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)

    max_length = model.config.n_positions if len(input_ids[0]) > model.config.n_positions else len(input_ids[0]) + 50
    beam_output = model.generate(
        input_ids,
        attention_mask=attention_mask,
        max_length=max_length,
        min_length=4,
        num_beams=5,
        no_repeat_ngram_size=2,
        early_stopping=True,
        temperature=0.9,
        top_p=0.95,
        top_k=70,
        length_penalty=2.0,
        do_sample=True,
        eos_token_id=tokenizer.eos_token_id,
        pad_token_id=tokenizer.eos_token_id 
    )
    
    text = tokenizer.decode(beam_output[0], skip_special_tokens=True)
    return text
h1 = "Löwolf Chat"
css = """
    body { font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif; }
    .gradio_app { max-width: 600px; margin: 50px auto; }
    .gradio_interface { box-shadow: 0 0 20px rgba(0,0,0,0.1); }
    .gradio_input_container { width: 100%; }
    .gradio_output_container { width: 100%; margin-bottom: 25px; }
    .gradio_input { border-radius: 20px; }
    .gradio_output { border-radius: 20px; }
    button { border-radius: 20px; }
"""

iface = gr.Interface(
    fn=generate_text,
    inputs=gr.Textbox(placeholder="Type a message..."),
    outputs="text",
    css=css
)

iface.launch()