app.py

import gradio as gr
import torch
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import spaces  # обязательно для ZeroGPU

model_name = "google/flan-t5-base"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

@spaces.GPU
def respond(message, history=[]):
    prompt = f"Answer the question: {message}"
    inputs = tokenizer(prompt, return_tensors="pt").to(device)
    outputs = model.generate(**inputs, max_new_tokens=100)
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    history.append((message, response))
    return history

demo = gr.ChatInterface(fn=respond)

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