app.py

import gradio as gr
from transformers import GPT2LMHeadModel, GPT2Tokenizer
import torch

# Initialisierung des Modells und des Tokenizers
tokenizer = GPT2Tokenizer.from_pretrained("Loewolf/GPT_1")
model = GPT2LMHeadModel.from_pretrained("Loewolf/GPT_1")

def generate_text(prompt):
    input_ids = tokenizer.encode(prompt, return_tensors="pt")
    attention_mask = torch.ones(input_ids.shape, dtype=torch.bool)  # Erstelle eine Attention-Mask, die überall '1' ist

    max_length = model.config.n_positions if len(input_ids[0]) > model.config.n_positions else len(input_ids[0]) + 20
    beam_output = model.generate(
        input_ids,
        attention_mask=attention_mask,
        max_length=max_length,
        min_length=4,  # Mindestlänge der Antwort
        num_beams=5,
        no_repeat_ngram_size=2,
        early_stopping=True,
        temperature=0.9,
        top_p=0.90,
        top_k=50,
        length_penalty=2.0,
        do_sample=True,
        eos_token_id=tokenizer.eos_token_id,  # EOS Token setzen
        pad_token_id=tokenizer.eos_token_id 
    )
    
    text = tokenizer.decode(beam_output[0], skip_special_tokens=True)
    return text

css = """
    body { font-family: Arial, sans-serif; }
    .gradio_container { max-width: 700px; margin: auto; padding-top: 50px; }
    .gradio_header { display: none; }
    .gradio_input_box { border-radius: 10px; }
    .gradio_output_box { border-radius: 10px; }
    button { background-color: #29B3FF; color: white; padding: 10px 20px; border: none; border-radius: 5px; cursor: pointer; }
    button:hover { background-color: #106ba3; }
"""

iface = gr.Interface(
    fn=generate_text,
    inputs=gr.Textbox(label="Schreibe hier...", placeholder="Stelle deine Frage..."),
    outputs=gr.Textbox(label="Antwort"),
    title="Löwolf Chat",
    description="Willkommen beim Löwolf Chat. Stelle deine Fragen und erhalte Antworten vom KI-Chatbot.",
    css=css
)

iface.launch()