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

# Initialisierung des Modells und des Tokenizers
tokenizer = GPT2Tokenizer.from_pretrained("Loewolf/L-GPT_1.1")
model = GPT2LMHeadModel.from_pretrained("Loewolf/L-GPT_1.1")

def generate_text(prompt):
    input_ids = tokenizer.encode(prompt, return_tensors="pt")
    attention_mask = torch.ones(input_ids.shape, dtype=torch.long)

    max_length = model.config.n_positions if len(input_ids[0]) > model.config.n_positions else len(input_ids[0]) + 90
    beam_output = model.generate(
        input_ids,
        attention_mask=attention_mask,
        max_length=max_length,
        min_length=1,
        num_beams=5,
        no_repeat_ngram_size=2,
        early_stopping=True,
        temperature=0.7,
        top_p=0.9,
        top_k=10,
        
        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


css = """
h1 {
  text-align: center;
}

#duplicate-button {
  margin: auto;
  color: white;
  background: #1565c0;
  border-radius: 100vh;
}

.contain {
  max-width: 900px;
  margin: auto;
  padding-top: 1.5rem;
}

"""

iface = gr.Interface(
    fn=generate_text,
    inputs=gr.Textbox(lines=2, placeholder="Type a message...", label="Your Message"),
    outputs=gr.Textbox(label="Löwolf Chat Responses", placeholder="Responses will appear here...", interactive=False, lines=10),
    
    css=css
)

iface.launch()