app.py

from flask import Flask, request, jsonify
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

# Modell: Deutsch, öffentlich zugänglich
MODEL_NAME = "dbmdz/german-gpt2"

# Tokenizer & Modell laden
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForCausalLM.from_pretrained(MODEL_NAME)

app = Flask(__name__)

@app.route("/chat", methods=["POST"])
def chat():
    data = request.json
    user_input = data.get("message", "")

    # Eingabe tokenisieren
    input_ids = tokenizer.encode(user_input, return_tensors="pt")

    # Modell antworten lassen
    output_ids = model.generate(input_ids, max_new_tokens=100, do_sample=True, top_k=50)
    response = tokenizer.decode(output_ids[0], skip_special_tokens=True)

    # Antwort ohne ursprüngliche Eingabe zurückgeben
    answer_only = response[len(user_input):].strip()

    return jsonify({"response": answer_only})

if __name__ == "__main__":
    app.run(debug=True)