app.py

import gradio as gr
import re
from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer

# Загрузка модели и токенизатора
model_name = "Dennterry/okt_bot"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

def respond(message, history, system_message, max_tokens, temperature, top_p):
    # Формируем текст, который будет передан в модель
    inputs = tokenizer(f'@@ПЕРВЫЙ@@{message}@@ВТОРОЙ@@', return_tensors='pt')

    generated_token_ids = model.generate(
        **inputs,
        top_k=50,
        top_p=top_p,
        num_beams=5,
        num_return_sequences=3,
        do_sample=True,
        no_repeat_ngram_size=2,
        temperature=temperature,
        repetition_penalty=1.5,
        length_penalty=0.6,
        eos_token_id=50257,
        max_new_tokens=max_tokens
    )

    # Извлечение и возврат текста ответа
    context_with_response = [tokenizer.decode(sample_token_ids) for sample_token_ids in generated_token_ids]
    result1 = re.sub(r'@@.*?@@', '', context_with_response[0])
    result2 = result1[len(message):]
    yield result2.strip()

# Настройка интерфейса Gradio
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="Чебупели", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=100, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=1.2, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95, step=0.05, label="Top-p (nucleus sampling)"
        ),
    ],
)

if __name__ == "__main__":
    demo.launch()