse poem about {prompt}. Make it creative and meaningful:\n\n",
            "tanka": "Write a tanka about {prompt}. Follow the 5-7-5-7-7 syllable pattern:\n\n"
        }
        
        template = prompt_templates.get(request.style.lower(), prompt_templates["free verse"])
        full_prompt = template.format(prompt=request.prompt)

        output = model(
            full_prompt,
            max_new_tokens=request.max_length,
            temperature=request.temperature,
            top_p=0.95,
            repeat_penalty=1.2
        )

        generation_time = time.time() - start_time
        
        return PoetryResponse(
            poem=output.strip(),
            generation_time=generation_time,
            prompt=request.prompt,
            style=request.style
        )

    except Exception as e:
        logger.error(f"Generation error: {str(e)}")
        raise HTTPException(
            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
            detail=f"Failed to generate poem: {str(e)}"
        )
    
def download_model():
    """Download the model if it doesn't exist"""
    import requests
    from tqdm import tqdm
    
    if MODEL_PATH.exists():
        logger.info(f"Model already exists at {MODEL_PATH}")
        return
    
    logger.info(f"Downloading model to {MODEL_PATH}")
    try:
        response = requests.get(MODEL_URL, stream=True)
        response.raise_for_status()
        total_size = int(response.headers.get('content-length', 0))
        
        with open(MODEL_PATH, 'wb') as file, tqdm(
            desc="Downloading",
            total=total_size,
            unit='iB',
            unit_scale=True,
            unit_divisor=1024,
        ) as pbar:
            for data in response.iter_content(chunk_size=1024):
                size = file.write(data)
                pbar.update(size)
                
        logger.info("Model downloaded successfully")
    except Exception as e:
        logger.error(f"Error downloading model: {str(e)}")
        if MODEL_PATH.exists():
            MODEL_PATH.unlink()
        raise

if __name__ == "__main__":
    import uvicorn
    uvicorn.run("app.main:app", host="0.0.0.0", port=8000, reload=True)