app.py

import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

"""
For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
"""

# Инициализация модели и токенизатора
MODEL_NAME = "yandex/YandexGPT-5-Lite-8B-pretrain"
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, legacy=False)
model = AutoModelForCausalLM.from_pretrained(
    MODEL_NAME,
    device_map="cuda" if torch.cuda.is_available() else "cpu",
    torch_dtype="auto",
)

def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    # Формируем контекст из истории
    full_prompt = f"{system_message}\n\n"
    for user_msg, assistant_msg in history:
        if user_msg:
            full_prompt += f"User: {user_msg}\n"
        if assistant_msg:
            full_prompt += f"Assistant: {assistant_msg}\n"
    full_prompt += f"User: {message}\nAssistant:"

    # Токенизация и генерация
    inputs = tokenizer(full_prompt, return_tensors="pt").to(model.device)
    outputs = model.generate(
        **inputs,
        max_new_tokens=max_tokens,
        temperature=temperature,
        top_p=top_p,
        do_sample=True,
        pad_token_id=tokenizer.eos_token_id,
        stream=True
    )
    
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    # Убираем начальный промпт из ответа
    response = response[len(full_prompt):]
    yield response


"""
For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docsx/chatinterface
"""
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, 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()