app.py

from functools import lru_cache
import os
from flask import Flask, request, jsonify, render_template
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
import logging
import time

app = Flask(__name__)
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Önbellek dizini ayarı
os.environ['TRANSFORMERS_CACHE'] = '/app/cache'
os.makedirs('/app/cache', exist_ok=True)

# Model konfigürasyonu
MODEL_NAME = "redrussianarmy/gpt2-turkish-cased"

@lru_cache(maxsize=1)
def load_model():
    try:
        logger.info("Model yükleniyor...")
        tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
        
        # Pad token kontrolü ve ayarlama
        if tokenizer.pad_token is None:
            tokenizer.pad_token = tokenizer.eos_token
        
        # CPU için float32 kullan
        model = AutoModelForCausalLM.from_pretrained(MODEL_NAME)
        model = model.to('cpu').float()  # Float32 formatına dönüştür
        torch.set_num_threads(1)
        logger.info("Model başarıyla yüklendi")
        return model, tokenizer
    except Exception as e:
        logger.error(f"Model yükleme hatası: {str(e)}")
        raise RuntimeError(f"Model yüklenemedi: {str(e)}")

@app.route('/')
def home():
    return render_template('index.html')

@app.route('/health')
def health_check():
    try:
        load_model()
        return jsonify({"status": "healthy"}), 200
    except Exception as e:
        return jsonify({"status": "unhealthy", "error": str(e)}), 500

@app.route('/generate', methods=['POST'])
def generate():
    try:
        start_time = time.time()
        data = request.get_json()
        prompt = data.get('prompt', '')[:300]  # 300 karakter sınır
        
        if not prompt:
            return jsonify({"error": "Prompt gereklidir", "success": False}), 400

        model, tokenizer = load_model()
        inputs = tokenizer(prompt, return_tensors="pt", padding=True, truncation=True).to('cpu')
        
        with torch.no_grad():
            outputs = model.generate(
                inputs.input_ids,
                attention_mask=inputs.attention_mask,
                max_length=80,
                do_sample=True,
                top_k=40,
                temperature=0.7,
                pad_token_id=tokenizer.pad_token_id,
                num_return_sequences=1,
                early_stopping=True,
                use_cache=True
            )
        
        result = tokenizer.decode(outputs[0], skip_special_tokens=True)
        processing_time = round(time.time() - start_time, 2)
        
        return jsonify({
            "result": result,
            "success": True,
            "processing_time": processing_time
        })
        
    except Exception as e:
        logger.error(f"Hata: {str(e)}")
        return jsonify({
            "error": str(e),
            "success": False
        }), 500

if __name__ == '__main__':
    app.run(host='0.0.0.0', port=7860, threaded=False)