app.py

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

if torch.cuda.is_available():
    tokenizer = AutoTokenizer.from_pretrained("ai-forever/ruGPT-3.5-13B")
    model = AutoModelForCausalLM.from_pretrained("ai-forever/ruGPT-3.5-13B", load_in_8bit=True, device_map="auto")
    pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)

@spaces.GPU(duration=600)
def predict(prompt, temperature, max_length):
    return pipe(prompt, temperature=temperature, max_length=max_length, top_p=0.95, top_k=50, do_sample=True)[0]["generated_text"]

demo = gr.Interface(
    fn=predict,
    title="ruGPT-3.5-13B Demo",
    inputs=["text", gr.Slider(minimum=0.01, maximum=1.0, value=0.7, label="temperature"), gr.Slider(minimum=1, maximum=1024, value=50, label="max_length")],
    outputs=["text"],
)

demo.launch()