app.py

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

MODEL_ID = "Lev384501/qwen3-0.6b-russian-dialogues"
SEP = "\n### Ответ:\n"

print("Загружаю модель...")
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
model = AutoModelForCausalLM.from_pretrained(
    MODEL_ID,
    dtype=torch.float16,
)
model = model.to("cuda")
print("Модель загружена.")


@spaces.GPU
def respond(message, history, max_tokens, temperature, top_p):
    prompt = message.strip() + SEP

    inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

    with torch.no_grad():
        output = model.generate(
            **inputs,
            max_new_tokens=max_tokens,
            temperature=temperature,
            top_p=top_p,
            do_sample=True,
            repetition_penalty=1.3,
            pad_token_id=tokenizer.eos_token_id,
        )

    generated = output[0][inputs["input_ids"].shape[1]:]
    answer = tokenizer.decode(generated, skip_special_tokens=True).strip()

    return answer


chatbot = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Slider(minimum=1, maximum=256, value=80, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=2.0, 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"),
    ],
    title="Qwen3-0.6B Russian Dialogues",
)

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