import os
import sys
import time
import threading
import traceback
from datetime import datetime
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from transformers import pipeline
from unsloth import FastLanguageModel

# === Ortam değişkenleri
os.environ.setdefault("HF_HOME", "/app/.cache")
os.environ.setdefault("HF_HUB_CACHE", "/app/.cache")
os.environ.setdefault("BITSANDBYTES_NOWELCOME", "1")

# === Basit log
def log(message):
    timestamp = datetime.now().strftime("%H:%M:%S")
    print(f"[{timestamp}] {message}", flush=True)

# === FastAPI başlat
app = FastAPI()
pipe = None
model = None
tokenizer = None

@app.on_event("startup")
def load_model():
    global pipe, model, tokenizer
    try:
        model_name = "atasoglu/Turkish-Llama-3-8B-function-calling"
        log(f"⬇️ [1] Model yükleme başlatılıyor: {model_name}")

        model, tokenizer = FastLanguageModel.from_pretrained(
            model_name=model_name,
            load_in_4bit=True,
            device_map="auto"
        )
        log("✅ [2] Model ve tokenizer çekildi.")

        FastLanguageModel.for_inference(model)
        log("✅ [3] Model inference moduna alındı.")

        pipe = pipeline(
            "text-generation",
            model=model,
            tokenizer=tokenizer,
            device_map="auto"
        )
        log("✅ [4] Pipeline başarıyla kuruldu, test etmeye hazır.")

    except Exception as e:
        log(f"❌ [ERROR] Model yükleme sırasında hata: {e}")
        traceback.print_exc()
        raise

class TestRequest(BaseModel):
    user_input: str

@app.post("/test")
def test(req: TestRequest):
    try:
        prompt = f"Kullanıcı: {req.user_input}\nAsistan:"
        log(f"💬 [5] Prompt alındı: {req.user_input}")

        inputs = tokenizer([prompt], return_tensors="pt")
        log("🧠 [6] Tokenizer çıktılarını hazırladı, generate başlıyor...")

        outputs = model.generate(
            **inputs,
            max_new_tokens=256,
            temperature=0.2,
            top_p=0.95,
            repetition_penalty=1.1,
            do_sample=True
        )
        log("✅ [7] Generate tamamlandı, cevap dönülüyor.")

        answer = tokenizer.decode(outputs[0], skip_special_tokens=True)
        answer_clean = answer.split("Asistan:")[-1].strip()
        return {"response": answer_clean}

    except Exception as e:
        log(f"❌ [ERROR] /test sırasında hata: {e}")
        traceback.print_exc()
        raise HTTPException(status_code=500, detail=str(e))

@app.get("/")
def health():
    return {"status": "ok"}

def run_health_server():
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=7860)

threading.Thread(target=run_health_server, daemon=True).start()

log("⏸️ [0] Uygulama bekleme modunda, startup bekleniyor...")
while True:
    time.sleep(60)