app.py

from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel, Field
from contextlib import asynccontextmanager
import re
import os
import logging

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

try:
    from llama_cpp import Llama
except ImportError:
    raise ImportError("Install llama-cpp-python: pip install llama-cpp-python")

MODEL_REPO = "bartowski/Phi-3.5-mini-instruct-GGUF"
MODEL_FILE = "Phi-3.5-mini-instruct-Q4_K_M.gguf"

llm = None
model_loading = False

@asynccontextmanager
async def lifespan(app: FastAPI):
    global llm, model_loading
    try:
        logger.info("🚀 Starting model load...")
        model_loading = True
        
        # Set cache directory for Hugging Face Spaces
        cache_dir = os.getenv("HF_HOME", "./models")
        
        llm = Llama.from_pretrained(
            repo_id=MODEL_REPO,
            filename=MODEL_FILE,
            n_threads=4,
            n_ctx=2048,
            n_batch=256,
            n_gpu_layers=0,
            verbose=False,
        )
        model_loading = False
        logger.info("✅ Model loaded and ready")
    except Exception as e:
        logger.error(f"❌ Model load error: {e}")
        model_loading = False
        llm = None
    
    yield
    
    logger.info("🛑 Shutting down...")
    if llm:
        del llm

app = FastAPI(
    title="AI Summarizer",
    description="Fast & Accurate AI Text Summarizer",
    version="1.0",
    lifespan=lifespan
)

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["*"],
    allow_headers=["*"],
)

class SummarizeRequest(BaseModel):
    text: str = Field(..., min_length=1, max_length=2000)
    length: str = Field(default="short", pattern="^(short|medium|long)$")

LENGTH_INSTRUCTIONS = {
    "short": "Summarize in 2–3 concise sentences.",
    "medium": "Summarize in 4–5 clear sentences.",
    "long": "Summarize in a detailed paragraph.",
}

def clean_output(text: str) -> str:
    """Clean model output from special tokens"""
    text = re.sub(r"<\|.*?\|>", "", text)
    text = re.sub(r"\s+", " ", text)
    return text.strip()

@app.get("/")
def root():
    """Root endpoint - returns status"""
    return {
        "status": "healthy",
        "model_loaded": llm is not None,
        "model_loading": model_loading,
        "message": "AI Summarizer API is running"
    }

@app.get("/health")
def health():
    """Health check endpoint for container orchestration"""
    if model_loading:
        return {
            "status": "starting",
            "model_loaded": False,
            "model_loading": True,
            "message": "Model is loading, please wait..."
        }
    
    if llm is None:
        return {
            "status": "unhealthy",
            "model_loaded": False,
            "model_loading": False,
            "message": "Model failed to load"
        }
    
    return {
        "status": "healthy",
        "model_loaded": True,
        "model_loading": False,
        "model_name": MODEL_FILE,
        "message": "Ready to summarize"
    }

@app.get("/ready")
def readiness():
    """Readiness probe - returns 200 only when model is loaded"""
    if llm is not None and not model_loading:
        return {"ready": True}
    raise HTTPException(status_code=503, detail="Model not ready")

@app.post("/api/summarize")
async def summarize(req: SummarizeRequest):
    if model_loading:
        raise HTTPException(
            status_code=503,
            detail="Model is still loading. Please wait and try again."
        )
    
    if llm is None:
        raise HTTPException(
            status_code=503, 
            detail="Model not loaded. Check server logs."
        )
    
    try:
        text = req.text.strip()
        length_instruction = LENGTH_INSTRUCTIONS.get(
            req.length, 
            LENGTH_INSTRUCTIONS["short"]
        )
        
        prompt = f"""<|user|>
You are an expert text summarizer.
{length_instruction}

Text:
{text}
<|end|>
<|assistant|>"""
        
        max_tokens_map = {
            "short": 140,
            "medium": 220,
            "long": 300
        }
        
        logger.info(f"Summarizing text (length: {req.length})")
        
        output = llm(
            prompt,
            max_tokens=max_tokens_map.get(req.length, 140),
            temperature=0.3,
            top_p=0.9,
            top_k=40,
            repeat_penalty=1.05,
            stop=["<|end|>", "<|user|>"],
            echo=False
        )
        
        summary = clean_output(output["choices"][0]["text"])
        
        if not summary:
            raise HTTPException(
                status_code=500,
                detail="Model produced empty output"
            )
        
        logger.info("✅ Summary generated successfully")
        
        return {
            "summary": summary,
            "success": True,
            "length": req.length
        }
    
    except HTTPException:
        raise
    except Exception as e:
        logger.error(f"Summarization error: {e}")
        raise HTTPException(
            status_code=500,
            detail=f"Summarization error: {str(e)}"
        )

if __name__ == "__main__":
    import uvicorn
    # Use PORT environment variable for Hugging Face Spaces
    port = int(os.getenv("PORT", 7860))
    uvicorn.run(app, host="0.0.0.0", port=port)